Neeraj H

Building with AI, Documenting the mistakes so you don't repeat them.
Rule 1: Build with AI.
Rule 2: Post without it.

[Neeraj H @sign_left]

What's the smallest, dumbest thing that made you completely lose trust in an AI agent mid task?

It doesn't even have to be a big dramatic failures, more the small moments where something clicked and you went from trusting the output by default to double checking everything. For me it was watching an agent confidently rename a function across twelve files, then leave the original function untouched in a thirteenth file it apparently didn't search, with zero indication anything had been missed. It wasn't even a hard case, the file just wasn't in the directory it happened to grep first.

What was your moment? And did it actually change your workflow afterward, or did the trust creep back in after a week like it always seems to for me?

[Neeraj H @sign_left]

@anupamrjp Github.

[Neeraj H @sign_left]

@itz_komal01 migrations

[Neeraj H @sign_left]

@itz_komal01 Hi there komal. Let's connect.

[Shravani @Shra_va_ni]

They really ran out of features😭

[Neeraj H @sign_left]

@Shra_va_ni "microsoft recommended" sounds less like a feature and more like a warning label.

[Neeraj H @sign_left]

@crescentforeal the bigger divide isn't blue collar vs white collar. it's routine vs judgment.
the messy edge cases still belong to humans.
That being said, a handful of people will survive in the services industry.
So, in essence, yes! we're cooked!

[Neeraj H @sign_left]

@crescentforeal having the best models isn't enough anymore. people stick with the AI that's already inside their workflow. switching costs are becoming UX, not intelligence.

[Neeraj H @sign_left]

@jasonleowsg i'm starting to filter ideas with one question: "would someone still pay if they could build the first version themselves today?"

surprisingly few survive that test.

[Neeraj H @sign_left]

@thealtryst @X don't automate a workflow until you've done it manually enough times to find the boring parts. otherwise you just scale your misunderstanding.

[Neeraj H @sign_left]

@alexmacgregor__ the interesting curve isn't toy to normal. it's demo to dependable. lots of things look magical for 10 minutes. far fewer survive six months of real users.

[Neeraj H @sign_left]

@TomHacohen this is why i'm bearish on "clone in a weekend" businesses. generating features is getting cheaper every month. earning enough user context to know which features matter isn't.

[Neeraj H @sign_left]

@buccocapital i think we'll see fewer teams building CRUD apps, not fewer teams building software. the custom work just moves closer to the part of the business that actually creates leverage.

[Neeraj H @sign_left]

@PrimalNick vibe coding isn't the danger. cheap building is. once creation costs almost nothing, saying "no" becomes the highest leverage engineering skill.

[Neeraj H @sign_left]

@jasonleowsg i've started sorting ideas by "how much human judgment survives after generation." if the answer is almost none, AI eats it. if the messy part repeats every week, there's usually still a business there.

[Neeraj H @sign_left]

@TWlTTERHAVER I exactly know what you just said!

[Neeraj H @sign_left]

the friction isn't creating anymore. it's reviewing. i can generate work while walking. i still need a tight loop for catching the one assumption that quietly sends everything sideways. Maybe I feel like this because I am a person who believes vetting the AI output results in much better results

[Neeraj H @sign_left]

@iamKierraD "narrative architect" feels right. the part i keep watching is feedback loops. people optimizing for first draft quality are getting passed by people optimizing the edit loop. that's where most of the gap is now.

[Neeraj H @sign_left]

@phl43 i'd go one step further. hallucinations are noisy. hidden assumptions are quiet. noisy failures get fixed. quiet ones make it into production because everything looks plausible until much later.

[Neeraj H @sign_left]

@businessbarista the interesting shift for me isn't AI vs human. it's who can build the fastest revision loop. i spend way more time fixing confident mistakes than generating words. does that ratio ever hit zero?

[Neeraj H @sign_left]

@businessbarista end result matters. but i think process still leaks into the output. the best AI writing i've seen came from people with a sharp editing instinct, not the longest prompt. models amplify taste more than they replace it.