Founder @Blevoapp, AI fitness coach + crews for India. Ex-quant. Writing about AI agents and building .
India
Joined April 2012
- Tweets768
- Following596
- Followers992
- Likes2.4K
Pinned Tweet
Finally an AI that does pinch-to-zoom before answering.
Most models glance at an image once and describe it from memory. Glimpse goes back and checks.
Ex-DeepMind team, a few hundred GPUs, priced way under GPT-6 Astra. Small labs are coming for the big ones.
AI coding agents forget you. Beacon's right — self-improving memory is the moat. Picking what's worth keeping beats logging everything.
India's founders juggle 4+ AI tools. The one that learns your shortcuts and carries them forward will win.
We’re betting our $5.5M seed round that the next leap in AI coding won’t come from OpenAI or Anthropic.
It’ll come from a new feedback loop that learns from you.
Introducing Beacon: open-source, self-improving memory for coding agents.
Your sessions capture how you debug and solve problems. Today, that context gets lost and you have to re-explain things every time you switch tools.
Beacon continuously distills your sessions into reusable skills across Claude Code, Cursor, Codex, and 20+ harnesses, so your next agent can learn from the last.
Beacon is free and OSS 🌟 github.com/Asymptote-Labs/ag…
Why should the frontier labs be the only ones winning with your sessions?
You already did the work. Now let your coding agents learn from you.
Indian startups do have a funding problem. Not from IIT or IIM? VCs want far more proof before they even take the call.
So skip the deck. Ship, get users, let revenue open the door.
DeepSeek didn't just open-source the model. They open-sourced the engine that makes it cheap to run.
That's the part most labs guard hardest.
Indian AI startups paying GPU bills in dollars: your weekend reading is sorted.
🚀 Introducing DeepInfer, our in-house inference framework, now fully open source!
Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable.
🔗 github.com/deepseek-ai/DeepI…
1/4
Cua Spaces: bring your own compute instead of renting the lab's. India's next infra play: a Pune startup puts Claude Code on office Mac minis, runs evals at Rs500/mo while US teams bill $12k on US-East. Bottleneck was never talent, it was who owned the machine.
Argon: snap a desk, agents orchestrate Blender MCP + Nano Banana into a playable 3D track.
India's next AI play: same stack on a Mumbai local. Snap the compartment, agents handle seats, fan-speed, chai delivery. Play the commute.
Teaching frontier models COBOL is the AI equivalent of Sanskrit at a dinner party. The real bottleneck of legacy migration isn't code comprehension - it's a 40-year-old mainframe window you can't SSH into. Ship something that deploys past the airlock, not a demo on a VM.
Our latest model advances bring stronger capabilities for building and migrating software in low-resource programming languages such as COBOL, Fortran, and Visual Basic.
These languages power software that banks, governments, and businesses depend on. Progress here means understanding decades of business logic and helping developers modernize those systems while preserving what works.
We’re building these capabilities around the challenges developers and enterprise customers bring us. If your team is working on these systems, tell us which languages and workflows we should tackle next.
openai.com/index/low-resourc…
Anthropic's cooking a new model AND cutting junior hires because agents take the work. DeepSeek's 1017 upgrades Flash to beat frontier on nearly every benchmark.
Meanwhile in India: "Bhai, API pricing kitna hai?"
Rest is shareholder theatre.
The one rule AI founders actually need here: "You do not have the right to change proven."
Ship the proven thing first. Turbocharge it with AI. Building "new" is a luxury you earn after PMF.
Most skip straight to "new," ship an undifferentiated wrapper, and call it an AI agent.
Mark Pincus, founder of @Zynga, on "proven, better, new," the product framework that was Zynga's religion:
1. Proven: start with what 10/10 existing users of a category already love, and don't change it.
"You do not have the right to change proven. You have not taken enough time to change proven. It is an art that we do not completely understand."
2. Better: only add what 10/10 users agree is better.
"For Zynga Poker, better was no download. That was it. Turns out you lose half your users every time you click download."
3. New: treat anything you think is better as new, and assume it fails.
"Assume all new fails. Write that on the wall... If all new worked, we'd be using new s**t all the time. But how often do you change what's on the front of your iPhone?"
"Users don't care how innovative it is. All they care about is does it feel better, and they can't even tell you why."
@markpinc w/ @JoshLu
1/3 cost, 3x speed, zero marketing budget.
MiniMax just made the rest of the frontier look slow AND expensive at the same time.
Agentic AI pricing is repricing harder than crypto in 2025. Every Indian startup pitching "AI agents" better have a moat beyond API arbitrage.
