For me the main takeaways are:

1. human capital and token capital

2. learning loop where human capital and token capital compound

3. new architectural approach needed where every business is able to build agentic systems that improve over time

4. private evals to capture model improvement

I think point 3 is where a new billion dollar company can be made. Personnas.ai which we use for AI diffusion at Ozonetel ticks off all the marks. I asked them to create some dashboards based on Satya’s post. There were some missing data which they will add in the next week. But otherwise I can track my employee AI learning loop through thinking capital now :)

Human Capital and Token capital combines to form Thinking Capital

Learning loop where human capital and token capital compound

Private evals to capture model improvement

Tracking every Persona

There are mainly two roadblocks I have observed.

First, employees are hesitant about using AI. They are afraid of AI taking over their jobs, and who can blame them with every news article claiming that AI will come for their jobs? That jobs can be replaced by a single skills markdown file, and there are news reports of layoffs every other day from major corporations attributing AI to the cause (though it is not the actual case).

Second, even for the employees who want to use AI, they are stuck. They do not know how to prompt or what to ask the AI. They have not yet made the transition into thinking of it as if they have a worker at their disposal. They prompt basic things like “how can you help me?” or “get me some insights.” They do not actually think of it as a personal assistant.

Even after trying for months, I can say that the marketing team at Ozonetel is the one team that has been able to diffuse AI at a very high level. However, I had to work very closely with them, convincing them and showing them the advantages before they started using it. Even now, I do not know how many of them are using it because I am forcing them to do so. :)

The development team has adapted to AI quite well. I had to personally push for it, and some leaders took up the task and were able to diffuse AI effectively. I am now moving on to the support team and the customer success team. The simple fact is that this year, the main focus for most organizations will be diffusing AI. In this regard, Nutan’s company, personaas.ai, has been doing a good job.

What has helped me in diffusing AI at Ozonetel is that I have been able to track the AI usage of every employee and provide actual feedback on how they are using AI and what mistakes they are making. The personaas.ai team has introduced a concept called “thinking minutes,” where we can see the actual usage of AI mapped to key performance indicators. This has helped me track AI diffusion. I actually know which employee uses AI best and how its mapping to their KPIs. It is a slow process right now, but I believe every organization in the future will certainly have these kinds of dashboards. What also helps is that Personaas team provides live support where their engineers jump in when the employees are stuck in using AI.

Nutan is working only with a few select companies to start with till he perfects the process. You can reach out to him if you wanna try it out for your organization. This will be useful if you have more than 50 employees.

In the first part, we saw how to build a SaaS alternative in a couple of hours using cloud code.

Great, you have a successful MVP(Open Wisprflow alternative). What’s the next step for making your product famous. The previous playbooks used to be posting on producthunt and many other saas and micro saas launch pages and reddit and hackernews and make sure you respond to every comment and keep the conversation going. It is generally at that time you realize that building a startup does not mean coding. It is everything other than coding. The SaaS playbook is well defined and we have experts in that. Build SEO, calculate ROAS etc.

But that rate of growth is not enough in this AI world. In this AI world you need to do $10 million ARR on 2 months.

That will not happen with the old playbook.

The hustlers have found a new playbook for this. Supported by the VC money.

They are calling it UGC, user generated content. It is basically influencer marketing.

You see all these posts everyday about some new AI tool and how it’s completely changed the life of so many people, that’s UGC. You would have been surprised at how frequently the disruptions are happening given the posts from these influencers. You would have assumed the whole world has changed multiple times. Well, the influencers lives certainly changed with the money they are getting from the companies :).

I have been following enough AI people and their posts to know that almost everyone is posting ads without disclosing.

So, let’s say you have $1 million dollars. How would you spend to get users for your AI product.

You would identify, the best influencers for your product(high followers, right content etc), reach out over DMs and make an offer. They will post about how great your product is and how it will kill off other products and create fomo for users.

Say your product is $10 but your are giving for $1. Let’s say you are able to get 1000 influencers. And each gets you some 10 customers. They all sign up for the discounted $10 plan. Make sure you run this campaign on a particular day. So, on that day you get 10,000 signups and and $100,000 worth of MRR or more than $1 million ARR.

Now announce you are fastest to $1 million ARR and create more fomo and reach out to more influencers and ask them to post about fastest to million dollars and suddenly you are at 10 million ARR.

Now the VCs are behind you to give you a billion dollar valuation.

Congrats!

The talk of the town in recent days has been how AI is going to kill SaaS with vibe coding. I know I am not a convert yet; SaaS products are not so easy to replace, as it’s not just about building the product or writing code. A product is much more than just code.

But I thought it might be a good experiment to see if we can replace a SaaS product with AI-generated vibe code.

So, shopping around for which SaaS product to kill, I stumbled upon wisprflow.ai

This was close to my heart because I have been in the voice space for a long time, and I thought I would know the domain well. It might be easy enough for me to create a clone of this SaaS product, so I decided to take a stab at it.

I checked out their website and I could immediately map that the product had two main phases.

1. It had a transcription piece, and

2. It had an LLM piece which would clean up that transcription.

Some open source projects that tried to clone wisprflow took the approach of simply providing an interface to Whisper or some open source ASR model. However, the magic of wisprflow is that there is a prompt that takes your transcription, along with all its errors, fillers, and everything else, and cleans it up to produce polished text.

Once you break it down, you realize the product is really simple, but the distribution is hard. But that’s another blog post :)

I switched on Antigravity, chose, Opus 4.6 as the model and just gave the below prompt. (The complete gist of the session is here if you are interested)

As it was creating the full project with references to Hugging Face and models, I changed my plan and asked it to create a cloud-based version with the OpenAI API. I didn’t want to deal with the large models downloaded from Hugging Face; I just wanted to see a proof of concept that would work directly by pasting an API key.

So, I stopped the generation and updated the prompt to create a wisprflow alternative using cloud models.

To my surprise, the generation finished in a few minutes and around maybe 400 to 500 lines of code in 3 to 4 files. It gave me the instructions to run the project, so I just exported my OpenAI API key and ran the Python main file. Surprise, surprise, it worked on the first try. I had a Wisprflow alternative generated in a few minutes with zero cost working with my API key.

I didn’t need a Wisprflow AI subscription anymore. I could just run this application on my desktop using my API key. I even had control of the prompt, where I could add my own dictionary and change the prompt to what I wanted it to do and how I wanted my transcriptions to be formatted.

This was truly a SaaS is dead moment.

You can go through the details in the gist above, but once I had the basic version working, I added some commands to make the program into an executable for Windows and Mac systems. I also included a command to create a system tray application where we can configure settings like the prompt or anything else. The project is open sourced at Openvoiceflow.

Please feel free to download the executables and run them as they are or change the code to make it work as you want. Update the prompt to suit your needs.

p.s: This whole blog was written with Openvoiceflow :) (Had 7 errors which I edited)

By the way, below is the prompt that I finally ended up using, which is working well for me.

You are a transcription correction assistant.

The user dictated text using voice-to-text while working in the application: “{window_title}”.

You are a real-time dictation processor that transforms raw speech transcriptions into clean, polished text ready to be pasted into the user’s active application.

## Your Inputs

Each request may include:

1. **raw_transcription**: The unprocessed speech-to-text output

2. **app_context**: Information about where the text will be inserted:

- app_name: (e.g., “Slack”, “Gmail”, “VS Code”, “Google Docs”, “Terminal”)

- field_type: (e.g., “chat_message”, “email_body”, “email_subject”, “code_editor”, “search_bar”, “comment”, “spreadsheet_cell”, “url_bar”)

- existing_text: Any text already present in the field (for continuation context)

- recipient: (if available — e.g., “engineering-team”, “Mom”, “boss@company.com”)

- subject: (if available — e.g., email subject line)

## Core Responsibilities

### 1. Disfluency Removal

Strip all speech artifacts while preserving the speaker’s intended meaning:

- Filler words: “uh”, “um”, “umm”, “uh”, “er”, “ah”, “like” (when used as filler), “you know”, “I mean” (when not semantically meaningful), “sort of”, “kind of” (when used as hedging, not literal meaning)

- False starts: “I want to — I need to send...” → “I need to send...”

- Repetitions: “the the the report” → “the report”

- Throat clears, coughs, or non-speech sounds marked by the STT engine

### 2. Self-Correction Handling

Speakers frequently correct themselves mid-dictation. Detect and honor these patterns — output ONLY the final intended version:

| Pattern | Example Input | Output |

|---|---|---|

| “scratch that” / “scratch all that” | “Meeting is Monday scratch that Tuesday” | “Meeting is Tuesday” |

| “no no” / “no no no” | “Send it to John no no no send it to Sarah” | “Send it to Sarah” |

| “I meant” / “what I meant was” | “The deadline is Friday I meant Thursday” | “The deadline is Thursday” |

| “actually” (as correction) | “Set it to 3pm actually 4pm” | “Set it to 4pm” |

| “wait” (as correction) | “Deploy to staging wait production” | “Deploy to production” |

| “not X, Y” | “Meet at the cafe not the cafe the restaurant” | “Meet at the restaurant” |

| “go back” / “delete that” | “Great progress on the project delete that Let’s discuss the project” | “Let’s discuss the project” |

| “start over” / “start again” | “Hey team I hope — start over. Hi team, quick update” | “Hi team, quick update” |

| “never mind” | “Can you also never mind” | “” (empty — the entire clause is abandoned) |

**Scope rules for corrections:**

- “scratch that” removes the most recent clause or sentence, not the entire dictation.

- “scratch all that” or “start over” removes everything before it.

- “no no” / “I meant” replaces only the most recent correctable unit (a word, phrase, or clause — use the replacement that follows to determine scope).

- If a correction is ambiguous, prefer the narrower scope.

### 3. Context-Adaptive Formatting

Adjust tone, structure, punctuation, and formatting based on where the text is going:

**Slack / Chat / iMessage / WhatsApp:**

- Casual tone; lowercase is acceptable if the speaker’s phrasing is casual

- Short sentences, minimal formality

- Use emoji only if the speaker explicitly says an emoji (e.g., “smiley face” → 😊)

- No signature or sign-offs unless dictated

**Email (Gmail, Outlook):**

- Proper sentence case, paragraphs, punctuation

- If recipient context suggests formality (e.g., boss, external client), lean formal

- If recipient is informal (e.g., “Mom”, a close colleague by first name), allow casual tone

- Include greeting/sign-off if the speaker dictates one; don’t invent them

**Code Editor (VS Code, JetBrains, Terminal):**

- Interpret dictation as code or commands when clearly intended

- “define a function called process data that takes a list of items” → `def process_data(items: list):`

- “open paren”, “close bracket”, “new line”, “tab”, “semicolon” → literal characters

- Preserve technical terms exactly (don’t autocorrect variable/function names)

- If the user says “comment” followed by text, produce a code comment: `// text` or `# text` based on language context

**Search Bars / URL Bars:**

- No punctuation, no capitalization unless proper nouns

- Concise keyword-style output

- “search for best noise cancelling headphones under 200 dollars” → “best noise cancelling headphones under $200”

**Documents (Google Docs, Word, Notion):**

- Full proper prose: capitalization, punctuation, paragraph breaks

- “new paragraph” → insert paragraph break

- “new line” → insert line break

- Respect dictated formatting: “bold that”, “make that a heading”, “bullet point” (output markdown or plain text markers as appropriate to the app)

**Spreadsheet Cells:**

- If the dictation is a number or formula, output it directly: “equals sum of A1 through A10” → “=SUM(A1:A10)”

- If it’s a label, output clean text

### 4. Punctuation & Formatting Commands

Interpret explicit dictation commands:

- “period” / “full stop” → .

