The password toggle worked with a mouse.
Our AI QA agent flagged it anyway.
It was an image, not a button. Keyboard users couldn’t reach it.
I built @testwithrecus to find broken software.
Now it’s making me question what “working” actually means.
here's a workflow you should probably save:
>use https://t.co/TBhiODNQbN to build your website
>use https://t.co/82gLdSjLPP to find bugs you didn't know existed
>take those bugs back to v0 and get them fixed
>run https://t.co/82gLdSjLPP again to catch what the fixes broke
>repeat until your website stops embarrassing you
>AI builds the product
>AI breaks the product
>AI fixes the product
welcome to software development in 2026 😭
a customer used @testwithrecus , fixed the bugs, came back to test again, then started finding bugs in the AI that finds bugs.
"your recordings look broken"
"where's the export button?"
"btw your app is down"
BRO !
i accidentally hired a QA engineer, PM, and CTO for $0.
customer feedback is the actual feature.
Every app I built, I kept running into the same problem: testing everything manually, over and over again.
So I stopped writing tests and built Recus.
An AI QA agent that explores your app, understands how it works, and finds bugs on its own.
Here’s what it does in 79 seconds 🎬
100+ users. 200+ runs.
holy fucking shit. 😭
today, @testwithrecus crossed 100 users and 200 test runs.
and weirdly, the numbers aren't even the most exciting part.
it's watching complete strangers show up, test their applications, discover bugs, go back and fix them, and then come back to test again.
like, wait a fucking minute. people are actually using something i built to make their own products better???
that's literally what i wanted @testwithrecus to become.
and no, it's not all sunshine and rainbows. sessions are failing. the agent still does stupid shit sometimes. half the things i want to build aren't even built yet. the product is nowhere near where i want it to be.
i wouldn't say we've cracked product-market fit yet, but we're seeing something that feels a lot more real than just another signup on a dashboard.
people are coming back because they actually need this thing.
i'm building a bunch of things in parallel, probably more than any sane person should, but there's something about @testwithrecus that just hits differently.
it's slowly going from "what if i build this?" to "holy shit, people actually want this."
still early. still broken. still figuring shit out.
but 100 people trusted a product that barely existed a few months ago, and that's fucking insane to me.
anyway, enough emotional founder bullshit.
back to shipping. we are not fucking stopping.
test with @testwithrecus: https://t.co/nGWeHVNqeL
we're just getting started, baby. 🚀
“bro, @testwithrecus is just a playwright click bot with ai slapped on top.”
yeah? and your entire tech stack is a wrapper around a wrapper around another fucking wrapper.
@viktor_com is “just an ai slack bot” until you realize execution, architecture, distribution, and actually solving a problem are what make a product works
the funniest part? the same people calling everything a wrapper are the first ones asking for 20 more features.
nobody gives a shit what you wrapped.
the person who actually builds it, ships it, and makes people pay for it wins.
that's the whole fucking game.
i think the best engineers have a serious fucking problem.
they can't just fix a bug and move on.
i was testing a random website and found a bug that only happens under one ridiculously specific condition.
normal engineer: fix it.
my brain: what if we could eliminate this entire category of bugs across every website we'll ever test?
suddenly a 20-minute fix turns into a 3-day architectural rabbit hole.
but that's the part of engineering i'm genuinely obsessed with.
solving a problem you've encountered once in a way that handles thousands of situations you haven't even imagined yet.
that's when you stop writing fixes and start building systems. ,ps make big w @testwithrecus
small wins for @testwithrecus .
a couple of people have started signing up through chatgpt recommendations. like, people are literally asking an ai what to use for testing, and somehow it's recommending the ai qa agent we're building.
what the heck 😭
are they our ideal customers? maybe not. do we care about these users? absolutely. every single person trying our product matters, regardless of where they come from.
meanwhile, we've been shipping some pretty massive changes.
right now, @testwithrecus is page-structure intelligent. it can see a page, understand its elements, interact with them, and find bugs.
but here's the problem: understanding a page isn't the same as understanding a product.
a human doesn't forget what they're doing every time they navigate to another page.
so we're building context bucketing. page-level understanding, shared application context, memory of previous interactions, and an evolving understanding of how the entire product actually works.
the goal isn't to build an ai that clicks buttons faster than playwright.
it's to build an ai that understands WHY it's clicking them.
i don't want @testwithrecus to behave like a test script.
i want it to behave like a human who actually gives a shit about breaking your product.
we're nowhere near done, but holy shit, this is getting interesting.
holy fucking shit. 😭
today i opened microsoft clarity for @testwithrecus and saw someone spent 37 MINUTES using our product.
37 fucking minutes.
not scrolling. not browsing. actually testing their app with something WE BUILT.
we're getting users from chatgpt, qa engineers, lovable builders, and people we've never spoken to.
our reddit posts crossed 100k views. people are telling us what they'd actually pay for. and we're watching strangers use recus without us begging them to.
do we have PMF yet? probably too early to say.
but holy shit, the signals are getting real.
@MomenticAI ' mo showed there's a market for ai qa. we're betting we can make autonomous testing radically cheaper and more accessible.
we're building. we're shipping. we're talking to users.
and this time we're not fucking stopping.
@testwithrecus is just getting started.
pricing an ai qa agent is somehow harder than building the fucking thing 💀
we're building @testwithrecus and can't decide:
subscription ($49–$99/mo) or pay-as-you-go credits?
if you're a qa engineer or dev, what would you actually pay for?
genuinely curious.
reddit is genuinely the most humbling place to build in public 😭
you spend days building an ai tool, thinking you've cooked, and then open reddit to find 17 people who've already built something better, open sourced it, and are calling your entire category "just another wrapper" 💀
but honestly, i love it.
the amount of insane stuff people are building is unreal. found so many things we could potentially use to scale @testwithrecus too.
one thing i've realized: if you can build something that actually makes redditors happy, you've probably built something people will pay for.
because impressing reddit might genuinely be harder than building the product itself lmao.
we're actually going insane building @testwithrecus AI QA agent.
imagine telling an AI agent "test the signup flow" and it just fucking figures everything out.
needs an email? creates one. needs an OTP? opens the inbox, grabs it, verifies it. needs to finish signup? just does it.
no 47-step instructions. no babysitting.
and this is just ONE skill.
we're adding 10–20 more.
the crazy part? building these agents is a fucking nightmare because you have to test the thing that's supposed to test things 😭
but genuinely, this is the first time in a while i've felt like we're building something people haven't seen before.
we're just getting started.