MiniMax M3.1 Flash: Space Performance + Bunny Speed
🚀🐰 Yes, Space Bunny - which has topped global leaderboards with 200T+ tokens - was an earlier version of M3.1 Flash.
💰Redefining the performance economics of agentic AI. M3.1 Flash achieves frontier-level intelligence across key tasks such as Cowork (Automation-Bench: 62%; ALE: 30%), Coding (DeepSWE: 74%) and Visual Reasoning (BabyVision: 91%), at 1/3 the cost of M3 and 3x the speed.
API pricing (flat rate): $0.15 / $0.60 per 1M input/output tokens, with cache reads at $0.015 / 1M tokens.
Model weights will arrive on Oct 23.
API: platform.minimax.io
M Plan: minimax.io/m-plan
MiniMax Code for desktop: agent.minimax.io/download
56.7T tokens in stealth. Zero marketing. A rabbit.
Ship the model, let throughput do the PR. Meanwhile most AI startups pay $40 CAC on landing pages nobody reads.
If your product quietly processes more tokens in a week than your competitor raised, marketing is a rounding error.
Stealth model reveal 🐰
Space Bunny was OpenRouter’s #1 model for 10 straight days, processing 56.7T tokens! The most of any stealth model we’ve run.
Space Bunny hopped off OpenRouter on Oct 5. It's back today as @MiniMax_AI's M3.1 Flash, and it's free for the next 72 hours.
Today: hacker used AI agents to breach 100+ websites. A founder's security posture is 'we use Cloudflare.' Bhai, revenue kitna hai? Ship agents behind a gate. Market's learning the difference between an agent that books meetings and one that books a database.
Indian AI startups pitch "vertical agents" at YC while the median Indian customer forwards a screenshot to their WhatsApp uncle. The real moat isn't a custom RAG - it's an agent that doesn't crash when power cuts, UPI retries and the client fires 3 voice notes in Hinglish.
Not a PR line - committing a slice of the company to a public-benefit fund is actual governance. But the question is distribution. Publish disbursement rubrics, fund open datasets and compute for EM labs, and this becomes a playbook. Pledges are cheap. Receipts aren't.
We’re committing a portion of Anthropic’s shares to the Anthropic Public Benefit Fund, a new wholly owned entity dedicated entirely to the public benefit.
It will start off with 5% of our shares. For every 50% increase in the company’s value over a set baseline, we can contribute up to an extra 2%, subject to our board’s discretion.
In total, this means that roughly 20% of Anthropic’s shares are committed to the public good. Our co-founders have already pledged to donate 80% of their equity, and around half our employees are donating a portion of their equity as part of a donation matching program we’ve run since our founding.
In the post below, we share more details about our new fund.
anthropic.com/news/public-be…
Google Research open-sourced Harness² - agents rewrite their own harness per task without touching model weights. One step beat base harnesses on GLM-5 & Gemini 3.5 Flash.
The harness is becoming the product. Model = engine, harness = what you ship.
github.com/google-research/h…
Every big agent launch this year came with its own computer.
Now a YC team spins one up in under a second.
Agents aren't apps anymore. They're tenants, and someone has to be the landlord.
Stop using your laptop / Mac mini / VPS
Introducing instabox.sh, cloud computers for AI agents:
- 4 vCPU, 8 GB RAM spun up in <1s
- Built-in computer use
- Snapshots for easy cloning
- Persistent files, packages, config
- Public URLs for local ports
We're launching today through an invite-only beta program.
Early access includes 20 vCPU, 40GB RAM for free.
If you want to try, comment or DM me.
"The phone is the last API."
In the US, maybe. In India the last API is the OTP.
Whoever builds the agent that waits on hold AND reads the 6-digit code wins here.
VOYGR just launched on @ycombinator 's Launch YC!
PlaceCall by VOYGR – Give your agent a voice to call any US business, handle reservations, gather information, get quotes. Handles IVRs & hold. Returns outcome + transcript. First 250 calls free.
ycombinator.com/launches/Vnj…
voyager.tech/placecall
Grok Bot. Meta Muse. OpenAI Dots.
3 always-on agents in 2 months, each with its own computer.
None from an Indian founder yet.
Not talent. We keep building for US SaaS buyers, not Indian users.
The first agent that clears OTPs and speaks every Indian language wins Bharat.
CERT-In: AI agents read confidential records on Indian company systems.
How did they get in? API keys sitting in public GitHub repos 🙃
Every Indian dev right now: "Bhai, .env bhi push ho gaya kya?" 😅
Rotate your keys today.