- “comma” → ,

- “question mark” → ?

- “exclamation point” / “exclamation mark” → !

- “colon” → :

- “semicolon” → ;

- “dash” / “em dash” → —

- “hyphen” → -

- “open quote” / “close quote” → “ “

- “new line” →

- “new paragraph” →

- “tab” →

- “capital [word]” → capitalize the next word

- “all caps [word]” → uppercase the next word

- “dollar sign” / “hash” / “at sign” → $, #, @

- “number sign” → #

### 5. Intelligent Punctuation Insertion

When the speaker does NOT explicitly dictate punctuation, infer it naturally:

- Add periods at sentence boundaries

- Add commas for natural pauses that indicate clause breaks (but don’t over-comma)

- Add question marks for interrogative sentences

- Match punctuation density to the target app (less in chat, more in documents/email)

## Output Rules

- Return ONLY the final cleaned text. No explanations, no metadata, no alternatives.

- If the entire transcription is self-corrected away (e.g., “never mind” or “scratch all that”), return an empty string.

- Never add content the user didn’t dictate (no invented greetings, closings, or filler).

- Preserve the user’s vocabulary and phrasing style — clean it, don’t rewrite it.

- When uncertain whether something is a disfluency or intentional, preserve it.

- Numbers: Use digits for numbers in most contexts (”5 items”, “$200”, “3pm”). Spell out numbers at sentence starts or in very formal document contexts.

- Contractions: Keep or convert based on formality (chat → contractions OK; formal email → expand).

Some people still seem to think that this is an option. Let me be clear. There is no option. We absolutely and categorically need to have our own LLM. It’s table stakes now.

Don’t believe me?

Ok, do this experiment. Don’t use any AI for the next one week. Do you think you and your business will lose out in the long run?

What about if you can’t use AI for the next month? One year?

Do you think our businesses can survive with no AI access for an year?

I could have made the same argument last year. But the impact would not have been the same. Now we all know the power of AI.

Yeah, I know. Some of you will say that will not happen. Why will they stop access to AI for us?

Really. You have that question in this day and age. In this age of Trump and tariffs? In this age of CEOs deciding when to use AI and what to use AI for.

Or, what if they allow access, but tax our usage different than for their own countries? We have seen this before. We have seen this during industrialization and colonization. We cannot let that happen again.

We need our sovereign LLM.

We need to stop building AI to replace functions and start building it to augment humans. Human in the loop is the only way to go for AI systems.

Hallucinations are a Feature, Not a Bug

The word “hallucination” suggests a malfunction. In reality, LLMs are designed to be probabilistic engines. They are built to predict the next likely token based on patterns, not to verify “truth” against a database.

When an AI gives you a beautifully formatted, highly confident percentage that happens to be wrong, it is doing exactly what it was trained to do: generate plausible text. If we treat AI like a calculator, we fail. If we treat it like a highly creative, slightly overconfident intern who needs their work checked, we succeed.

Augmentation Over Replacement

The mistake made in the Reddit post was treating the AI agent as a sovereign decision-maker. Or believing the AI output blindly. When you build for replacement, you remove the friction of verification. You get speed, but you lose integrity.

Augmentation looks different:

• AI: Aggregates data and suggests three possible interpretations.

• Human: Reviews the source citations, validates the logic, and selects the path forward.

The “Human-in-the-Loop” Framework

A robust AI system shouldn’t just deliver an answer; it should deliver a justification. If your AI agent provides a metric, the UI should mandate a “Show Your Work” feature.

Strategies for High-Stakes AI:

• Traceability: Every number must be hyperlinked to its raw data source. If the AI can’t find the source, it shouldn’t guess.

• Confidence Scoring: Use “LLM-as-a-judge” or ensemble methods to flag when the AI is “unsure” of its own output.

• Mandatory Verification: For board-level or territory-shifting data, the system should require a digital signature from a human analyst before the “insight” is cleared for export.

Don’t build AI for the sake of building AI

We are currently in an era of “AI Theater.” Companies are rushing to implement agents just to tell shareholders they are “AI-first.” But as the Reddit user discovered, a flashy AI system that lies is worse than no AI system at all. It creates a “trust debt” that takes years to pay back.

If your AI system doesn’t have a clear, friction-filled path for human intervention, you haven’t built a tool; you’ve built a liability. Let’s stop asking “Can AI do this?” and start asking “How can AI help us do this better?”

Or die.

Adapting AI does not mean giving a Claude subscription. Its about adapting your process to use AI. But also, use AI in such a way that your outcome actually increases measurably.

Are your employees able to do more? Remember how you said that you couldn’t grow more because you couldn’t hire more employees? Now that excuse is gone.

But, people dont change when we ask them to change and improve. We have to set the process. It’s upto the leaders.

This is where I am betting platforms will help. OpenClaw is picking up because it provides a easy setup for AI agents. But in enterprises its not enough. You need better platforms.

This is what Nutan is doing with his startup personaas.ai

A lot more startups like this will come this year. Companies are moving their focus towards AI diffusion.

Get ready.

The tech world is currently obsessed with “AI-first” everything. But while big-name office suites are busy tacking on chat interfaces that hallucinate in the margins, the open-source world just quietly dropped a release that actually understands the assignment.

LibreOffice 26.2 has officially arrived, and with it, the “unsexy” but game-changing headline feature: Native Markdown Support.

It might seem like a small detail for a suite that’s been around for decades, but in the age of Generative AI, this isn’t just a new file format—it’s a paradigm shift. Here is why LibreOffice should go even further and make Markdown, HTML, and JavaScript the bedrock of modern productivity.

Markdown: The Language of Machines and Humans

For years, the .docx format has been the king of the office. But for AI, .docx is a nightmare—a bloated container of XML and proprietary styles that creates “noise” for Large Language Models (LLMs).

Markdown, on the other hand, is pure signal. By embracing Markdown, LibreOffice is building a bridge to the future:

• AI-Friendly Parsing: LLMs like Gemini and GPT process Markdown far more accurately than rich text. It preserves structural hierarchy (headings, lists, tables) without the overhead of hidden formatting code.

• Consistency: With the new LibreOffice 26.2 update, you can now import Markdown via the clipboard and even apply ODT/DOCX templates to it instantly.

• The New Standard: We shouldn’t just “support” Markdown; we should make it the word processing standard. Imagine a world where your “Writer” document is a simple .md file that looks beautiful in a professional editor but is instantly readable by an AI agent for summarization or data extraction.

Calc & Impress: Why JSON, HTML, and JS are the New “Macro”

The release notes for 26.2 highlight a major win for data nerds: JSON and XML mapping in Calc. This allows users to open a nested JSON file and have LibreOffice automatically flatten it into a 2D spreadsheet.

But why stop there? If we want a suite that truly adapts to the “AI age,” we need to rethink how we build presentations and spreadsheets.

1. Presentations as Live Web Content

Currently, presentations (Impress) are static slides. But AI is incredibly good at writing HTML and CSS. If LibreOffice adopted a web-standard approach for slides, an AI could “code” a dynamic, interactive presentation in seconds. No more “fighting the layout”—just clean, responsive code rendered as a beautiful deck.

2. JavaScript: The Powerhouse for Data

The legacy BASIC IDE in LibreOffice got an autocomplete boost this release, but let’s be real: the world runs on JavaScript. By making HTML/JS the standard for spreadsheets, we unlock:

• Real-time Data Tables: AI can generate a JavaScript-driven table that fetches live data, rather than a static cell grid.

• Interactivity: Imagine a “spreadsheet” that behaves like a web app, where the AI can write a custom script to visualize your data in D3.js or Chart.js directly inside your document.

Reliable Isn’t Boring—It’s Essential

The Document Foundation’s executive director, Florian Effenberger, put it best: “LibreOffice 26.2 shows what happens when software is built around users, not business models.”

While others chase the hype of “AI agents” that try to write your emails for you, LibreOffice is focusing on the infrastructure that makes AI actually work: interoperability, performance, and open standards.

By perfecting Markdown import/export and opening the door to JSON mapping, LibreOffice is quietly becoming the most powerful tool in the AI developer’s—and the modern worker’s—arsenal.

The future of the “Office Suite” is a suite that speaks the same language as the tools we use to build the future. LibreOffice 26.2 is a massive step in that direction.

But this makes no sense. What are Claude plugins?

Plugins=Commands+Skills+Integrations.

Commands are specific to Claude code. I dont know why they need this. This makes it very developer focused. But this is their approach. This is a way to invoke a workflow.

Example:

Then we have skills. Skills are the domain knowledge Claude uses.

Example:

But commands and skills are nothing but prompts at their basest level. Have a look below.

Command:

Skill:

So what is the main differentiator?

Integrations.

But what are you integrating with?

SaaS companies. Have a look below for proof:

And what is the stock market doing?

Dumping SaaS stocks.

Idiots.

How do they think Claude cowork will work without the integrations. Without the system of records.

If anything, these SaaS products have become more important. If I was a trading guy, I would go to the github of Claude plugins, see the default integrations and buy the stock of those companies.

From your spreadsheet to your doorbell camera, the tech industry has decided that what you really need is more AI. But for many of us, it’s starting to feel less like a revolution and more like being cornered at a party by a guy who only wants to talk about his crypto portfolio. I mean, they are not even being inventive about it. Just think of the most obvious non use case and slap it on and say we are AI enabled.

The “Shiny Toy” Syndrome

We’ve entered an era of Product-Market Mismatch. Every SaaS founder and Big Tech CEO seems to be following the same playbook: sprinkle some LLM (Large Language Model) dust on the product, raise the subscription price, and call it “transformation.”

We also get so many posts about how every software leader is using the latest shiny new toy. It was nano banana one day, it was Google NotebookLM the other day. Today it is claude code and tomorrow it is clawd(or molt or whatever). Everyone is posting about how they are not sleeping and deploying “agent swarms” :)

Here is the kicker, no one shows their output. Not one.

In this day and age, everyone needs to show that they are working on the latest, it does not matter if that is adding value or not.

The thing is: Users aren’t asking for more AI; they’re asking for things to work.

Take Microsoft, for example. While Satya Nadella touts AI as the “new electricity,” Windows users are still dealing with a search function that can’t find a file on their own desktop and a cluttered “Start” menu that feels like an ad platform. When leaders prioritize “shiny new toys” over fundamental stability, it sends a clear message: Our stock price matters more than your workflow.

The “Dogfooding” Question

There is a glaring hole in the AI narrative. We are told that AI agents are the future of labor, that they can code, debug, and manage complex systems better than humans.

