Udayan Walvekar

Unpopular opinion I agree with- learning should not be fun. This blurb from @karpathy summarises it very well.

@Ujjwal__xyz Thanks!

@udayan_w Good article

@saransh92garg Congratulations Saransh 💙

Happy to share this milestone in the company building journey!

@saxenasaheb 😎

@udayan_w S tier caption game man!

@natashamalpani hahaha! thanks 💙

@udayan_w love the title

Worth a read - most knowledge workers will evolve into agent managers, albeit with a function-specific nuance.

such a clarity read

Article number #2 on twitter.

Thank you @EconomicTimes @kpswathi for featuring and giving spotlight to the real builders shipping 🚀 economictimes.indiatimes.com/tech/artificia…

What happens when 150 OpenClaw builders are in one room? The room was insane. Openclaw on custom hardware. Voice agents. Agent to agent payments. AI infrastructure. SaaS. CRM. Security. Health care. Finance. Builders shipping across 20+ industries. The room was packed. People didn't want to leave. The builder energy was infectious. @harshilmathur kicked off the evening by demoing the openclaw agent he uses daily. It was clear. The best people in the ecosystem are building. One of the largest OpenClaw Showcase in India. Couldn't have done it without 👇🏻 The @GrowthX_Club members who made the energy in that room what it was. The @OpenAI team for supporting and backing the builder community in India. @peakxvpartners and @Razorpay for always providing the right platform for disruptors and builders.

@waynesutton @ah20im +100. Need Convex and Codex integration. @gabrielchua @harshitm29

@ah20im convex.dev integration

What would you like to see in Codex?

there's a reason some people's agents keep getting better and others stay stuck. it's not the model. it's not the prompt. it's something simpler. everyone optimising skills and memory for their agents right now is creating a feedback loop. most don't call it that. they call it "fixing the prompt" or "tweaking the skill." but the structure is the same every time: run the agent, look at the output, find the gap, write a correction, run again. foundation model companies are doing the same thing. optimising feedback loops in post-training and fine-tuning. automatic feedback loops produce better systems. always have. always will. i started seeing this when i was building agents last year. the patterns felt familiar. not from engineering. from building teams. any founder who's good at building high feedback loop teams (clear signal, fast correction, right incentives) will train their agents the same way. the mental model transfers directly. the other thing that stuck with me was a professor @karpathy video where he walked through how GPT was built. that broke something open for me. the principles of how you improve a model through structured feedback became so foundational that i still apply them at the application layer. skills, memory, evals. make the feedback loop better. make the feedback rewards better. and you will see magic.

@utkarshsphotos @dastarkhwanclub Thanks for making this happen 😁

@udayan_w @dastarkhwanclub It was really nice to meet you yesterday!

ending saturday with THE best food community in bengaluru >> @dastarkhwanclub also, congratulations on the new place & name? the navu project is now navu bistro.

no hype but design, product & engineer is truly one role now.

@p_naix @peakbengaluru no, very raw conversation around ai, defensibility, moats, and so much more.

@udayan_w @peakbengaluru Was it the PMF talk?

Builders, Demos and VCs in Koramangala at the Sanctuary. @peakbengaluru maxxx

@saaketsharmax @peakbengaluru had so much fun.

@udayan_w @peakbengaluru It was lovely hosting you! 🙂‍↕️

yea, cause i am an opus snob.

AI Festival 2026