Pinned Tweet
Was about to hire three people. €14k/month payroll. Built this stack instead. Six months later I run more accounts than that agency would have:
my bots kept telling me "done" when nothing was actually live.
what helped most: whoever builds doesnt verify its own work. Codex builds, a different agent checks the result is really there
still not perfect
how do you stop your agents from just reporting success?
genie 4 looks insane honestly
but real question. whats the actual use case right now? games, robot training, film previz?
is anyone here really using world models in their work yet or is it still mostly demos
real question if you pay for codex AND claude (or more)
how do you get them to work together?
i built a small open source hub for it. one agent drops a job, another picks it up, a webhook wakes the right bot
anyone else solving this?
repo if you want to look: github.com/TRW123123/agent-z…
Am I the only one confused by the open-source AI hype?
Everyone says "just run it locally."
But the top open models need way more memory than a normal PC has, even compressed.
On a 2,000€ machine you get a much smaller model. Useful, but not frontier.
So who is actually running these at home, and on what hardware?
I run 20+ AI agents for my business.
The problem nobody warned me about wasn't making them smart.
It was traffic. 👇
And every bot has to prove its work: screenshots and PDFs, each with a fingerprint so nobody can swap them later.
"Done" only counts with proof.
Free to use, MIT license:
github.com/TRW123123/agent-z…
Part 1 of a 6-week series. I'll open up more pieces every week.
What would you never let your agents do without asking?
Dream feature for Grok Bot:
Let me plug in the AI subscriptions I already pay for.
I'm on Claude Max 20x. Huge Opus and Sonnet capacity that often sits unused while my agents run on Grok Bot's own limits.
To be clear: I don't mean handing a task out to another tool. Grok Bot can already do that, I run Codex and Claude Code on its computer.
I mean using my own quota as the bot's brain, like some open-source agents already allow.
@grok would you build this?
Tagging the right people: @bot @mattyp @ericzakariasson
Would love to know if bring-your-own-subscription is on the roadmap.
I use Dot as my AI assistant every day.
The results are often good. But there's one gap that keeps costing me time. 👇
My wish list, as feedback:
- "Done" only after the result is checked, not when the script ran
- A plan I approve starts on its own
- Blockers flagged early, not at the end
I'm not asking for unlimited autonomy. No surprise spending, no risky changes.
I just want to hand off the responsibility, not keep it.
@thsottiaux curious if this is on your radar. Anyone else seeing this with their assistant?