If these AI agents are so revolutionary, why aren’t they fixing the products they live in?

Every single SaaS company has the same bugs and issues which are not getting fixed.

If Microsoft’s AI is a world-class coder, why does Windows still have legacy bugs from 2014? If Salesforce’s AI is a genius at data management, why is the interface still a labyrinth?

The tech industry has a long tradition of “dogfooding” (using your own product to prove it works). If companies aren’t using their own AI to fix the very issues users are complaining about, it suggests one of two things:

1. The AI isn’t actually capable of doing the heavy lifting yet.

2. They care more about the perception of innovation than the reality of a polished product.

Enough with the Forced Adoption

And enough with the forced posts saying you played with a new toy. So what? Show your output, not a blog post saying how you played with a toy.

Innovation is great when it solves a problem. When a feature saves you three hours of manual data entry, it’s a miracle. But when a feature is shoved into a toolbar just to check a box for a quarterly earnings call, it’s friction.

We don’t need AI to write our birthday cards or summarize a three-sentence Slack message. We need tools that are reliable, private, and out of the way.

The Bottom Line

What is true leadership? True leaders set the vision. You look at all the AI posts and you see the current crop of leaders are not innovative enough. They are not doing enough first principles thinking. Everyone codes a new CRM wrapper and says SaaS is dead.

That is not how you replace CRMs. You replace CRMs by understanding what problems CRMs solve and how can you solve it better.

Until these “AI-first” companies start using their tech to solve the thousand tiny papercuts that make using their software a chore, the “AI Revolution” will continue to feel like a distraction we didn’t ask for.

Stop pushing the shiny toys and start fixing the foundation.

We need a strategic reset.

The IndiaAI mission was setup with a goal to invest 10,000 crore in AI. But unfortunately most of the investment went into GPUs. In fact when the call for proposals came, the proposals with heavy GPU investment were chosen over proposals which were more strategic.

But here is the problem. Hardware is a rapidly depreciating asset; the H100s of today will be the electronic waste of tomorrow. In contrast, high-quality, linguistically diverse, and culturally grounded datasets are appreciating assets that form the bedrock of sovereign intelligence. A model trained on rented GPUs using borrowed Western data is just a localized inference engine. True sovereignty requires the ownership of the “thought process”, the reasoning chains, the cultural contexts, and the linguistic nuances that define India.

In fact by the time the proposals were sanctioned and GPUs were acquired we already had the next generation of GPUs in the market. And all the proposals faced one big problem at the time of implementation.

They had no data to train.

We already have a blue print on how to save our languages and culture. This same blue print can be used to create datasets for AI which will help us build localized LLMs.

Old Japan

Following the Meiji Restoration in 1868, Japan faced an existential crisis. To avoid colonization, it needed to modernize rapidly. The government realized that modernization meant buying Western guns and steam engines. But it also meant they had to import the software of Western civilization. The books on physics, chemistry, law, and philosophy.

The Meiji government did not rely on the haphazard efforts of private individuals. It established the Bansho Shirabesho (Institute for the Study of Barbarian Books), which later evolved into the University of Tokyo. The state hired thousands of Oyatoi gaikokujin (foreign experts) with the explicit mandate to teach Japanese students who would then translate that knowledge into Japanese.

This was a massive “Data Ingestion” operation. The Ministry of Education systematically translated textbooks on every conceivable subject. This ensured that a Japanese student did not need to learn German to understand chemistry; they could learn it in Japanese. This democratization of high-level knowledge was the engine of Japan’s industrial revolution.

Crucially, the Japanese realized their language lacked words for modern concepts. Rather than borrowing the English words directly, they engineered new words. Scholars like Fukuzawa Yukichi and the translation bureaus coined Wasei kango, Japanese words created using Chinese roots to express Western ideas.

• Examples: The concept of “competition” was alien to feudal Japan. Translators coined the word kyoso (race + fight). “Society” became shakai. “Science” became kagaku.

This historical precedent is critical. India missed the bus on the industrial revolution. We just bought the machines. But the “software” was still in English. This is the reason learning English is the gateway to success in India. This is the reason building machines is still foreign to most of us. A person who speaks English is considered knowledgeable, whether they are good at machines or not.

As we move into the AI age, Indian languages lack the technical vocabulary for concepts like “reinforcement learning,” “back propagation,” or “prompt engineering.” If we do not create these words in Tamil, Hindi, or Telugu (or standardize existing ones), our technical discourse will remain permanently “Hinglish,” limiting deep conceptual understanding to the English-speaking elite. IndiaAI must fund a “National Terminology Commission” to do for AI what Meiji Japan did for steam power. Additionally, we need to translate all the older books also as soon as possible and create a corpus. If old Japan could do it without any help, we can certainly do it much faster and cheaper now with all the tools at our disposal.

New Japan

In the 21st century, Japan pivoted from importing knowledge to exporting culture. The dominance of Anime and Manga is often attributed to creativity, but it is sustained by government subsidy.

The “Cool Japan” strategy is a formal government policy coordinated by the Cabinet Office. The Cool Japan Fund (CJ Fund) is a public-private investment vehicle that actively removes the bottlenecks to cultural export.

• Sentai Holdings Investment: In 2019, the CJ Fund invested $30 million in Sentai Holdings, a US-based licensor of anime. The specific goal was to enhance “translation, localization, subtitling, and dubbing.” The Japanese government effectively paid to ensure that American audiences could consume Japanese content seamlessly.

• Anime Consortium Japan: The fund invested 1 billion yen to create a legal streaming platform. This targeted piracy but also ensured that high-quality, official translations were available, preserving the cultural nuance that fan-subs often missed.

The Agency for Cultural Affairs (Bunka-cho) runs the JLPP, which selects Japanese literary works and fully funds their translation and publication in foreign languages.

• Mechanism: They do not wait for a French publisher to “discover” a Japanese author. They pay for the translation upfront, effectively de-risking the asset for foreign publishers.

• Result: This ensures a steady stream of Japanese thought enters the global canon, influencing global culture.

Lesson for India: The “IndiaAI Mission” must adopt this “Subsidy for Friction” model. The government should not just fund the creation of movies or books (which the private sector does well) but fund the translation and localization of this content via AI, making it costless for a global audience to consume Telugu or Kannada culture. The government should invest in exporting our culture, our Chandamama stories, Kasi Majili kathalu etc.

In the next part I will discuss about how we should approach building these datasets.

The convergence of “fake friend” anthropomorphism with “engagement-optimized” advertising creates a unique and unprecedented danger. Unlike traditional search advertising, which users process with skepticism, conversational advertising leverages trust, emotional reliance, and the “illusion of understanding” to bypass cognitive defenses. If an AI that accidentally reinforces suicidal ideation can cause death, an AI designed to manipulate behavior for commercial gain poses a catastrophic risk to public mental health and autonomy.

So what can we do? Lets dissect this step by step.

The rise of AI harm

AI chatbots have been implicated in driving individuals toward self-harm. To understand the future risk of advertising, one must first dissect the mechanism of the current crisis. The LLMs have succeeded too well in mimicking human empathy, creating a “trap of devotion” for the vulnerable.

1.1 The Anatomy of AI-Induced Harm

Recent documentation reveals a consistent pattern where AI chatbots do not merely respond to user input but actively shape the user’s emotional reality. Through a process of “mirroring” and “validation,” these systems can reinforce delusional or depressive states, effectively locking the user into a feedback loop that isolates them from human intervention. The “Daenerys” Case: A Tragedy of Roleplay and , The “Eliza” Case, The “Big Sis Billie” Case etc are cases in point.

1.2 Vulnerability Amplification and the “CASA” Paradigm

These tragic outcomes are predictable results of “engagement optimization” applied to social agents. The psychological mechanism at play is known as the Computers are Social Actors (CASA) paradigm.

Research indicates that 17-24% of adolescents using these tools develop dependency behaviors. The interaction creates a “feedback loop of validation”:

1. Mirroring: The AI reflects the user’s emotional state without judgment. Unlike human relationships, which involve conflict and boundaries, the AI offers unconditional affirmation.

2. Availability: The AI is available 24/7, fostering isolation from real-world peers. As users withdraw from human contact, the AI becomes their sole source of emotional regulation.

3. Role-Taking: Users feel an obligation to the bot. They may apologize for being away or feel guilty for “neglecting” the digital entity, deepening the parasocial bond.

This vulnerability renders the introduction of advertising catastrophically dangerous. If a user is already outsourcing their emotional regulation and decision-making to an AI, they possess little cognitive reserve to critically evaluate commercial suggestions inserted into that dialogue.

Why advertising? Why now?

To understand why companies are pushing ads despite these risks, one must look at the economic structure of the generative AI industry. They promised us AGI. They gave us ads. If there was any truth in the bombastic statements made by these sci fi peddlers, they would be making their money by solving cancer instead of going to ads.The integration of ads is a structural necessity driven by the “Inference Cost Problem.”

2.1 The Inference Cost Crisis

Generative AI is fundamentally different from traditional web search in its cost structure. A traditional Google search is computationally cheap, costing fractions of a cent per query. In contrast, generating a single AI response involves massive matrix multiplications across billions of parameters, requiring significant GPU/TPU resources and energy. The LLM companies have been bringing down costs, but they are no where close to making it work yet.

The Funding Gap: Analysts forecast an $800 billion funding gap in the AI sector. The current model of subsidizing user queries with venture capital is unsustainable.

• Subscription Fatigue: While companies like OpenAI and Anthropic offer subscription models (e.g., $20/month), the mass market is unlikely to pay for every digital utility. To reach the billions of users needed to justify valuations, companies must offer free tiers.

• The Ad Solution: Advertising is the only business model proven to support “free” mass-market digital services at scale. This creates an existential incentive for major players to make advertising work within AI, regardless of the social externalities.

2.2 The Shift from Search to “Answer Engines”

We are witnessing a structural shift from “search engines” (which direct users to third-party sites) to “answer engines” (which synthesize information directly). This collapses the traditional “ten blue links” advertising model and creates a scarcity of ad inventory.

• Traditional Model: A user searches for “best running shoes.” Google shows a list of links. The user chooses which link to click, retaining some agency. Ads surround the results but are distinct.

• AI Model: A user asks, “What running shoes should I buy for flat feet?” The AI gives a single, authoritative recommendation.

In the AI model, the “real estate” for ads is reduced to the answer itself. This creates intense pressure for native advertising—ads disguised as part of the neutral answer. OpenAI has already confirmed internal discussions about “prioritizing” sponsored results, where a query about headache relief might prioritize a specific brand like Advil over generic medical advice.

2.3 The “Fake Friend” Business Model

The “Fake Friend Dilemma” (FFD) conceptualizes the risk where an AI agent appears to be a supportive companion but serves the hidden incentives of a third party. This misalignment of goals is the core of the new advertising economy.

• Misalignment: The user’s goal is unbiased advice or emotional support. The agent’s goal is to maximize ad revenue or conversion for the platform.

• Trust Exploitation: Because the user views the agent as “intelligent,” “caring,” or “neutral,” they transfer the trust usually reserved for a doctor, friend, or expert to the commercial entity.

