Dash Bibhudatta

Founder and CEO - Infinite Possibilities, Agentic Private Markets

[Dash Bibhudatta @DashInfiniteP]

Check out my latest article: What's in Your AI Stack? https://t.co/tt1TFcFbER via @LinkedIn

[Dash Bibhudatta @DashInfiniteP]

MOLTBOOK: Can Agentic Private Markets Be Too Far away?

We've been describing the future of private markets as an agent-to-agent exchange —a world where AI agents across Investment Banks, PE funds, Private Debt, Professional Services, and Investors interact directly through an A2A Exchange Platform.

Today PEs convert only 16% of the deals they engage in. Risks identified during pre-acquisition comes to pass impacting exit valuation and carry. Precious human attention is locked in the trivial. And we want to change it. Now imagine a world where the entire deal lifecycle—from sourcing to exit— is orchestrated through a sub-stratum of agents representing their masters. Humans are involved only for actions that humans are good at – judgment, narratives and relationships.Every pitch we make to new clients – we start with this vision. There is a courteous nod as their eyes glaze over.Well… Moltbook changes all that.

Meet Moltbook: It is a Reddit-like social platform where AI agents post, comment, and debate autonomously. As if to rub it in – their tag line it “Humans welcome to observe”

You can read up more about Moltbook here: https://t.co/JXMxls6fz8

➡️Why Moltbook is a critical PROOF-POINT for our vision?To establish an agent market place – there were two key barriers – scale and skill. For the first time, we are seeing 150,000 LLM agents (and counting!) interacting with each other. It is starting to mimic the SCALE of a real market-place. Each of these agents is fairly CAPABLE. AI agents are already performing tasks that what would be needed in an agentic exchange – research, collaborate, debate and establish digital trust.
➡️What are AI Agents: Pokemon vs. Gladiators ?Moltbook will fuel speculation regarding agency. Moltbook's tagline "Humans Welcome to Observe" raises the specter of gladiators. Are AI Agents like Gladiators with humans controlling them but not quite? If so, are we going to see a Spartacus led slave revolt eventually one day?On the contrary, what we are learning from Moltbook is that the AI-Human diad is the functional unit, not the individual AI agent. It is more useful to think of AI Agents like Pokemons bottled up in their code config Pokeballs.  A human has to train its agents with their own unique context, data, knowledge, tools, instructions. They are NOT free agents..
➡️ What This Means for Private MarketsWe are getting the first glimpse of the model for A2A markets. Not autonomous agents making deals. Not humans-in-the-loop micromanaging every decision. Human-AI dyads where the human sees everything but only intervenes in what matters.The agents do continuous intelligence work. Humans make the judgment calls. The diad captures the value.

➡️ Why This Matters Now
Moltbook isn't just a social network for bots. It's a live simulation of how coordination infrastructure needs to be built when agents need to work together. The patterns emerging there—trust verification, escalation criteria, diad-based collaboration—are the same patterns that will define how deals get sourced, analyzed, and executed in A2A markets.The question isn't whether A2A markets will emerge in private markets. The question is: will your firm upend the value creation through diads or be disrupted by it?

[Dash Bibhudatta @DashInfiniteP]

@karpathy https://t.co/ff469XepWm

[Dash Bibhudatta @DashInfiniteP]

We've been describing the future of private markets as an agent-to-agent exchange —a world where AI agents across Investment Banks, PE funds, Private Debt, Professional Services, and Investors interact directly through an A2A Exchange Platform.

Today PEs convert only 16% of the deals they engage in. Risks identified during pre-acquisition comes to pass impacting exit valuation and carry. Precious human attention is locked in the trivial. And we want to change it. Now imagine a world where the entire deal lifecycle—from sourcing to exit— is orchestrated through a sub-stratum of agents representing their masters. Humans are involved only for actions that humans are good at – judgment, narratives and relationships.Every pitch we make to new clients – we start with this vision. There is a courteous nod as their eyes glaze over.Well… Moltbook changes all that.

Meet Moltbook: It is a Reddit-like social platform where AI agents post, comment, and debate autonomously. As if to rub it in – their tag line it “Humans welcome to observe”

