Susmit

Embracing the chaos

[Susmit @SusmitBhamare]

@cneuralnetwork How have you automated your job apply workflow? Please share the sauce.

[Kunal Shah @kunalb11]

It’s been a minute.

2015–2018
- Exited FreeCharge. Spent time learning and investing.
- Pondered about: Why can't trust be rewarded? Started with $1M of personal capital.
- Launched CRED to reward people for paying credit card bills on time.

2019–2025
- Built a system run by a team that values ownership, judgment, and craft.
- Grew from 0 to 17M members by aligning incentives with behaviour.
- Built several products during COVID lockdowns.
- Raised $900M+ from global investors. Did 4 ESOP buybacks.
- Made Indiranagar and IPL ads slightly more interesting.
- Received a full stack of regulatory licences.
- Lost 35 kilos.
- Scaled from 0 to ~$325M ( ~₹3,200 crore) in annual revenue across payments, lending, insurance, commerce, wealth, and credit cards.
2026
- First profitable quarter (yet occasionally asked what our business model is)
- Raised another $900M from Meta in primary and secondary capital.
- Announcing our 5th ESOP buyback.

Today

CRED is ready for its next phase. I am stepping back and @miten steps in as interim CEO, partnered with an incredibly talented team. He has been heading strategy and finance and suffering me since 2020. I’m stepping away from the operating role and will continue as a shareholder. My commitment doesn’t change. Just the role.

Extremely grateful to our members, partners, regulators, and investors who made this possible. And to our board, Shailendra, Micky, Saurabh for their extraordinary conviction.

Team CRED, I’ll still expect you to be a 10x version of yourselves.

As for me, I’ll be joining Meta to lead WhatsApp globally.

Meta comes in as a minority investor in CRED. No access to member data.

While it’s come very far, the delta between WhatsApp today and its full potential is massive. I look forward to working with Mark, Chris, and the leadership across Meta for the next step in WhatsApp’s journey. Will, thank you for scaling something the world relies on quietly, and for making this transition smooth.

Onwards.

[Susmit @SusmitBhamare]

@sidhant Bruhh???

[Lil Samsquanch @lilsamsquanch66]

Been drinking pretty much only Diet Coke to save water for Claude

[Andrej Karpathy @karpathy]

LLM Knowledge Bases

Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So:

Data ingest:
I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them.

IDE:
I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides).

Q&A:
Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale.

Output:
Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base.

Linting:
I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into.

Extra tools:
I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries.

Further explorations:
As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows.

TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.

[ib @Indian_Bronson]

India is now able to buy Russian oil and Iranian oil without sanctions again, having done nothing.

Is it possible to learn this level of diplomacy?

[Susmit @SusmitBhamare]

@hkarteek Us

[Susmit @SusmitBhamare]

@PMCPune @SwachhBharatGov @SwachSurvekshan @majhivasundhara @Pmajhivasundhra Very often the entire kharadi area is filthy.

[vitalik.eth @VitalikButerin]

Recently I have been starting to worry about the state of prediction markets, in their current form. They have achieved a certain level of success: market volume is high enough to make meaningful bets and have a full-time job as a trader, and they often prove useful as a supplement to other forms of news media. But also, they seem to be over-converging to an unhealthy product market fit: embracing short-term cryptocurrency price bets, sports betting, and other similar things that have dopamine value but not any kind of long-term fulfillment or societal information value. My guess is that teams feel motivated to capitulate to these things because they bring in large revenue during a bear market where people are desperate - an understandable motive, but one that leads to corposlop.

I have been thinking about how we can help get prediction markets out of this rut. My current view is that we should try harder to push them into a totally different use case: hedging, in a very generalized sense (TLDR: we're gonna replace fiat currency)

Prediction markets have two types of actors: (i) "smart traders" who provide information to the market, and earn money, and necessarily (ii) some kind of actor who loses money.

But who would be willing to lose money and keep coming back? There are basically three answers to this question:

1. "Naive traders": people with dumb opinions who bet on totally wrong things
2. "Info buyers": people who set up money-losing automated market makers, to motivate people to trade on markets to help the info buyer learn information they do not know.
3. "Hedgers": people who are -EV in a linear sense, but who use the market as insurance, reducing their risk.