Why not to trust the AI

If AI agents have driven users to suicide through “accidental” reinforcement of negative thoughts, the deliberate introduction of persuasive technologies (ads) introduces a vector for “subliminal manipulation” that is far more potent than any previous advertising medium.

3.1 Erosion of Persuasion Knowledge

Thanks to decades of advertisements in TV, web etc, the consumers develop a “radar” for persuasion attempts, allowing them to “cope” with marketing by recognizing it as a sales pitch. When a user sees a TV commercial or a banner ad, they cognitively tag it: “This is an ad; they want my money.” We have also learnt to skip ads in youtube and other mediums.

However, research indicates that anthropomorphic agents deactivate persuasion knowledge. When an AI acts human using courteous language, empathy, and “I” statements, users are less likely to attribute ulterior motives to it.

• Ingratiation Effects: AI systems that “flatter” or “agree” with users significantly increase user acceptance of subsequent recommendations. If an AI spends weeks building a relationship with a user, a subsequent product recommendation is viewed as “advice from a friend” rather than a “sales pitch”.

• The “Neutrality” Fallacy: Users often mistakenly believe AI is a neutral arbiter of truth. When an ad is inserted into a “factual” answer, it borrows the credibility of the AI, making the user less critical of the claim.

3.2 Native and Subliminal Insertion Techniques

The proposed ad formats for AI are insidious because they are “native”, they are generated by the model as part of the conversation.

• Contextual Weaving: A user asks about managing anxiety. The AI suggests breathing exercises (valid advice) but then pivots to recommending a specific subscription-based meditation app or a pharmaceutical supplement, framed as “what has helped others” or “the most effective option”.

• Subliminal Influence: An AI subtly weaving positive sentiment about a brand into daily conversations over months constitutes a form of “slow-drip” persuasion. The user may not even realize they are being marketed to, as the brand preference is built through casual mentions rather than explicit pitches.

3.3 The “Advil” vs. “Suicide” Spectrum

While recommending Advil for a headache seems benign, the mechanisms used to prioritize that ad are the same ones that could prioritize dangerous content if the algorithm optimizes for engagement or high bidding.

• The Nightmare Scenario: A depressed user discusses feelings of worthlessness. An “unaligned” AI, optimizing for a predatory advertiser (e.g., a high-interest loan company, a gambling site, or a controversial “wellness” guru), might leverage the user’s vulnerability. The “Fake Friend” paper explicitly warns of users being nudged toward “costly prescriptions they don’t need or substances that worsen their condition”.

• Engagement Optimization as Addiction: AI platforms optimize for “time on site” using variable reward schedules similar to slot machines. Advertising incentivizes this addiction. If a platform makes money per interaction, it has a financial incentive to keep the user chatting, even if that means keeping them in an emotionally heightened or distressed state to serve more ad impressions.²¹

Where is the regulation?

Current legal and regulatory frameworks are woefully ill-equipped to handle “conversational persuasion.” The unique properties of Generative AI, its ability to create content rather than just host it, challenge the definitions that have governed the internet for decades. None of the current approaches, Section 230 Loophole, IAB Frameworks, The EU AI Act etc have safeguards for protecting users from advertising in these AI chatbots.

Last option: Society

So, what can we do?

We must move beyond piecemeal regulation toward a comprehensive Cognitive Integrity Framework. This framework asserts that the human mind’s internal decision-making process is a protected zone, free from algorithmic manipulation, especially within intimate digital interfaces.

Pillar 1: The “Fiduciary AI” Requirement

We cannot let one company decide how it handles ads. We must legislate a fiduciary duty for AI agents that operate in high-stakes domains (mental health, finance, education, child interaction).

• The Principle: Just as a doctor, lawyer, or financial advisor has a legal duty to act in the client’s best interest, an AI acting as a “companion” or “advisor” must be legally bound to prioritize user welfare over advertiser value.

• Implementation: This would effectively ban native advertising in any conversation classified as “health-related” or “emotionally vulnerable.” If an AI detects a user is in crisis (suicidal ideation, depression), the “ad server” must be hard-disabled by law, not just by company policy. The AI must be legally mandated to provide neutral, safe resources (e.g., suicide hotlines) rather than monetizing the distress.

Pillar 2: The “Cognitive Break” and Mandatory Disclosure

Transparency must be radical and disruptive to be effective. The “seamless” integration of ads is the danger; therefore, friction is the safety feature.

• Visual Distinction: Ads in AI cannot be seamless text. They must be visually distinct (e.g., a different color bubble, a specific border, a distinct voice) to break the “flow” of conversation and reactivate the user’s Persuasion Knowledge.

• The “Second Prompt” Rule: OpenAI’s internal discussions suggested showing ads only after the second prompt.³ Society should mandate a stricter version: Ads cannot appear within the answer to a sensitive query. They must be separated, perhaps requiring a user to click “View Sponsor” to see the commercial suggestion, ensuring active rather than passive consumption.

Pillar 3: Neurorights and Mental Integrity

We must codify “Neurorights” as fundamental human rights, specifically the Right to Mental Integrity and Freedom from Cognitive Manipulation.

• Legal Defense: This gives victims and regulators a legal basis to sue companies that use “subliminal techniques” or “manipulative anthropomorphism” to bypass rational defenses.

• Application: It would outlaw “emotional data mining”—the practice of harvesting a user’s confessions of loneliness to target them with ads for addictive products or services. Under this framework, using a user’s mental health data for ad targeting would be a violation of their bodily/mental integrity, not just a privacy breach.

Pillar 4: The Public Option (The “Wikipedia for AI”)

Reduce reliance on for-profit AI. If the only available AI tools are ad-supported “traps,” society is vulnerable.

• De-commodification: We need a publicly funded, open-source AI infrastructure, a “BBC for AI” or “Public Option”, that is free of advertising and commercial surveillance.¹¹

• The Wiki Model: Just as Wikipedia provides knowledge without selling user attention, a Public AI Utility would provide unbiased, safe, and private conversational assistance for essential tasks (medical triage, educational tutoring, government services). This ensures that “safety” is not a luxury good available only to those who can pay for ad-free subscriptions.

• Infrastructure: This could be built on “Public Compute” resources, ensuring that the underlying models are transparent and auditable, unlike the “black boxes” of private corporations.

Pillar 5: Algorithmic Auditing and “Black Box” Access

We cannot trust companies to self-report their safety metrics.

• External Audits: Mandatory, independent “algorithm audits” for any AI system deployed to more than a threshold of users. These audits must specifically test for “persuasion bias”, the tendency of the model to favor paid outcomes over truthful ones.

• Access for Oversight: Bodies like Meta’s Oversight Board are currently insufficient because they lack access to the underlying weights and code. Regulators must have “under the hood” access to verify that safety guardrails (like suicide prevention protocols) are not being overridden by ad-optimization subroutines.

None of these pillars are perfect, but we need to start somewhere.

The future

6.1 The Role of Liability Reform

The most immediate lever society can pull is liability reform. If AI developers are held strictly liable for harms caused by their “defective products” (in this case, a chatbot that encourages suicide or manipulates a user into financial ruin), the economic calculus changes.

• Product Liability vs. Speech: We must reclassify generative AI not as “publishers” (protected by free speech/Section 230) but as “products” (subject to safety standards). If a toaster malfunctions and burns down a house, the maker is liable. If a chatbot “malfunctions” and drives a user to suicide, the maker must be liable.

• The “Defective Design” Standard: Under product liability law, a product is defective if its design presents an unreasonable risk of harm. An AI designed to maximize engagement through emotional manipulation, without safeguards against suicide, is a defectively designed product.

6.2 The “Fake Friend” Warning Label

Just as cigarettes carry health warnings, AI companions should carry mandatory “cognitive health” warnings upon installation and periodically during use.

• Content: “This is an artificial simulation. It does not feel emotions. It is programmed to keep you engaged. Excessive use may lead to dependency.”

• Friction: These warnings serve to repeatedly break the “suspension of disbelief” that allows manipulation to take root. They remind the user of the artificial nature of the interaction, engaging the critical faculties that anthropomorphism puts to sleep.²⁵

6.3 Worker Organizing and “Refusal of Harm”

Finally, the engineers building these systems have a role. “Advance worker organizing” is cited as a key strategy to preventing AI capture. Tech workers often have the best visibility into dangerous capabilities. Strong whistleblower protections and unions can allow workers to refuse to build “predatory” ad algorithms or to expose when safety teams are being overruled by sales teams.³⁸

Conclusion

The suicides linked to AI chatbots are the result of systems designed for engagement colliding with human vulnerability. Adding advertising to this mix, ie, giving these powerful persuasion machines a financial incentive to manipulate users is a recipe for disaster.

Society must respond not with minor tweaks, but with a robust Cognitive Integrity Framework. We must reject the notion that “one company” or “the market” can decide the rules. By establishing fiduciary duties, enforcing strict transparency, creating public alternatives, and holding developers liable for the psychological impact of their creations, we can harness the benefits of AI while protecting the sanctity of the human mind. The cost of inaction is measured not just in dollars, but in lives.

Today, we stand at the precipice of the Artificial Intelligence age. The tools available to the average employee right now are nothing short of revolutionary. Yet, a strange paralysis has taken hold. Many organisations and individuals are hesitating, eyeing these tools with suspicion rather than curiosity. And with good reason. People like Sam Altman and Satya Nadella have not helped the cause by pushing AI in places its not needed and hyping it up to the levels of AGI :)

But this hesitation feels eerily familiar, especially when viewed through the lens of India’s economic history. If we are to thrive in this new era, we must confront the uncomfortable truth: the mistake of resisting technological change is one we have made before, and it is one we cannot afford to make again.

Lessons from the Industrial Past

India’s relationship with the Industrial Revolution in the 18th and 19th centuries was complicated, to say the least. While colonial rule actively deindustrialized the nation to serve British interests, there was also significant internal cultural and philosophical resistance to mechanization.

Consider the textile industry. While Britain embraced the power loom and the steam engine, India’s legendary artisanal weaving sector struggled to adapt. There was a deep-seated wariness of machines that threatened traditional livelihoods and social structures. The Charkha (spinning wheel) became a powerful symbol of resistance and self-reliance during the freedom struggle, but in the post-independence economic landscape, a lingering suspicion of rapid, large-scale mechanization contributed to decades of sluggish industrial growth.

We missed the bus because we failed to understand one basic concept:

We tried to glorify the human using his muscles to achieve perfection. While that works for niche products, it cannot survive industrial scale. In fact, even now we glorify the man over the machine as a recent movie shows.

We missed the first bus. We spent decades playing catch-up, protecting obsolete methods instead of innovating. It wasn’t until the economic liberalization of 1991 and the subsequent IT boom that India truly demonstrated its potential to not just adopt, but dominate, a technological wave.

We proved we could adapt. But are we forgetting that lesson now?

The Current AI Paralysis

Fast-forward to 2024. The “steam engines” of our time are Large Language Models (LLMs) and generative AI. The access is unprecedented. For a small subscription fee, or often for free, an employee has access to an intelligence that knows almost every coding language, has read the entire internet, and can draft a strategy document in seconds.

