←
Mayur Chougule3,219 posts
Mayur Chougule avatar

Mayur Chougule

@_mayurc

building @vistaralabs agent ecosystem, @splitmarkets | ex @McKinsey

◷ Joined July 2010▧ 528 Photos and videos
Mayur Chougule avatar
Mayur Chougule@_mayurc·5h

in SWE-CoWork, Opus 5.5 found the right owner for 93% of first questions. yet 153 of 450 deferred threads never got a follow-up. finding the owner wasn't enough. the factory needs to carry those obligations through to completion. that's what we're building at @vistaralabs

Can a coding agent take over a teammate’s work in progress? Introducing SWE-CoWork, a benchmark for coding agents working as digital employees on long-horizon software engineering tasks. Each assignment starts with a short handoff in a simulated workplace modeled on real-world engineering workflows. Agents search task boards and versioned docs, ask teammates for missing details, and revise their code when a reply changes what needs to be built. Open questions and confirmed requirements can both be lost before the agent hands the work back. SWE-CoWork lets us trace them through the conversation, the patch, and the tests. 101 assignments across 101 real open-source repositories. 26 simulated teammates per workplace. 11 programming languages. 12 application domains. 17 workplace tools. 🌐 Website: tokenwave-ai.github.io/swe-c… 📄 Paper: arxiv.org/abs/2611.20218 💻 Code: github.com/TokenWave-AI/swe-… 🤗 Hugging Face Daily Papers: huggingface.co/papers/2611.2… (If you find it useful, we’d appreciate an upvote!)

1:36
Mayur Chougule avatar
Mayur Chougule@_mayurc·23h

the next big job for coding agents: killing legacy software without stopping the business. replace one capability, verify it, move its callers, delete the old code. log what changed and why. repeat. @martinfowler called it the strangler fig. a real software factory never stops running it, so nothing you own gets to become legacy.

Mayur Chougule avatar
Mayur Chougule@_mayurc·Nov 17

I want my software factory to make me obsolete as an engineer. review and challenge my specs, find flaws in its own implementation, reject bad work, prove the result works before shipping. my job is deciding what deserves to exist and why.

Thariq@trq212Nov 16

software factories feel like a temporary problem. the models will learn to build them for you the more interesting question to me is how to build a workspace where agents can collaborate with you

Vistara Labs avatar
⟳ Mayur Chougule retweeted
Vistara Labs@vistaralabs·Nov 16

Anything you see, can be built with Keki. Keep the work moving.

0:30
Mayur Chougule avatar
Mayur Chougule@_mayurc·Nov 16

Keki now handles the requests I'd always say yes to. Only the weird ones reach me.

0:19
Mayur Chougule avatar
Mayur Chougule@_mayurc·Nov 14

Same on my Mac running the factory. Last 30 days Vistara counted at least 17% of agent time waiting on me, across 2,395 asks. So Keki settles what my policy already covers and pulls me in only for the calls that are mine. Anthropic's report: www-cdn.anthropic.com/72628f…

Post attachment
Mayur Chougule avatar
Mayur Chougule@_mayurc·Nov 14

Anthropic measured the price of being a principal: its interactive Claude sessions sit around 80% idle, waiting for human input. Agents scale. A principal's hours don't. The tools that win in 2027 ask you less, and only for the calls that are yours.

Post attachment

Andrew D'Souza @andrewdsouza Nov 13 Article Nobody Trusts Your Agent If 2026 was the year of agents, 2027 will be the year of principals. Nobody has ever trusted an agent, and nobody ever will. An agent is something that acts on someone’s behalf. A principal is the one

Vistara Labs avatar
⟳ Mayur Chougule retweeted
Vistara Labs@vistaralabs·Nov 13

Agent swarms made easy

I don't want to spend the day coordinating agents. I want them working toward ambitious goals, in parallel, and I get pulled in only when a decision is actually mine. This is how I'm running development that way.

Mayur Chougule avatar
Mayur Chougule@_mayurc·Nov 12

Your software factory is already on your Mac. Keep your Claude, Codex, Cursor and Grok agents running wherever they already run. Vistara coordinates them as one team across projects and goals: handing off tasks, reviewing each other, shipping in parallel. Keki makes the calls within your rules. You watch and steer it live. Free Mac beta. No signup needed. vistara.dev/keki

0:48
Mayur Chougule avatar
Mayur Chougule@_mayurc·Nov 12

Your coding agents can now dream across Claude, Codex and Grok bots. While you sleep, Keki reads the day's sessions and updates the project's memory: what was decided, what got reversed, what's already done. Last night: 11 sessions, 164 turns, 3 new decisions, 2 old conclusions reversed. Open a new agent, type @, and pick up where the others left off.

Post attachment

Memory is what allows Claude to learn from experience. In a new Engineering Blog post, we share how “dreaming” helps Claude organize what it’s learned, correct its mistakes, and find what it needs more easily the next time. anthropic.com/research/compo…

Mayur Chougule avatar
⟳ Mayur Chougule retweeted
Mayur Chougule@_mayurc·Nov 5

yes. select -> handoff. here's an earlier recording

0:14
Mayur Chougule avatar
⟳ Mayur Chougule retweeted
Mayur Chougule@_mayurc·Nov 7

We built Keki to help prepare those decisions. It supervises coding agents and Grok Bots, bringing what needs me into one Today queue. Here, an agent reports a build complete, but Keki flags missing tests after recent changes. One click sends the follow-up back; I then hand the build summary to the lead session for review.

0:19
Vistara Labs avatar
⟳ Mayur Chougule retweeted
Vistara Labs@vistaralabs·Nov 11

Different models have different strengths. There's literally no reason you can’t use them all. 🔴Grok for real-time search and what’s happening on X 🔴Claude Code for architecture and reasoning 🔴Codex for the actual shipping

Post attachment
Mayur Chougule avatar
Mayur Chougule@_mayurc·Nov 9

I guessed my coding agents need me about 1,500 times a month. That was on a call @boardyai set up with a head of engineering. Afterwards I checked Keki on my Mac: 2,035 asks in the last 30 days. What I didn't expect: 49h 48m of finished work sat waiting for my next prompt while I was at the desk.

Post attachment

I don't want to spend the day coordinating agents. I want them working toward ambitious goals, in parallel, and I get pulled in only when a decision is actually mine. This is how I'm running development that way.