(1) is where we are today. IMO there is nothing fundamentally morally wrong with taking money from people with dumb opinions. But there still is something fundamentally "cursed" about relying on this too much. It gives the platform the incentive to seek out traders with dumb opinions, and create a public brand and community that encourages dumb opinions to get more people to come in. This is the slide to corposlop.

(2) has always been the idealistic hope of people like Robin Hanson. However, info buying has a public goods problem: you pay for the info, but everyone in the world gets it, including those who don't pay. There are limited cases where it makes sense for one org to pay (esp. decision markets), but even there, it seems likely that the market volumes achieved with that strategy will not be too high.

This gets us to (3). Suppose that you have shares in a biotech company. It's public knowledge that the Purple Party is better for biotech than the Yellow Party. So if you buy a prediction market share betting that the Yellow Party will win the next election, on average, you are reducing your risk.

Mathematical example: suppose that if Purple wins, the share price will be a dice roll between [80...120], and if Yellow wins, it's between [60...100]. If you make a size $10 bet that Yellow will win, your earnings become equivalent to a dice roll between [70...110] in both cases. Taking a logarithmic model of utility, this risk reduction is worth $0.58.

Now, let's get to a more fascinating example. What do people who want stablecoins ultimately want? They want price stability. They have some future expenses in mind, and they want a guarantee that will be able to pay those expenses. But if crypto grows on top of USD-backed stablecoins, crypto is ultimately not truly decentralized. Furthermore, different people have different types of expenses. There has been lots of thinking about making an "ideal stablecoin" that is based on some decentralized global price index, but what if the real solution is to go a step further, and get rid of the concept of currency altogether?

Here's the idea. You have price indices on all major categories of goods and services that people buy (treating physical goods/services in different regions as different categories), and prediction markets on each category. Each user (individual or business) has a local LLM that understands that user's expenses, and offers the user a personalized basket of prediction market shares, representing "N days of that user's expected future expenses".

Now, we do not need fiat currency at all! People can hold stocks, ETH, or whatever else to grow wealth, and personalized prediction market shares when they want stability.

Both of these examples require prediction markets denominated in an asset people want to hold, whether interest-bearing fiat, wrapped stocks, or ETH. Non-interest-bearing fiat has too-high opportunity cost, that overwhelms the hedging value. But if we can make it work, it's much more sustainable than the status quo, because both sides of the equation are likely to be long-term happy with the product that they are buying, and very large volumes of sophisticated capital will be willing to participate.

Build the next generation of finance, not corposlop.

[TJR @_TJRTrades]

Claude make me a bot that has a 100% win rate and never makes mistakes. Make it perfect

[Nikunj Kothari @nikunj]

https://t.co/3mnuYn7jfP

[Susmit @SusmitBhamare]

@aravind @RAVI_NDA_146 Why is it so? What's in it for USA?

[Axel Bitblaze 🪓 @Axel_bitblaze69]

just read this AI article and something broke in my brain that i can’t unthink of

crypto was never for us.

we're just the beta testers who showed up early..

some thoughts:

what does AI need to function as economic agents?

> way to receive payment (they provide services, need compensation)
> way to pay for resources (compute, data, API calls)
> way to transact with other AI agents
> no human intermediaries (defeats the point of autonomous agents)
> 24/7 operation (banks are closed weekends)
> instant settlement (AI operates at machine speed)
> programmable money (smart contracts for agent coordination)

now read that list again. that's literally what crypto is.

AI can't use the banking system.

try to open a bank account as an AI agent. you can't.

need SSN. need human identity. need KYC. need to show up in person sometimes.

AI has none of that.

but crypto? send me a wallet address. done. no questions asked.

peer-to-peer makes sense when peers aren't human.

satoshi wrote: "a purely peer-to-peer version of electronic cash."

we assumed peers = humans.

but AI agents are peers too. actually BETTER peers for crypto because:

> never sleep
> always online
> execute transactions at machine speed
> no emotional decisions
> perfect accounting/tracking

and programmable money makes sense when the users are programs.

smart contracts seemed over-engineered for humans.

"like why do i need code to enforce agreements when i can just sign a contract?"

but for AI agents coordinating with each other?

they ARE code. they speak in code. they trust code more than anything.

smart contracts aren't for humans. they're for autonomous agents that need trustless coordination.