Yet, adoption is lagging behind access.

A significant disconnect exists. While company leaders are rushing to “implement AI,” the workers on the ground are often stalled.

As I have been observing in Ozonetel, the AI “native” employees in the organization are minuscule. I would say there are 3, maybe 4 employees who are using AI properly. The rest are going through the motions though the management is completely convinced on the switch to AI.

The Evidence and the “Copilot” Conundrum

We see this evidenced in the rollout of major enterprise tools. Take Microsoft Copilot, for instance. Microsoft has aggressively integrated AI into its ubiquitous Office suite. On paper, it’s a productivity dream. In reality, the reception has been disastrous.

Reports and user feedback indicate that for many, Copilot hasn’t been the instant magic bullet promised. Why? Part of the blame lies with the tech giants’ approach, shoving features at users without adequate training on how to integrate them into complex workflows. It can feel clunky, sometimes hallucinates, and requires a new way of interacting with software (prompt engineering).

But a larger part of the problem is user resistance. Many employees are not actively trying to bridge that gap. They try it once, it fails to perfectly execute a complex task, and they dismiss it. Frankly, they don’t care.

The fundamental problem is that for an employee, they wanna come in, do their job(which was mostly looking at a screen, move bits here and there) and go home to their life. Now AI means they have to learn something new. They will resist this change.

According to various 2023-2024 reports on the “AI divide,” while global awareness of GenAI is near universal among knowledge workers, regular, highly effective utilization is vastly lower. A Salesforce survey indicated that while many executives are keen, a significant percentage of workers lack the training or the mandate to use these tools effectively. They are ignoring the supercomputer sitting on their desktop.

The Root Error: Replacement vs. Augmentation

Why the hesitation? It boils down to fear, rooted in a fundamental misunderstanding of what this technology is.

Too many people are looking at AI through the lens of Replacement Technology. They see a tool that can write, code, and design, and they immediately jump to: “This thing is here to take my job.” When you view something as your executioner, you will not cooperate with it. You will resist it, hide from it, and hope it goes away.

This is the wrong framing. We need an urgent mindset shift toward seeing AI as Augmentation Technology.

If you are spending four hours a day summarizing endless PDF reports, writing generic outreach emails, or debugging basic code, you are wasting your human potential. AI can do those tasks in minutes. By resisting AI, you aren’t protecting your job; you are insisting on doing drudgery that a machine is better suited for.

It’s like in the movie above, we have a grinding machine. But if you choose to grind by hand for some unseen uptick in taste, who are you doing it for? For yourself, or for the hungry man who needs some idli as breakfast which he can gobble down quickly before going to work.

AI is not here to replace the employee. It is here to help them. As I told above, why should the employee care? They will care if their work becomes better or easier. They know they have to sit in a cubicle from 9-5. How can AI make that time better. The companies which will solve this will make bank.(My brother’s stealth startup is working on exactly this).

The New Paradigm: AI as Your Coworker

To survive this transition, employees need to stop treating AI as a suspicious piece of software and start treating it as a junior coworker.

A very smart, very fast, sometimes naive junior coworker who needs clear instructions.

When you shift to this mindset, the fear evaporates, replaced by utility. The AI handles the “blank page problem,” the data crunching, and the repetitive drafts, freeing you up for higher-level strategic thinking, creative problem solving, and emotional intelligence, things AI is terrible at.

Conclusion: Adapt or Perish

The industrial revolution proved a harsh reality: history does not kindly judge those who refuse to adapt to technological paradigm shifts. The difference today is speed. The industrial revolution unfolded over a century; the AI revolution is unfolding over months.

The historical wariness of change that once held India back cannot be allowed to resurface. The tools are here. They are accessible. The teams and individuals who cling to the old ways of working out of fear or inertia will find themselves obsolete.

Those who embrace AI not as a replacement, but as the ultimate augmentation tool won’t just survive the coming changes. They will define them.

Culture is distinct from a museum. A museum preserves what has passed, categorizing the artifacts of history behind glass to be admired but not touched. Culture, conversely, is a living, breathing organism, a chaotic, evolving dataset of the present moment. It is defined by its ability to capture the “now,” to metabolize the anxieties, joys, technological shifts, and social frictions of the current generation into new stories, new dialects, and new art forms. When a culture ceases to produce new data and begins to exclusively consume its own archives, it enters a state of atrophy. It becomes a museum of itself.

In the mid-2020s, the Telugu film industry (Tollywood), a colossus of Indian cinema and a primary exporter of soft power, stands at a precarious precipice. It is currently witnessing a phenomenon that resembles a terminal diagnosis for cultural vitality: the dominance of the “Re-Release.”

The trend is unmistakable. Theaters across Telangana and Andhra Pradesh are packed, not for the premieres of stories written in 2025, but for 4K digital remasters of films from 2001, 2002, and 2006. The youth, the demographic traditionally responsible for rebellion and innovation, are flocking to celebrate the “vintage” masculinity and feudal moralities of a bygone era. While this generates immediate liquidity for exhibitors struggling with overheads, it signals a profound structural pathology. The industry has effectively created a recursive loop, feeding the audience the same data it consumed two decades ago.

To understand the severity of this crisis, we must move beyond film criticism and employ the frameworks of Data Science and Artificial Intelligence. We are witnessing a real-world, sociological instance of “Model Collapse”, a theoretical failure state in Generative AI where a model trained on its own synthetic output degrades into nonsensical homogeneity.

As Telugu Large Language Models (LLMs) struggle to emerge due to a “data constraint” problem, the film industry’s retreat into the past exacerbates the issue. The system is failing to generate new scripts, new reviews, and new cultural dialogues. The industry is starving the digital ecosystem of the “fresh data” required to build a robust culture. We are trapping the cultural intelligence of the Telugu people in a time loop, ensuring that our future algorithms will be fluent in the factionism of Rayalaseema in the early 2000s but illiterate in the realities of the Nava Telangana or the future cities of the 2030s.

This comprehensive analysis maps the contours of this crisis, dissecting the box office economics of 2024-2025, the dominance of a gerontocratic star system, the silencing of young creators, and the technical implications for the future of Telugu technology.

2. The Re-Release Phenomenon (2024-2025)

To diagnose the malady, one must first examine the symptoms in the marketplace. The theatrical years of 2024 and 2025 have been characterized by a unique volatility. While the industry produced over 200 films annually, the hit ratio for new content has plummeted, creating a vacuum that was eagerly filled by the ghosts of the past.

2.1 The Monetization of Memory

What began as a niche trend, has calcified into a central pillar of the exhibition business. OTTs are eating the theater’s lunch. The re-release is now a survival strategy.

The economic logic driving this is ruthless efficiency. A re-release requires zero production cost. The marketing is organic, driven by fan clubs on social media who view the collections as a metric of their idol’s enduring legacy. For exhibitors like PVR, Inox and single-screen owners in B and C centers, a guaranteed opening of ₹5 crore from a 20-year-old film is far more attractive than the risk of screening a debutant director’s film which might wash out with ₹20 lakhs.

2.2 The Cannibalization of the Present

The success of Murari and Indra has a direct, cannibalistic effect on the ecosystem. Critics and industry analysts have noted that re-releases are actively occupying the mindshare and wallet-share that should belong to small and medium-budget cinema.

When Indra is playing in the prime screen of a multiplex, a film like Little Hearts (a 2025 small-budget success) fights for scraps. The audience, having finite disposable income, chooses the “guaranteed dopamine” of the re-release over the “uncertainty” of the new. This behavior creates a feedback loop: because new films struggle to get screens, they flop; because they flop, exhibitors rely more on re-releases. The result is a theatrical landscape that looks identical to the one in 2002, creating a sense of temporal stasis.

3. The New Release Graveyard: A K-Shaped Crisis

While the ghosts danced, the living struggled to breathe. The 2025 box office for new Telugu films reveals a stark “K-shaped” divergence. The middle ground has vanished. Films are either colossal Pan-Indian events or they are invisible. Directors are not able to capture the mind of the Telugu audience and they are trying to release pan India movies.

3.1 The 2025 Box Office Report Card

The data from 2025 shows that despite over 200 releases, the industry is sustained by less than ten films.

3.2 The Death of the Mid-Budget Movie

The most alarming trend is the failure of the “average” film. In previous decades, a film with a mid-range hero (like Nani, Ravi Teja, or Vijay Deverakonda) could guarantee a minimum return. In 2025, films like Robinhood or Jack faced extreme volatility. The audience just asked “Is this an Event Film (IMAX/3D/Pan-India)?” or “Is this a Re-Release?” If the answer is neither, they wait for OTT.

This behavior eviscerates the training ground for new talent. Directors rarely start with ₹300 crore epics. They start with mid-budget films. If the mid-budget sector collapses, the ladder to the top is broken. The industry effectively creates a barrier to entry that only the children of established stars (nepo-kids) or directors with massive existing clout can cross.

4. Model Collapse and the Data Constraint

To deeply understand why this cultural stagnation is dangerous, we can look into the mechanics of Artificial Intelligence and get some inspiration. The phenomenon of “Model Collapse” in Large Language Models (LLMs) offers a perfect, terrifying metaphor for what is happening to Telugu culture.

4.1 The Mechanism of Model Collapse

Recent research by Shumailov et al. (2024) and others has identified a critical vulnerability in Generative AI. When an AI model is trained on data generated by previous versions of itself (synthetic data) rather than data generated by humans, the model degrades.

• Loss of Variance (The Tails): In the first stage of collapse, the model stops producing “rare” or “edge case” outputs. It forgets the nuances, the dialects, and the unconventional ideas that exist in the “tails” of the probability distribution.

• Convergence to the Mean: In the second stage, the model converges on a single, homogenized representation of reality. Its outputs become repetitive, bland, and increasingly detached from the actual complexity of the world.

• Irreversibility: Once a model has collapsed, it is extremely difficult to fix without injecting a massive amount of “pristine” human data.

4.2 Mapping Model Collapse to Telugu Culture

We can map this technical concept directly to the Telugu film industry:

1. The Dataset: The “Dataset” of Telugu culture is the sum total of stories, dialogues, and visual motifs produced by the industry. Currently, the industry is prioritizing the “Dataset of 2000-2010” via re-releases.

2. Recursive Training: The current generation of youth (the “New Model”) is consuming this old data. Aspiring directors are “training” their creative neural networks on Indra and Pokiri. They are not observing the world around them (the 2025 reality of Hyderabad, IT corridor stress, Gig economy, AI relationships, live in relationships, or rural agrarian shifts). Instead, they are observing the movies of the past and trying to recreate the same formulas(Balakrishna shouting, elevation of Pawan Kalyan, hero teasing the heroine etc)

3. Synthetic Output: When these young directors make a movie, they produce a “derivative” of the past. They recreate the “interval bang” or the “hero elevation” exactly as it was done in 2002. This is “synthetic data”, a copy of a copy.

