Neil Kakkar

I love finding problems. Insatiably curious. I tweet about what 🤯 and ask lots of questions

Eng @ Tano Ex: @PostHog @Bloomberg

Writing:

[Neil Kakkar @neilkakkar]

I love it when things click!

A thread exploring things that surprised me.

1. https://t.co/JO3GoeE6Z0

[Neil Kakkar @neilkakkar]

Something that took me hilariously long to understand:

Earning millions in revenue doesn't mean you're making money.

Revenue gives no indication to profitability. People assume that would be the case, since well, you'd notice something wrong once you get to a few million.

[Neil Kakkar @neilkakkar]

@VictorTaelin Really enjoy your posts on this! I've found that you don't need to go as far as pointing out the root cause, asking a question like "why does mem flow only in one direction" is enough to put it on thr right track in most cases. Do you think this is getting rarer tho?

[Neil Kakkar @neilkakkar]

Tano is hiring :)

[Sagar Shah @sa64r]

Looking for an engineer to join Tano in London.

AI-native influencer marketing. Small team, agents in production, brands paying for it.

I'm the CTO. DM me and I'll tell you exactly what we're building and what you'd own. No recruiter, no form, I read every message.

Link in bio

[Neil Kakkar @neilkakkar]

An interesting, unexpected effect of working with AI: Not only did it make me faster, but it made me braver.

https://t.co/5BM0BBNP7K

[Neil Kakkar @neilkakkar]

3/ Wrote about it more in depth here: https://t.co/lYtPmqThLo

[Neil Kakkar @neilkakkar]

1/ The highest-leverage work I've done hasn't been writing features. It's been building the plumbing that makes AI agents effective.

The kind that compounds as models get better, not the kind that becomes obsolete.

[Neil Kakkar @neilkakkar]

2/ I'm in a tight loop: kick off task → agent writes code → check preview → read diff → feedback or merge → next task.

The feedback loop is so tight there's no gap for my attention to leak out.

And as models get better, this harness lets me move faster.

[Neil Kakkar @neilkakkar]

4/ The engineering role is shifting. Less writer, more gardener.

You're not writing the code anymore. But you're the one making sure it stays coherent.

Wrote more about this here: https://t.co/bfBYy2eint

[Neil Kakkar @neilkakkar]

1/ New concept I've been thinking about: agentic debt.

It's like technical debt, but self-reinforcing. An agent writes code. Works. Ships. Next agent inherits the codebase. Gets confused by the first agent's choices. Makes worse choices. Repeat.

[Neil Kakkar @neilkakkar]

3/ The fix isn't bigger context windows. Three duplicated patterns don't become clearer with more context, instead they become three times as confusing.

The fix is making the code simple enough for humans to model. Human-understandable turns out to be agent-actionable.

[Neil Kakkar @neilkakkar]

@_bhuvanesh_ thanks for the kind words :D cheers!

[Neil Kakkar @neilkakkar]

1/2 Haven't tweeted in years. Burned out, had a panic attack, quit my job, rebuilt.

The biggest thing I learned: every lasting behaviour change I've made has been environmental, not motivational. Willpower-based resolutions: 0% success rate for me.

[Neil Kakkar @neilkakkar]

@rexan_wong @vercel Sweet, this is great! Get @posthog analytics setup, should be pretty easy and you're on the free tier so you'll get much more accurate analytics for free :)

(i used to work there, but not anymore)

[Neil Kakkar @neilkakkar]

Trying to debug a CI workflow, exhibit #24357

[Neil Kakkar @neilkakkar]

@TomFromRealLife @raquel_m_smith @posthog You don't want to random assign inside bootstrap. Instead, use your middleware to add this info on page render.

The reason it's not slow is you'd use local evaluation in your middleware - which will compute the flag without going to our servers, and return within 15ms.

[Neil Kakkar @neilkakkar]

@deliberatecoder Congratulations Michael! It's been amazing to follow your journey the past few years :')

good luck with the baby and whatever comes next 🙌

[Neil Kakkar @neilkakkar]

@veryayskiy @posthog thanks, responded there, will continue convo there :)

[Neil Kakkar @neilkakkar]

@jameschambers @posthog Hey @jameschambers just took a look for you - the problem here is you have very few absolute conversions, which is why (1) results can easily flip quickly and (2) probability is not very high.

You can input your numbers here to see a similar probability: https://t.co/LPjuzWGxNp