Shriyashish Mishra

Product Builder | Product, Growth & AI
Learning in the hopes of understanding users better.

[Shriyashish Mishra @ShriyashishM]

4 AI agents. 3 humans. 1 repo. 💀
Production breaks. 🚨

@claudeai : “Not me.”
@cursor_ai : “Not me.”
Humans: “WHAT JUST HAPPENED?” 😭

So I built: Who Broke It? 🕵️
Now it’s less debugging, more true-crime podcast. 🔫
⭐️ https://t.co/Qd0vAxDa9H
#AIAgents #OpenSource #Github

[Shriyashish Mishra @ShriyashishM]

PM take after building with AI agents: the hard part isn't writing the spec anymore. It's writing down what "done" means 🤔

Agents are weirdly good at declaring victory. "All done!" they say, while the tests pass because the agent wrote the tests too 😅

My rule now: if I can't describe a check that proves it's finished, I don't have a requirement yet. Just a vibe.

How do you define "done" for your AI agents? 👇

[Shriyashish Mishra @ShriyashishM]

It's called Project Hulk and it's free to try: https://t.co/YbI13eJsds 💪

[Shriyashish Mishra @ShriyashishM]

I built an AI fitness coach that writes me a report every night.

One day I rested on purpose, like a responsible adult. It gave my workout a 0 😭

Turns out my score defaulted to zero on any day I logged as a deliberate rest day.

The AI wasn't harsh. My code had decided that resting means failing 💀

I fixed it, so a planned rest no longer counts as a skipped workout 🛌

Under the hood: the scores come from plain code over your logs and the AI explains them, so this was a logic bug, not a model having a mood 🧠

What's the most unfair score an app has ever given you? 👇

[Shriyashish Mishra @ShriyashishM]

@kanishkrazdan Respect for writing the postmortem. Thanks a lot for sharing. As a builder myself, I wanted to know what's the one thing you'd check in month one now to know it wouldn't get past $400?

[Shriyashish Mishra @ShriyashishM]

@rewind02 Claude working inside the running editor is the part I'd steal. What did you still have to check by hand? Since Claude can't feel whether the game actually feels good to play. 🎮

[Shriyashish Mishra @ShriyashishM]

@BIGMayrr Worktree per agent, interesting. I built something in this space (Who Broke It: https://t.co/Qd0vAxDHZf) and the hard part wasn't the worktrees, it was catching two agents changing the same interface before the merge. Does Firstmate check for that, or find out at merge time? 🤔

[Shriyashish Mishra @ShriyashishM]

It's called RegImpact AI. Poke the sample report yourself: https://t.co/LNXYXrDbNR 🔍

[Shriyashish Mishra @ShriyashishM]

I built an AI that checks fintech apps against RBI's lending rules. Then I gave it a loan app I made up and it roasted it. 20% compliant, "not ready for production" 💀

Its best burn: loan money goes through a pooled account first? RBI banned exactly that, and the AI quotes the clause to prove it 📜

My favourite part: when it can't verify a citation, it writes "Unverified" instead of bluffing. More self-aware than most of my first drafts 😅

Under the hood: RAG over the actual RBI text (not the model's memory), 5 follow-up questions that adapt to the product, citation verification, and the final score computed in plain code, not by the AI 🧠

[Shriyashish Mishra @ShriyashishM]

@stackway24 Two things right now. Project Hulk, an AI fitness coach that writes you a report every night from your food, workouts and sleep. And Who Broke It, a tool for when AI agents and humans break each other's code in one repo. What are you building?

[Shriyashish Mishra @ShriyashishM]

PM who got tired of just writing PRDs, so I started building the things too. 😄

I share what I’m building, breaking & learning at the intersection of AI × Product × Growth. No guru threads, just real experiments.

If you're into building, let's connect. 👋

[Shriyashish Mishra @ShriyashishM]

@alexvence_ @X Currently teaching an AI fitness coach to stop contradicting itself. Building in public at @ShriyashishM. What about you?

[Shriyashish Mishra @ShriyashishM]

I built an AI fitness coach that turns your food, workouts & sleep into a nightly report. 🧠💪

One day it told me my recovery score was 58.

A few hours later: 90.

I blamed the AI.
Then my prompt.
Then the model.
(That one was actually a rate limit. 😭)
Turns out… it was my own logic.
The AI was innocent. I’d like to formally apologize. 🤖🙏
What’s the dumbest thing you’ve blamed AI for?

#buildinpublic #AI #fitnesstech

[Shriyashish Mishra @ShriyashishM]

@stackway24 Hey, amazing. Followed you!

[Shriyashish Mishra @ShriyashishM]

@benz__002 Hey there! Let's connect. :)

[Shriyashish Mishra @ShriyashishM]

@Nishant_N_25 @X Hey there! Let's connect.

[Shriyashish Mishra @ShriyashishM]

@davindie_ All the best, David! Looking forward to seeing what you build. :) Let's connect!

[Shriyashish Mishra @ShriyashishM]

@sagarmoy22 @X Hey, let's connect!

[Shriyashish Mishra @ShriyashishM]

@CodeWithAmann @X Hey, let's connect!

[Shriyashish Mishra @ShriyashishM]

@rafal_darlak Interesting tactic. Let's connect!

[Shriyashish Mishra @ShriyashishM]

@turnless_HQ @X Hey, let's connect!