Swarochish Chekuri

What I was wrong about

I used to underestimate how much relationships, conflict and emotional intelligence affect the quality of execution. There were years when I thought being right about the problem, doing good work and being able to defend the reasoning should carry more weight than they actually do. It does not work that way, and difficult personality conflicts affected my own performance more than I would have admitted at the time.

Resolving disagreement is not a distraction from the work. Inside an organisation it is part of the work. I still value intellectual disagreement, probably more than most people. What changed is that I no longer confuse being willing to disagree with being indifferent to how disagreement is handled. You can hold a hard position without making the relationship collateral damage. I learned that later than I should have.

How I actually work

I still build things myself. Python is the constant, with a lot of TypeScript and web work, graph and knowledge systems, agent runtimes, databases, local models, infrastructure, and increasingly Rust where it earns its place. Claude Code has changed the level I can work at, not as a substitute for knowing what I am doing, but because it lets me move between product, architecture, implementation, testing and experimentation much faster than before.

I prototype. I write code. I break things. I benchmark them. I throw things away. I start again. Nothing here should suggest I sit above the work drawing architecture diagrams while somebody else builds it.

I build to understand something. Sometimes the experiment becomes a product and sometimes it answers the question and dies, and both are useful. The recurring ideas are a better description of the work than the repository names are.

• Agent orchestration
• Persistent memory
• AI-native programming
• Software-building systems
• Knowledge graphs
• Local and private AI

Most of it is private, some is public, and a fair amount should probably never have become a repository. It is on GitHub, including a journalism toolkit of thirty specialist agents for investigative reporting and fact-checking, MIT licensed.

Reading, and who I pay attention to

I read widely and I have no five-book canon that made me who I am. I read for mental models rather than allegiance. Product, psychology, technology, history, philosophy, systems, design. If an idea changes how I see a problem I take it, stress-test it against my own work, and discard the rest. The pattern matters more than the catalogue.

There are people whose work I keep returning to, which is not quite the same as following them. Steve Jobs, less for the mythology than for the insistence that technology, product, interaction and taste cannot be separated, and that a person should not have to understand the machinery underneath a product. Kent Beck, for treating software as something that evolves through feedback rather than something that becomes correct because the architecture looked elegant on day one. Martin Fowler, for spending decades making complicated software ideas legible without making them simplistic. Andrej Karpathy, when he shows what is actually happening at the edge of AI engineering rather than commenting on it.

What they share is that they make and explain. I am much less interested in people who only have opinions about making.

Outside it

I box. Part of the appeal is that there is nowhere to hide in it. You can understand the mechanics perfectly well and find out in about four seconds that knowing what to do and being able to do it under pressure are entirely different things. That distinction probably appeals to me for obvious reasons. Before boxing there was jiu jitsu, and alongside it strength training and rock climbing.

I travel when I can, and I am more interested in experiencing a place than getting through its landmarks. I will spend an afternoon in a local restaurant or a neighbourhood working out how people actually live there. Art has become a larger interest than I expected. I can spend a long time on how the same subject changes across medium, period and circumstance, with no ambition to know enough art history to sound knowledgeable about it. I like noticing things.

I keep choosing hobbies I am bad at and interested enough to become less bad at.

I am curious to a degree that is sometimes impractical, and can go from agent runtimes to etymology to architecture to monetary history to a painting technique to why a particular interface creates cognitive load. The through-line is usually that I found something I did not understand well enough. I like teaching for the same reason: the test of whether I understand something is whether I can explain it without hiding behind the vocabulary of the field.

What draws me to products, startups, boxing, software and travel alike is that the gap between theory and reality shows up quickly in all of them. Reality is a useful editor. And I have a high tolerance for starting again, some of it voluntary and some of it not, which has left me much less attached to preserving an answer simply because it took a long time to reach.

The constraint

I work on reducing the distance between intent and execution. And on what software becomes when systems can sense what is happening, decide what needs to happen, act on it, and show the work back to the person without taking away their understanding or control.

Analytics was slow, procurement was bureaucratic, enterprise tools were hostile. Those are the same complaint in different vocabulary. Somebody knew what they wanted, and the system made them assemble it.

The positions on this site are where I have got to, not where I expect to stop.

Contact

If something here overlaps with a problem you are working on, write to me. I am particularly interested in hearing from people thinking seriously about autonomy, trust, and how software changes as systems begin to act rather than wait. And if something on this site is wrong, that is the most useful message you could send.

• swarochish.c@gmail.com
• linkedin.com/in/swarochishchekuri
• github.com/swarochish