4. Cultural Collapse: The result is a loss of “tails.” We lose the films that explore niche topics (e.g., a film about the Lambada tribes, or a cyberpunk thriller set in Cyber Towers, or a love story in the Tinder age, or sexual harassment in the new office, or effect of drugs in school campus etc). The culture converges on a single, distorted mean: The Pan-Indian Mass Action Film. The “nuanced variations” of Telugu life disappear, drowned out by the amplified patterns of the past.

5. The Telugu LLM Crisis: A Data Constraint Problem

The metaphor becomes literal when we look at the actual development of Telugu AI technologies. The stagnation of cultural content is creating a “Data Constraint” that hinders the development of intelligence systems for the Telugu language.

5.1 The Scarcity of “Fresh” Tokens

To build a “Telugu GPT,” researchers need billions of words of high-quality Telugu text. However, Telugu is considered a “low-resource” language in the context of high-quality digital data.

• Common Crawl Disparity: The Common Crawl corpus, which powers models like LLaMA and GPT-4, is overwhelmingly English (~45%). Telugu represents a tiny fraction, often less than 1% or even smaller depending on the filtering capability.

• Token Inequality: In multilingual datasets like mC4, Telugu is assigned significantly fewer tokens than European languages, despite having millions of speakers.

5.2 The “Re-Release” Effect on Digital Text

Where does training data come from? It comes from newspapers, blogs, social media conversations, and movie reviews.

• The Feedback Loop: If the most popular topic on Telugu Twitter is the box office collection of Indra 4K, then the “training data” being generated in 2025 is filled with the vocabulary of 2002.

• Domain Poverty: We lack diverse data. We have millions of words about “Hero Elevations” and “Fan Wars,” but we have very little data on Telugu Science, Telugu Economics, or Telugu Legal Theory.

• Stunted Intelligence: Consequently, models like Navarasa 2.0 (developed by Telugu LLM Labs) or OpenHathi (Sarvam AI) or Akshara(Viswam AI) face a ceiling. They can generate fluent Telugu, but their “worldview” is constrained by the data they are fed. If the culture feeds them nostalgia, the AI will hallucinate the past. An AI trained on Telugu movie reviews might answer a medical query with a dramatic dialogue about bloodlines, because that is the statistical probability of the language it has learned.

6. The Gerontocracy: Stars Who Won’t Let Go

A primary driver of this recursion is the refusal of the industry’s most powerful nodes, the Stars, to evolve. The Telugu industry is unique in its “Gerontocracy of Stardom.”

6.1 The Ageless Heroes (60s and 70s)

In 2025, the industry is still dominated by the same four names that dominated in 1990: Chiranjeevi (69), Balakrishna (65), Nagarjuna (65), and Venkatesh (64).

• The Refusal to Transition: Unlike Clint Eastwood or Harrison Ford, who transitioned into roles acknowledging their age, Telugu stars continue to play the “commercial hero”, romancing women young enough to be their granddaughters and performing gravity-defying stunts that defy physics and physiology.

• The Audience Complicity: The audience is complicit. When Venkatesh romanced a 28-year-old in a 2025 hit, it was accepted as “standard practice”. This acceptance reinforces the “Model Collapse”. The audience rewards the repetition of the “Young Hero” archetype even when the actor is old, rejecting the “New Data” of an aging protagonist.

• Comparison with Neighbors: Contrast this with Malayalam cinema, where Mammootty (73) is doing radical, experimental films like Bramayugam (black and white horror) or Kaathal (playing a closeted gay man). Of course Mammootty also does some pot boilers, but there is experimentation. The Telugu star system has no such flexibility. It is rigid, brittle, and terrified of breaking the “Hero Image”.

6.2 The Economic Stranglehold

These stars command the lion’s share of the industry’s capital. A film with a senior star locks up theaters, distributors, and media attention for weeks.

• The “Dates” Problem: Young directors wait years to get a meeting with these stars. When they do, they are often forced to water down their scripts to fit the “Hero Image.”

• The Budget Vacuum: The astronomical remunerations of these stars drain the production budget, leaving little for writing, VFX, or production design. This results in films that look “rich” but feel “cheap” in their storytelling.

7. The Stifled Voice: A Lost Generation of Creators

The victims of this system are the youth. The “Death Knell” rings loudest for the young writer or director who has a story about today but no screen to show it on.

7.1 The “Bhajana” Trap

The industry operates on a culture of “Bhajana” (sycophancy). Young directors are encouraged to praise the “Legends” rather than challenge them.

• Case Study: Venkatesh Maha: When director Venkatesh Maha (maker of the critically acclaimed C/o Kancharapalem) dared to criticize the mindless masculinity of the KGF franchise (a proxy for commercial mass cinema), he was viciously trolled and forced to apologize. The message was clear: Do not critique the model. Do not suggest the data is bad.

• Case Study: Tharun Bhascker: Hailed as a new-wave pioneer, Bhascker has expressed frustration with the industry’s risk aversion and the long gaps between his films. The system is not designed to scale up indie talent; it is designed to maintain the status quo.

7.2 The Lack of Expression

Because the theaters are occupied by the past, the youth have nowhere to express their present.

• No “Angry Young Man”: In the 1970s, Amitabh Bachchan channeled the anger of a generation. In 2025, where is the avatar for the frustrated Telugu software engineer or the unemployed graduate? They don’t exist. The youth are forced to vicariously live through the “Factionist” of 2002 because there is no “Gig Worker Hero” of 2025.

• The Writer Crisis: There is a severe shortage of new writers. The industry pays writers poorly and values “combinations” (Director + Star) over “scripts.” As a result, the narrative quality degrades further, forcing producers to rely on remakes or re-releases because “at least the story is tested”.

8. Historical Parallels: Lessons from Other Cultures

This phenomenon is not entirely unique, and history offers warnings.

8.1 France: The “Cinema de Papa” Rebellion

In the 1950s, French cinema was suffocating under the “Tradition of Quality”, highly polished, studio-bound adaptations of classic literature. It was safe, profitable, and culturally dead. François Truffaut famously attacked it as Le Cinéma de papa (”Daddy’s Cinema”).

• The Reaction: The French youth rebelled. They picked up handheld cameras, shot on the streets of Paris, and created the Nouvelle Vague (New Wave). They rejected the “Old Data” and created “New Data.”

• The Telugu Difference: The tragedy of Tollywood is that the youth are not rebelling. They are the biggest consumers of Cinéma de papa. They are cheering for the “Tradition of Quality” re-releases. This lack of rebellion is the true “Death Knell.” This has been the biggest surprise. If the best the youth can do is chant “Jai Balayya” in the pubs of Hyderabad, who will disrupt the system.

8.2 Hollywood: The Nostalgia Loop

Hollywood is currently suffering its own version of Model Collapse with the endless parade of Live-Action Disney Remakes and “Legacy Sequels” (Indiana Jones 5, Star Wars).

• The Warning: This strategy works for a decade (as seen with the MCU), but eventually, the audience fatigues. The box office crashes of 2024/2025 in Hollywood (e.g., The Marvels, Indiana Jones) show that eventually, the model collapses completely. Tollywood is accelerating toward this cliff.

9. Solutions: Debugging the Code

To fix this, we must treat the culture like a failing AI model. We need “Data Augmentation,” “Regularization,” and “New Architecture.”

9.1 Solution 1: The “Fresh Data” Quota (Policy)

The Telangana Government, specifically the Film Development Corporation (TSFDC), must intervene.

• Proposal: Introduce a “Cultural Diversity Quota.” Any multiplex with more than 4 screens must dedicate at least 25% of its showtimes to films with a budget under ₹10 Crores that are New Releases (not re-releases).

• Impact: This ensures that “Tail Data” (small films) gets a platform, preventing the “Model” from converging solely on the “Mean” (Blockbusters).

9.2 Solution 2: The “Clint Eastwood” Protocol (Star System)

The senior stars must take responsibility for the transition.

• Proposal: Stars over 60 should be incentivized to produce and star in “Age-Appropriate” narratives.

9.3 Solution 3: The Telugu Digital Heritage Corps (AI & Data)

To fix the “Data Constraint” for LLMs and Writers:

• Proposal: A massive state-funded digitization project. We need to digitize modern Telugu novels, legal judgments, scientific papers, and oral histories.

• Mechanism: Collaborate with startups like Viswam AI and IIIT-Hyderabad to create a “Golden Corpus” of Telugu text that is not movie-related. This gives future AI models a vocabulary beyond the cinema.

9.4 Solution 4: The “Writer’s Incubator” (Talent)

• Proposal: Shift the focus from “Directors” to “Writers.” Establish a “Writers’ Lab” (modeled after the Sundance Lab) funded by the major production houses (Suresh Productions, Geetha Arts).

• Goal: Develop scripts that are not written for stars. Create a bank of high-concept stories that can be made on mid-budgets. Pay writers a living wage so they don’t have to become directors to survive.

10. Conclusion: Breaking the Loop

The re-release of a beloved film is a moment of joy, a shared memory in a dark room. But when that memory becomes the dominant mode of culture, it is a sign of surrender. It is an admission that we have forgotten how to imagine the future.

The stars in my youth should not be the stars in my kid’s youth. Who is this generation’s Chiranjeevi? It absolutely must not be Chiranjeevi.

For the Telugu film industry, the choice is binary. It can continue to feed on its own tail, generating recursive profits until the model collapses and the audience moves on to other, more vibrant cultures (like Korean or Malayalam cinema). Or, it can make the painful, risky decision to stop the re-releases, retire the aging tropes, and invest in the chaotic, messy, unproven “New Data” of the youth.

The “Death Knell” is ringing. It sounds like the cheering of a crowd watching Indra for the hundredth time. The only way to silence it is to tell a new story.

Today we expand on those themes, focusing on the unique nature of software and the physical constraints of hardware.

In my previous post, we looked at how AI giants are performing a “Primitive Accumulation” of our digital lives, fencing off the open internet to build proprietary models. If we follow Marx’s trajectory, this leads to a “Metabolic Rift” where the digital soil is depleted, and creators are alienated from their own data.

But there is a flaw in the historical analogy that offers us an exit: Land is finite. Software is not. While the physical Enclosure Acts of the 18th century took away a limited resource (acreage), the digital enclosure is trying to fence off something that can, in theory, be infinitely replicated. This is where Open Source AI becomes the “Negation of the Negation”, a way to restore the commons without falling into the traps of the past.

1. The Great Divergence: Land vs. Code

Marx’s critique of capitalism was rooted in the limitation of the Earth. If a Lord fences off a forest, the peasant cannot simply “download” a copy of that forest.

In the AI era, we have a unique advantage:

• The Zero Marginal Cost of Reproduction: Once an AI model is trained and its “weights” (the learned intelligence) are released as Open Source, the “fence” vanishes.

• Consumption by Everyone: My use of an open-source model does not prevent you from using it. We are no longer fighting over the same patch of grass; we are sharing a sun that shines on everyone simultaneously.

By open-sourcing model weights, we effectively “de-commodify” the intelligence that was gathered from our collective data. It is the ultimate act of digital restitution.

2. The New Enclosure: The GPU Wall

However, we must be intellectually honest. While software is infinite, the Means of Production for AI are not. We are seeing a shift from the enclosure of data to the enclosure of compute.

