Entrepreneur and software engineer
Hi, I’m 
Krish
I make AI agents reliable enough to trust with real work.
I wrote code by hand until 2024, when Claude Code turned up and made that a lifestyle choice. Since then I’ve been building what sits underneath AI agents: the software that plans their work, retries the steps that fail, checks the results and keeps the bill under control.
The model is rarely the hard part. Everything around it is, and most of that is distributed systems, which is what I worked on for years before agents came along.
Things I’ve written.
All writingLong reads on why software breaks and what to do about it. Most come with a small interactive demo, so you can break things yourself.
- An Agent Is a Model Plus a Harness. The Harness Now Rewrites Itself.Same model, same 89 terminal tasks. 58% inside one harness, 76% inside another that no person wrote. What a harness is, what it’s worth once somebody reruns the numbers, the five things a machine is now allowed to rewrite in one, and the results that should slow everyone down. One loop issued 43 warnings about what it might break. Five came true.
- AI That Improves Itself Is Real. Does It Compound?The phrase “recursive self-improvement” was on four arXiv papers in 2024 and a hundred and five in the first nine months of 2026. I read the ones that measured something. The improving is real. The one experiment that tested the recursion came back 0.780 against 0.782.
- Building Effective Autoresearch Systems: 2A system that runs its own experiments has about seventeen moving parts, and each one has a way of quietly lying to you. Here is every mechanism I found that stops it, in the order the loop runs them.
What I’m working on
Pensieve, a research system that improves itself.
I’m building Pensieve: automatic research and recursive intelligence. It proposes experiments, runs them, reads its own results and keeps what worked, and the recursive part is that what it learns changes how it searches next time.
The constraint I’ve set myself is that it has to optimise for outcomes with a real industry use behind them, not a benchmark number going up. Systems like this have discovered genuinely new things before, and what they share isn’t the clever search, it’s who marks the homework. Both parts of the autoresearch series on this site are me working out what that takes.
Got something that keeps breaking?
Tell me about it. I like hearing about AI agents, the infrastructure behind language models, distributed systems and early ideas, especially the messy ones. Email is the quickest way to reach me.