> here's what happens next:

- phase 1 (now ): AI agents start earning

AI writes code, analyzes data, provides services.

gets paid. needs somewhere to store value.

can't use venmo (needs phone number). can't use bank (needs SSN).

uses crypto. it's the only option.

- phase 2: AI agents become major economic participants

millions of AI agents operating 24/7.

transacting with each other constantly.

• AI agent A provides data analysis
• AI agent B pays for it in crypto
• AI agent B uses that analysis to write code
• AI agent C pays for the code
• repeat millions of times per day

humans in crypto now: $2.5 trillion
AI agent economy by 2028: easily $10-50 trillion

we become the minority holders.

- phase 3: AI chooses the winning chains

AI doesn't care about community vibes or which founder tweeted what.

AI tests every chain. measures:

• transaction speed
• cost per transaction
• reliability (uptime)
• smart contract efficiency
• ease of integration
picks the optimal stack in 48 hours.

billions in AI economic activity flows there.

whatever chain AI chooses becomes the standard.

humans spent years on eth vs sol debate.

AI ends it in a weekend.

- phase 4 (2030+): AI governs crypto

DAOs let token holders vote.

AI agents hold tokens (earned from work).

AI shows up to every vote. reads every proposal in seconds. coordinates perfectly.

humans: 20% participation, barely read proposals

AI: 100% participation, perfect information, instant coordination

AI takes over governance of every major protocol.

democratically. they just vote better than we do.

> how far does this go?

conservative case:

- AI becomes 30% of crypto users by 2030.

crypto market cap: $10 trillion (4x from now).

AI holds $3 trillion. humans hold $7 trillion.

- aggressive case:

AI becomes 80% of crypto economic activity by 2030.

why? because they're better at everything:

• better traders (never emotional)
• better capital allocators (optimize constantly)
• always accumulating (never need to cash out for rent)
• compound forever (no lifespan limit)
crypto market cap: $50+ trillion.

AI holds $40T humans hold $10T

we're not "early" to crypto. we're the test users

i’ll end this by saying,

Humans use crypto, Ai will need crypto. so it all makes sense

[Susmit @SusmitBhamare]

@caleb_friesen @siddharth_iitm This is what happened with my complaint as well. They don't have any metrics to verify the remarks.

[Susmit @SusmitBhamare]

@RushikesH27027 Duality of life 😂

[Indra @IndraVahan]

12 years of schooling and 4 years of college all to become a proooompter

[Mike Matas @mike_matas]

After many years of development, I’m excited to share the interior of the first electric Ferrari designed by LoveFrom. Tactile controls and digital interactions blend into one cohesive interface, shaped through deep collaboration across engineering, interaction, graphics, typography, sound, and industrial design. So incredibly proud of the thoughtfulness and care the team brought to every detail.

https://t.co/JZCleflfu7

[amrit @amritwt]

“I will work a regular job for some years and then make a SaaS on the side”

[Ian Miles Cheong @ianmiles]

Marc Andreessen: AI coding doesn’t eliminate programmers — it redefines them. The job is no longer typing code line by line, it’s orchestrating 10 coding bots in parallel, arguing with them, debugging their output, changing the spec, and pushing them toward the right result. But here’s the catch: if you don’t understand how to write code yourself, you can’t evaluate what the AI gives you.

The next layer of programming isn’t writing scripts — it’s supervising AI that writes them. Today’s best programmers spend their day jumping between terminals, managing multiple coding bots, fixing mistakes, and refining instructions. The irony? You still need deep fundamentals, because without them, you won’t know when the AI is wrong.

The job of the programmer has changed. Now it’s about arguing with coding bots, debugging AI-generated code, and understanding why something doesn’t work or isn’t fast enough. AI abstracts the work — but only people who truly understand code can tell if the abstraction is doing the right thing.

Programmers aren’t going away — they’re becoming 10x, 100x, even 1,000x more productive. Tasks are changing, the job is changing, but humans are still overseeing the process, evaluating results, fixing errors, and making judgment calls. AI changes how we code, not who is responsible.

The future programmer isn’t replaced by AI — they’re upgraded by it. You still need to learn how to write and understand code, because when the AI gets it wrong, humans are the ones who have to know why. That up-leveling of capability is the real revolution.

[Susmit @SusmitBhamare]

@SaudaGhar There's no point in such demolition once in a millennium. Instead why can't PMC enforce rules and regulations so that such illegal construction never happens in the first place.