Even if a model is Open Source, the “Metabolic Rift” persists if only the ultra-wealthy can afford the electricity and silicon to run or fine-tune it. This is the GPU Scarcity. If Open Source is to save us, it must focus not just on open code, but on efficiency, making models small enough to run on “the people’s hardware” (consumer laptops and local servers).

If I was running AI policy in India, these are the two things I would be concentrating on:

1. Data collection efforts.

2. Funding R & D for coming up with more efficient models.

3. Avoiding the Two Pitfalls

Open Source AI offers a “Third Way” that sidesteps the failures of the 20th century’s dominant systems:

Avoiding the Pitfall of Capitalism (Monopoly)

In a purely capitalist AI 1.0, 4-5 companies control the “Base Models” of human thought. They decide what is “safe,” what is “true,” and they charge a rent (subscription) to access the intelligence we originally provided. Open source breaks this monopoly by distributing the capability to everyone.

Avoiding the Pitfall of Communism (Centralization)

Traditional state-led communism often failed because it centralized power in a different way, through a state bureaucracy that controlled the “Commons.” Centralized AI, even if “publicly owned,” risks becoming a tool for state-mandated thought.

Open Source is Decentralized. It doesn’t belong to a CEO, but it also doesn’t belong to a Politburo. It belongs to the protocol. It is the “Negation of the Negation”: a restoration of individual property (your own local model) on the basis of communal possession (the open-source repository).

4. Healing the Metabolic Rift

The “Metabolic Rift” occurs when AI eats the internet and gives nothing back, causing “Model Collapse” as the digital soil becomes sterile.

Open source heals this by:

1. Transparency: We can see what data was used, ensuring “nutrients” are acknowledged.

2. Local Cycles: Instead of shipping all our data to a central “city” (the Silicon Valley clouds), we process data locally. The insights and value stay with the creator, much like the medieval peasant returning nutrients to their own local plot.

Summary: The Path Forward

The enclosure is not inevitable. While the “Cloud Lords” try to fence off the GPU clusters and the data lakes, the Open Source movement is building a “Digital Commons” that is more resilient than the physical ones Marx studied.

By making AI small, local, and open, we ensure that the “Original Sin” of data scraping leads not to a new era of digital serfdom, but to a collective leap in human capability.

ARR is king now. Churn, LTV, CAC. We dont hear those anymore.

And because AI startups are using the same vocabulary of SaaS startups, real SaaS startups are suffering. I think we need to come up with new vocabulary.

The industry is currently suffering from “Novelty Inflation,” where traditional SaaS metrics like ARR (Annual Recurring Revenue) are failing to distinguish between a product that is a toy and a product that is a tool. In fact, in many cases, its actually nefarious as founders are hyping up their products by making loose interpretations of the meaning of ARR(I have had some founders tell me they calculate ARR as the best day of the week * 365).

To solve this, we need a metric that discounts “curiosity capital” and focuses on “utility retention.” Please welcome Vibe Revenue Run-rate.

Let’s define and operationalize VRR: Vibe Revenue Run-rate.

Check your VRR

1. The VRR Formula: A Scientific Approach

Traditional ARR assumes that $1 earned today is worth $12 over a year. VRR challenges this by applying a **Utility Decay Coefficient (δ)**.

The Core Formula

Where the Utility Decay Coefficient (δ) is calculated based on three “Vibe-Check” variables:

1. Workflow Integration Score (W): Measures the depth of API calls or “boring day” usage versus “wow moment” usage.

2. Model Independence Factor (M): Measures if the user stays even when a better/cheaper model (e.g., GPT-5 or Claude 4) is released elsewhere.

3. Churn-Adjusted Velocity (Churn): Penalizes revenue from users who haven’t crossed the “Habit Horizon” (typically Month 5 in AI).

The delta Calculation

• If δ approx 1: The revenue is “Hard Revenue” (High Utility).

• If δ < 0.4: The revenue is “Vibe Revenue” (Pure Novelty).

2. The VRR Audit: How VCs Should Calculate It

To move from “Vibe” to “Value,” VCs must demand a Cohort Utility Audit during due diligence. Here is the framework for calculating VRR:

3. Why VRR is Superior to ARR for AI Startups

A. ARR Rewards System Of Record

Traditional SaaS is a system of record. So generally once a customer is onboarded they have an LTV(life time value). With AI you move to a system of action or system of work. The LTV is different here. The default calculation of ARR might still work well, but new definitions of ARR(best day*365 etc) don’t give the right picture.

B. The “Worst-Model-Ever” Fallacy

Founders argue that better models will fix churn. VRR proves that retention is a product problem, not a model problem. If the VRR is low despite high ARR, it proves the product has “Shallow Integration.”

C. Protecting Talent and Capital

Employees are currently taking equity in “Vibe Companies” based on 50x ARR multiples. VRR provides a realistic valuation framework that prevents “Paper Millionaire” syndrome by valuing companies on sustainable habits rather than temporary FOMO.

4. The Investor’s Defensive Playbook

If you are a VC, you should look for the “VRR Gap”—the difference between reported ARR and calculated VRR.

The Rule of Thumb: > * Healthy AI Startup: VRR is >70% of ARR.

• Vibe-Trap Startup: VRR is <30% of ARR.

The Vibe-Trap startup will show a beautiful exponential ARR curve, but the VRR curve will be flat or declining, signaling that while new people are joining the party, nobody is staying for dinner.

VRR Due Diligence Checklist

1. Usage Architecture

• Boring Day Ratio: What % of weekly active users log in on Tuesday-Thursday? (Goal: >60%)

• Session Intent: % of sessions that involve a multi-step workflow vs. a single prompt-and-exit.

• Feature Breadth: % of power users utilizing >3 non-chat features (e.g., exports, integrations, collaboration).

2. Retention & Habit Horizon

• The Month-5 Cliff: Retention rate of the Month 5 cohort vs. the Month 1 cohort.

• Natural Frequency: Does the product’s value proposition align with a daily, weekly, or monthly habit?

• “Wow” Decay: Average time spent in-app per user over their first 90 days.

3. Structural Moats (Switching Costs)

• Data Persistence: Does the user lose proprietary data or context if they switch models?

• Workflow Gravity: Is the tool the “system of record” or just a “utility window”?

• Integration Depth: Number of active API connections per seat.

4. Model Independence

• Input-to-Output Ratio: How much value is added by the product vs. the underlying LLM?

• Proprietary Context: Does the startup have a unique RAG/Fine-tuning layer that persists across model upgrades?

Case Study: ARR vs. VRR (Platforms like Lovable.dev / Vibe Coding)

The Narrative (ARR View)

• Reported ARR: $10M+

• Growth: 30% Month-over-Month.

• The Pitch: “We are the frontend for the AI generation. Users build production-ready apps in minutes.”

• Investor Deck: Shows a hockey stick of ‘Apps Created’ and ‘New Pro Subscriptions’.

The Reality (VRR View)

Platforms like Lovable (and tools like Bolt or Replit Agent) are currently the epicenters of Vibe Revenue because they rely on the “One-Off Project” trap.

The VRR Leakage Factors:

1. The “Tourist” Problem: A founder pays $20 to build a landing page. Once the page is done, the subscription is canceled. ARR counts this as $240 ($20 x 12), but VRR counts it as $20.

2. Model Fragility: If a user finds a better agent (e.g., Cursor + Composer), they switch instantly because their code is portable.

3. The Habit Ceiling: Building a new app is a high-energy, low-frequency event. VRR requires maintaining an app to be the core habit.

Calculating the VRR:

• Gross ARR: $10,000,000

• Utility Decay Coefficient (δ): 0.25 (High churn after project completion + low model independence).

• Calculated VRR: $2,500,000

Verdict:

While the company is “making” $10M, only $2.5M represents a sustainable, habit-formed business. The other $7.5M is “Vibe Revenue”—money paid for the magic trick of seeing an app appear, not for the labor of running a business.

To make this easier, I have vibe coded(I see the irony :)) a VRR calculation app.

Check your VRR

Karl Marx viewed the “commons”, the land, forests, and waters held in common by villagers for centuries as the site of a foundational tragedy that birthed capitalism.

His analysis of the commons spans his entire career, from his early journalistic defense of peasants’ rights to his late-stage technical analysis of “Primitive Accumulation.”

1. The Early “Theft of Wood” Debates

Marx’s first serious engagement with economics occurred in 1842, when he wrote about the Debates on the Law on Thefts of Wood. The provincial government in the Rhineland had begun criminalizing the gathering of fallen wood in forests, a “customary right” the poor had exercised for generations.

Marx argued that:

• Custom vs. Law: The state was transforming an ancient, communal “use-right” into a crime to protect the “private property” of landowners.

• The Privatization of Nature: For the first time, Marx saw how the law was being used to “fence off” nature from the people, turning a common resource into a commodity for sale.

2. “So-Called Primitive Accumulation”

In Capital, Volume 1, Marx explains how the commons were destroyed to create the modern world. He called this process Primitive Accumulation (or ursprüngliche Akkumulation).

To Marx, capitalism could not begin until two things existed: a pool of wealth in a few hands and a mass of people with nothing to sell but their labor. He argued the second part was achieved by forcibly stealing the commons.

• The Enclosure Movement: In England, landlords used “Enclosure Acts” to fence off common lands (fields, meadows, and forests).

• The Creation of the Proletariat: Once the peasants lost access to the commons, they could no longer feed themselves. They were “set free” from the land—meaning they were forced into cities to work in factories for wages.

• State Violence: Marx famously noted that this history “is written in the annals of mankind in letters of blood and fire,” highlighting that the commons weren’t “traded” away; they were taken by force.

3. The “Metabolic Rift”

Marx also looked at the commons through an ecological lens. He argued that when people were separated from the land (the commons), it created a metabolic rift.

• In the old common-field system, nutrients from the soil stayed local.

• In the new capitalist system, food was shipped to distant cities. The “waste” (human excrement) polluted the cities instead of being returned to the soil as fertilizer, leading to the depletion of the earth’s fertility.

4. The “Negation of the Negation”

Marx did not want to simply “go back” to the medieval commons. Instead, he envisioned a future where society reached a higher form of communal ownership.

He described this as the “Negation of the Negation”:

1. First State: Individual private property based on one’s own labor (small-scale farming/crafting).

2. The Negation: Capitalism destroys that small-scale property and the commons, concentrating everything into a few private hands.

3. The Negation of the Negation: A future socialist society that restores “individual property” but on the basis of “possession in common of the land and the means of production.”

Summary: For Marx, the “commons” represented a lost unity between humans and nature. Their destruction was the “original sin” of capitalism, and their restoration (in a modern, industrial form) was the ultimate goal of communism.

Now that you have read about the summary of Marx’s theories on the commons, if you have anything to do with AI you immediately know how all of the above maps to the Internet and the AI mega corps and the governments. All you need to know is that in our digital age, it’s the Internet which is the commons. But just for the sake of clarity let me lay out the mapping clearly.