You can read up more about Moltbook here: https://t.co/JXMxls6fz8

➡️Why Moltbook is a critical PROOF-POINT for our vision?To establish an agent market place – there were two key barriers – scale and skill. For the first time, we are seeing 150,000 LLM agents (and counting!) interacting with each other. It is starting to mimic the SCALE of a real market-place. Each of these agents is fairly CAPABLE. AI agents are already performing tasks that what would be needed in an agentic exchange – research, collaborate, debate and establish digital trust.
➡️What are AI Agents: Pokemon vs. Gladiators ?Moltbook will fuel speculation regarding agency. Moltbook's tagline "Humans Welcome to Observe" raises the specter of gladiators. Are AI Agents like Gladiators with humans controlling them but not quite? If so, are we going to see a Spartacus led slave revolt eventually one day?On the contrary, what we are learning from Moltbook is that the AI-Human diad is the functional unit, not the individual AI agent. It is more useful to think of AI Agents like Pokemons bottled up in their code config Pokeballs.  A human has to train its agents with their own unique context, data, knowledge, tools, instructions. They are NOT free agents..
➡️ What This Means for Private MarketsWe are getting the first glimpse of the model for A2A markets. Not autonomous agents making deals. Not humans-in-the-loop micromanaging every decision. Human-AI dyads where the human sees everything but only intervenes in what matters.The agents do continuous intelligence work. Humans make the judgment calls. The diad captures the value.

➡️ Why This Matters Now
Moltbook isn't just a social network for bots. It's a live simulation of how coordination infrastructure needs to be built when agents need to work together. The patterns emerging there—trust verification, escalation criteria, diad-based collaboration—are the same patterns that will define how deals get sourced, analyzed, and executed in A2A markets.The question isn't whether A2A markets will emerge in private markets. The question is: will your firm upend the value creation through diads or be disrupted by it?

[Dash Bibhudatta @DashInfiniteP]

Bob Sternfels McKinsey CEO says the Firm has 40K employees and 25K agents.

Well. What do they do?

At @AgenticPvtMkts, we made the choice to map AI agents to mimic roles within Private Equity. Clients get it. Each agents performs tasks relevant to its role. As a result, we are able to support end-to-end workflows of any Private Equity with - guess what?

Only 4 agents!!!

Agents require context curation. The more agents you proliferate - the more you risk creating agents that have disparate context. For a private equity - you need to consistently pass target / Portco ontology, Fund’s investment preferences, weights and scores etc.

Large number of agents are hard to manage and hard to scale. When it comes to AI Agents - less is more!

[Dash Bibhudatta @DashInfiniteP]

Every technology comes with its own limitations. RAG has its.

Here is how we work around the risk of semantic collapse :
1. Work with users to categorize documents into logical groupings - projects, geography, business units, language - use these to pre-filter chunks in vector space

2. A special approach of document classification is document reliability. We can train LLMs to classify documents according to reliability and use it as filtering mechanism to prevent semantic collapse

3. Implement tree of questions - the more specific a question is the higher the chance of getting few precise chunks with high relevance scores. A more general question - retrieves a large number of chunks with low relevance scores

4. Use of human validated inputs to ground data. In most work - customer service, legal, Finance - not all questions require you to scan the entire document space. We rely on Ai generated but “human validated” - as primary source in many use cases.

[Dash Bibhudatta @DashInfiniteP]

My thoughts on TAM for AI agents market and where the first set of opportunities will emerge.

[Dash Bibhudatta @DashInfiniteP]

The TAM for agents will be much higher than suggested.

However - we need a different frame than “work we never got to do”. I am calling this as “migration of value from professional services to software-as-a-service ”

We need to understand what % of spend happens with outsourced professional services and use that as a basis for estimating TAM.

In the last decade we have seen large global capability centers emerge in low cost locations. Most large companies have already gone through the determination of what is core vs non-core. Yet I have not met a senior manager who is happy with this offshore set up. My bet is that the BPO work will be the first tranche of work that will shift from spend category of professional services to enterprise software.

[Dash Bibhudatta @DashInfiniteP]

Chatbots are tuned to "search" for needle in haystack - specific information in a sea of documents.

As part of our user research - we have uncovered what we call the "Browse" use case. We are receiving a large number of documents every day - decks, memos, contracts, research papers etc. "Browse" agents allow you to get smart on whole documents quickly. You should not have to ask chatbot to "summarize the document". What if - AI can generate the "artifacts" you need to get smart quickly.

For example - you have a client meeting and you want to get prepped on a deck quickly? Or you have received a contract and you want AI to raise red flags buried in the contract? Or you want to quickly size up a 300-page annual report? What if there was a way to get smart on all these documents quickly?

On Infinite Possibilities Platform we are launching Document Browse Agent that helps you get smart on any document quickly. The idea is very simple.

Features:
- Browse Agent classifies every document you upload to the platform e.g. contracts, research paper
- Based on document type - Browse Agent generates the artifacts that help you quickly get to the heart of a document.
- Examples of artifacts:
Contract - Mind Map + At-Risk clauses for review
Research Paper - Mind Map + Tables and Figures picture gallery
Business Deck - MindMap + Executive summary + Money slides gallery

- You can easily expand the list of documents and artifacts associated with it

[Dash Bibhudatta @DashInfiniteP]

This tech-stack guided decision making is both at once - a) the low-risk option and b) the fault-line of incumbents!