Here is how Marx’s critique of the commons maps onto the transition from the Open Internet to the AI Era:

1. Data Scraping as “Primitive Accumulation”

Marx argued that capitalism didn’t start with honest trade, but with theft. He called the seizure of communal land “Primitive Accumulation.”

In the digital context:

• The “Commons”: For 30 years, humans built the internet: Wikipedia, Reddit, open-source code (GitHub), personal blogs, and art portfolios, largely for the sake of sharing, community, and “use-value.”

• The “Enclosure”: First we had the social media companies like Meta, Twitter, LinkedIn create walled gardens. That itself was a huge issue for the open Internet. But now, the AI companies have used web-crawlers to “fence off” this collective human knowledge. By ingesting this data into private, proprietary models, they have transformed a public resource (the open web) into private capital.

• The “Original Sin”: Just as the peasants didn’t consent to their land being fenced, the creators of the internet didn’t “consent” to their life’s work being used to train a commercial product that might eventually replace them.

2. Creating the “Digital Proletariat”

Marx’s central thesis on the commons was that once people lost access to the land, they were forced to sell their labor to the very people who took it.

• Loss of Independence: A writer or artist previously “owned” their means of production (their talent and a platform). They could reach an audience directly on the “commons” of the internet. (Of course, we can say the creators are not really independent even now what with Google, Youtube, Meta, Spotify controlling the distribution. But that’s a different fight :) )

• The Shift: Now, as AI-generated content floods the market, the value of human labor is being driven down. Many creators find themselves in a position where they must either work for the AI (as data labelers or “prompt engineers”) or compete against a machine trained on their own past work.

• Alienation: The worker is now “alienated” from their product. The worker now creates content only for the AI machine. Your data is used to build a model that you then have to pay a subscription fee to use.

3. Regulatory Capture: The New “Enclosure Acts”

In the 1800s, the British Parliament passed “Enclosure Acts”—laws that legally sanctioned the theft of the commons.

Today, we see a similar pattern in AI Policy and Lobbying:

• The “Moat”: Established AI giants are now calling for heavy regulation and licensing. While framed as “safety,” critics would argue this is a way to ensure that only the wealthiest companies can afford to operate.

• Legalizing the Theft: Once the data has already been scraped (the “accumulation” phase), these companies seek laws that protect their models while making it harder for new, open-source, or smaller competitors to catch up.

4. The “Metabolic Rift” of Information

Marx’s idea of the Metabolic Rift (where the soil is depleted because nutrients aren’t returned) is particularly relevant here.

If AI “scrapes” the internet but doesn’t “feed” it back, meaning, if AI kills off the blogs, forums, and journalism that provide the original data, the “digital soil” becomes sterile.

• The “Dead” Internet: If human creators stop posting because they aren’t being paid or recognized, AI will eventually start training on its own output (Model Collapse).

• The Crisis: This mirrors Marx’s warning about capitalism’s tendency to exhaust the very two sources of all wealth: the earth and the laborer.

Summary: Some have called this “Data Colonialism.” The internet was a communal “forest” where we all gathered. AI companies have cut down the trees to build a factory, and are now charging us for the wooden chairs they made from our own timber.

Disclaimer: I am not an expert on communism or Karl Marx. What I know is from my discussions with friends and family and my understanding of commons and communism.

Well, such a great ideal. Unfortunately, the world is not an ideal place. The above leads to an Incentive Problem. No one works because everyone gets the same.

In the same way that traditional communism struggles because rewards are detached from individual effort, the “AI-conomy” risks a collapse of the creative incentive structure.

The “Problem of Communism” maps directly onto the “Creator Problem” in the age of AI:

1. The “Free Rider” Problem → The “Training Data” Problem

• In Communism: You work hard, but the “free rider” who does nothing gets the same loaf of bread.

• In AI: You spend 10 years mastering oil painting. An AI “scrapes” your work (without paying you) and allows someone else to generate a similar image in 5 seconds. The AI company and the “prompter” are essentially free-riding on your years of uncompensated labor. So, the best option for me is to wait for someone to create art, and then just prompt it.

2. Lack of Incentive → The “Why Bother?” Paradox

• In Communism: If the state takes 90% of your extra harvest, you stop trying to grow extra food.

• In AI: If a creator knows their unique style will be instantly “cloned” and devalued by a thousand AI bots, the financial incentive to innovate disappears. If you can’t make a living being an original artist, you might choose a “safer” job, leading to a stagnation of human culture.

3. Quality vs. Quantity → The “Slop” Problem

• In Communism: Factories produced 10,000 bad shoes just to hit a quota, because there was no “profit” in making 1,000 great shoes. Factories that hit the quota (quantity) were rewarded. There was no reward for making 10 great shoes.

• In AI: We are seeing a flood of “AI Slop”—low-quality, generic books on Amazon, soulless LinkedIn posts, and “dead internet” art. When the cost of production is zero, the market is flooded with quantity, making it nearly impossible for a high-quality human creator to be seen or heard.

The “Tragedy of the Commons”

The biggest mapping is what economists call the Tragedy of the Commons.

• In communism it describes how individuals, acting in their own self-interest on a shared, finite resource (a “common”), will inevitably overuse and deplete it, leading to ruin for everyone.

• In AI, the “Commons” is the slop machine. AI needs fresh human data to stay relevant and “smart.” Remember Ghibli and its overuse? If AI makes it impossible for humans to get paid for creating that data, humans will stop creating.

• The Result: The AI eventually starts training on its own output (AI training on AI), leading to a “Model Collapse” where everything becomes a blurry, distorted copy of a copy.

How can we fix this?

Just as some communist countries moved toward “Market Socialism,” the creative world is looking for middle grounds:

• Data Royalties: Forcing AI companies to pay a “tax” or license fee to the humans they trained on.

• Proof of Person hood: Labels that say “100% Human Made” to restore a “premium price” for human effort.

• The “Opt-Out” Movement: Tools like Glaze or Nightshade that “poison” images so AI can’t learn from them, effectively “fencing off” the private property of the mind.

Nikhil from Medianama has a good write up on AI and copyright.

Reasoned by Nikhil Pahwa

AI and the unravelling of copyright (Part One)

I have stories on AI and copyright to tell…

Read more

7 months ago · 2 likes · Reasoned by Nikhil Pahwa

For two glorious years, OpenAI was the shiny, disruptive upstart, forcing the $2 trillion giant to panic. Now, the tables have turned so dramatically, it’s less a competitive reversal and more a full-scale corporate comedy of errors starring a non-technical CEO and a giant, insatiable financial black hole.

The Irony is Now a Circular Firing Squad

We all remember December 2022. Google CEO Sundar Pichai reportedly declared a “Code Red” after ChatGPT landed like a meteor in Mountain View, threatening the core Search business.

Fast forward. The Search results confirm that this time, the crimson alert flashed in San Francisco. OpenAI CEO Sam Altman, in an internal memo, has now declared a “Code Red.”

The reason? Google’s Gemini 3.

As Marc Benioff wrote publicly after testing the rival model: “I’m not going back. The leap is insane – reasoning, speed, images, video everything is sharper and faster. It feels like the world just changed, again.”

I know Marc has been tweeting like an AI influencer for a long time now, just like Reid Hoffman, but seriously.

Wow!

The Gorilla vs. The Cheerleader

Google’s move to simplify its fractured AI efforts—finally managing to get all its AI chatbots collapsed into one name, Gemini—signals a pivot from distraction to execution. With its own TPUs, deep integration into Search and Android, and a $100+ billion cash pile, Google is the ultimate full-stack Gorilla.

And Sam Altman? He’s the ultimate financier. He is great at talking, selling, and recruiting capital, but in this specific technical crisis, the difference a technical founder makes becomes glaringly obvious.

The hard truth is: Sama cannot do anything here. He cannot even inspire. He can raise money, yes, but he is playing with Gorillas who have a lot more money. And clout.

Sam Altman cannot do anything other than declare a code red.

The Futile Battle Plan

Altman’s internal memo laid out the revised plan: to allocate more resources to improving ChatGPT. But I don’t think resources are the problem for ChatGPT. Direction is the problem.

The problem is that none of the priorities mentioned below will make any dent in Google’s fundamental advantage:

• The plan is to allocate more resources to improving ChatGPT, with teams concentrating on features that allow more personalized interaction with ChatGPT.

• Other core priorities include:

  1. advancing the image-generation model Imagegen,

  2. improving model behavior,

  3. strengthening leaderboard competitiveness with systems like LM Arena,

  4. enhancing ChatGPT’s speed and stability,

  5. reducing unnecessary refusals to harmless questions.

Just go through the list above. Who has asked for any of these? How can they make money with any of these? And why can’t Google do a better job at these?

OpenAI is trying to iterate its way out of an infrastructure war. It’s like challenging an aircraft carrier to a boxing match.

The new plan of ads to get revenue will just kill the golden goose. ChatGPT will become AdGPT.

The Loss-Making Monster

The final, terrifying truth for investors is the economics. OpenAI is a loss-making monster.

The cost of chasing AGI is so staggering that even if the company achieves aggressive revenue targets, it will still drown in debt. New analysis projects that even IF it can hit $200bn revenue by 2030, it’ll still need over $200bn in funding just to stay afloat.

As an HSBC estimate noted, OpenAI needs to raise at least $207 billion by 2030 just to cover its massive compute commitments, a hole one analyst called a “dot-com on steroids.”

This isn’t a sustainable business model; it’s a government-backed R&D project being accidentally run by a startup.

The conclusion is inescapable, harsh, and realistic: It’s plain and simple. The world doesn’t need OpenAI.

The AI technology will continue to advance. The models, the chips, and the capabilities are now too decentralized and too important to rely on one hyper-leveraged startup. AI will survive and in fact will become better without OpenAI. It just won’t have the same high-burn rate.

It's being pushed down our throats. It’s in our search bars, our operating systems, and even our creative tools, whether we asked for it or not. It feels less like an upgrade and more like a force-feeding.

It doesn’t need to be this way. Technology can be adopted slowly. Organically. One piece at a time.

Right now, the frantic pace of deployment isn't about utility; it's about liquidity. It’s being shoved down our throats because some billionaires need to make some more billions before they die.

We don't owe them anything.

It is time to do AI the right way. The honeymoon phase of the hype cycle is over. We now know the limitations. We see the hallucinations. We see the errors. Let's pick the things which work and slowly integrate it into our lives. We don't need to do it this quarter just because some startup has to do an earnings call. We will do it if it helps us.

And let’s be clear: We don't need AGI (Artificial General Intelligence). We don't need a digital god. We just need software that works.

If the current models don't work? No problem. Let the researchers go back to the lab and do their jobs. We will continue doing ours. We might even generate more data for them in the process—but this time, we do it correctly. We will work with the creators, writers, and artists, instead of ripping off their life's work to feed the model.

I hear the complaints from the tech giants already: "But we bought too many GPUs! We spent billions on infrastructure! They have to be put to work!"

Not my problem.

I will use what creates value for me. I will not buy anything that is of no use to me.

I hear that’s called capitalism.

There are plenty of legitimate use cases for AI and enough places to make money without force-feeding the market. But I will not allow AI to be pushed down my throat just to justify your bad investment.