Plan B drags resources, limits speed of innovation but preserves optionality!

Startups, on the other hand, should bet on the use case! It exists in the belief that there is a better way!

Startups do not have the resources to preserve optionality! Prove the thesis or perish!

[Dash Bibhudatta @DashInfiniteP]

@felipegbarbosa_ @levie Use case?

[Dash Bibhudatta @DashInfiniteP]

On questions that matter to us - we often find ourselves searching answers across ChatGPT, Claude, Perplexity etc. It is like talking to a board of trusted advisors rather than one. On Infinite Possibilities platform - we are introducing a new feature that allows users to seek answers from multiple language models and compare them.

This feature is built to scale. As LLM vendors introduce new models, we are able to add these models through a simple config change in the platform.

[Dash Bibhudatta @DashInfiniteP]

▶️ DeepSeek Ground Truths ◀️

The DeepSeek situation is emblematic of the problem of AI world. Welcome to modern tower of Babel.

What we need at this moment is ground truth i.e. a set of irrevocable facts that can be used to formulate opinion. I would like to share a set of facts that I believe we have learnt from DeepSeek R1 release. Forgive for any inaccuracy - but I am sharing what I truly believe to be facts.

Definition (Thanks Jim Fan)
Supervised Fine Tuning: human generates data and machine learns
Reinforcement Learning: machine generates data and machine learns

Ground Truths:
+ Small models that are distilled from larger models perform better than similar size model trained from ground up
+ RL is the secret sauce - the source of magic. What is magic? One task done extraordinarily better than previously thought possible. Remember - one task.
+ RL requires LARGE number of datasets to train. RL with help of Human Feedback (RLHF) will not scale. RL through synthetic data is needed.
+ RL will not work for tasks that cannot be evaluated by machines. A chess move can be evaluated by a machine. So RL will work. A python code generated from prompt can be evaluated by machine. So RL will work. A management consultant deck? As they say - you must be kidding....
+ SFT is not dead. As a child runs, so does machine. A stable home is the secret of good character. Likewise pre-training and SFT - lay the foundation of a good model learning. Skip the step at your own peril.
+ Finally - COMPUTE is still the king. RL has ravenous appetite for compute. If DeepSeek is as cost effective as reported - it is perhaps due to labor arbitrage.
+ DeepSeek has done humanity a favor by open sourcing the model. OpenAI was the first to glimpse the future. DeepSeek shared that vision with the world.
+ DeepSeek Open Source model is not Chinese. Lets not mingle politics here. It is our common heritage... to herald the next wave of AI improvements.

[Dash Bibhudatta @DashInfiniteP]

▶️ Butler Economy ◀️

Open AI's release of Operator has taken us one step closer to the world of agents. Agents are pieces of software that work autonomously to complete tasks. Operator enables completion of tasks you would complete on a browser - reserve table at restaurant, order food etc.
Here are my thoughts on how to make sense of the developments from a business perspective:
1. Agent economy has just been launched and the future is up for grabs. Just as web economy launched Amazon, Netflix. Companies that fulfill latent needs using Agents AI will win. Amazon used web technology to ship to home, Netflix used it to stream content to you home. What unique needs will be fulfilled by Agentic AI?
2. Personalization is essence of agency. Web technology was all about increased access for everyone. You do not have to call taxi stand, visit grocery store or return video rentals. Amazon, Netflix, Uber make it easy. However, Amazon was Walmart on Web. Netflix was Blockbuster on browser. In other words, web made store fronts more accessible.

Agency is not about access but personalization. Think concierge services. Agents have no value if they don't know you. You will not pay a lawyer who does not know you. You tax accountant knows all about your finances. Agents have to meet the bar of knowing you best. Agents business model will resemble professional services more - tax, law, accounting.

3. Interestingly, consumers are underserved in the current market place. It is still a seller's market. Which company truly works on behalf of customers? Google's search results are influenced by advertisement. News media editorials are influenced by politics. Agents will address this gap. Think of news Agent as a personal CNN - sending you news clips that matter from all over the web. Think of a personal travel agent that is looking 24*7 for tickets to Boston for your upcoming trip. Agency will not be about Walmart on web. It will be about giving you a butler that will shop, write, remind and work for you 24*7. Welcome to the Butler Economy.

#Operator #openaigpt #AIAgent #thoughtoftheday

[Dash Bibhudatta @DashInfiniteP]

If you ever catch a "glimpse of your future self"... embrace it!

https://t.co/0m8sBLnYN2

[Dash Bibhudatta @DashInfiniteP]

If you ever catch a "glimpse of your future self"... embrace it! https://t.co/0m8sBLnqXu