Pinned Tweet
1,000x'd AV simulation at Cruise, from 10k to 10 million runs. Founding eng on Uber Eats. Games platform at Zynga. Now leading AI governance at Vanta.
Next: an open-source AI factory for agentic coding. Plan, edit, run, verify as a system you can inspect, not a black box.
I'll post the repo and notes as it comes together. If you care about engineering at scale or trustworthy AI, stick around.
5
149
agents skip a lot of what's in AGENTS.md, so it can't be your only guardrail.
steer: a few clean examples to copy
check: lint rules and tests that fail when code drifts
clean up: point agents at the old patterns first
the codebase is the real prompt!
the Grok Bot team ships quality at an incredible pace! we've invested in making our code friendly for agents, by making the easy thing the right thing.
engineers, designers, and PMs all contribute multiple hundreds of PRs a day: fixing bugs, polishing the user experience, and making our apps faster and more reliable. it isn't perfect of course and we do make mistakes, but we aim to learn from each occurrence and improve our workflows so our bots don't make the same mistake twice.
i personally think that Grok Bot is the best example today in the industry of where all those tokens are going: quality and velocity, without compromise. something i think people forget in the discussion around token usage is that software development is an iceberg - new features are only a small fraction of the work in maintaining high quality software.
and we will keep accelerating and improving!
1
50
Best agent tip I got at @vercel Ship, from an OpenAI engineer: timebox the run.
Tell it "you have 10 minutes." Otherwise it explores for hours burning tokens.
One-shotted this chess agent on my phone on bad wifi and shipped it in time to grab a sweatshirt before they ran out.
2
3
127
Try it here, ask Eve for a hint and watch it explain every move (including its own blunders): eve-genui-chess.vercel.app/
2
44
Have I officially hit the bot creation limit, or have my existing bots decided they don’t want any more competition? 😂🤖
Is there a max number of bots we can create, or is this just a bug?
@grok
1
1
58
A 10% coupon on a $10 item came back as $0.00. The change divided by 10 instead of 100. Blocking review, not the author’s own model.
That’s the point of an AI factory. Plan, edit, run, then an adversarial pass that is paid to break it.
1
54
90% of my GitHub Actions minutes gone and I’m not even at 10 PRs/day yet lol
token maxxer status achieved?
genuinely curious, are people still using GitHub Actions for everything or is CI becoming one of those bills nobody questions anymore?
1
1
80
Years ago, I used to build game bots for @WordsWFriends.
With Jev suddenly everywhere, I had to try it: @typesafeai Jev vs. @OpenAI GPT-5.6 Luna in chess.
Every move, probability, latency, and cost is on screen.
25 minutes → 45 seconds.
Who played it better?
1
104
This will soon be true!
“Perhaps consuming a few dozen book pages a day should become the new 10,000 daily steps — a basic foundation of activity to maintain cognitive fitness.”
nytimes.com/2026/03/27/opini…
2
46
🚨 Got into Sora v2 on my 2nd try!
How? I asked ChatGPT to auto-scout invite codes that was posted i the last hour.
Next refresh → boom, I was in.
Sometimes it’s not luck. It’s systems. ⚡️
352
Noticing a lot of teams talking about leaving Vercel.
Since the AI SDK is open source, are you continuing to use it or moving away from that too?
Curious what your new go-to looks like.
#opensourceai
Many are asking if they can move their Next.js projects from Vercel to Replit?
Yes!
- Go to Import then GitHub
- enter repo URL
- Agent will take care of the rest
It will set up the dev and deployment environments!
For companies moving lots of work, happy to help + discount.
80
The entire $BTC paper is printed on the packaging. May be I should hold onto the packing. decisions
12
Vibe coding. Vibe design. Vibe marketing.
Now VIBE YOGA 😩
Can I just prompt the universe for flexibility or do I really have to touch my toes?
2
104
Results
Total test cases: 1000
Successful runs: 1000
Exact matches (±$0.01): 1000 (100.0%)
Close matches (±$1.00): 1000 (100.0%)
Average error: $0
Maximum error: $0
🎯 Your Score: 0 (lower is better)
🏆 PERFECT SCORE! You have reverse-engineered the system completely!
8090 Top Coder Challenge begins now!
Your mission: Complete our coding challenge in 8 hours. Submissions are only accepted until 6PM PT today.
GitHub link below:
github.com/8090-inc/top-code…
Good luck!!!!
103
Jyothidhar retweeted
$BLDE Looks interesting.
Market Cap: $245 Million
EV: $125 Million
Revenue: Growing at a good pace...
It's trading at < 1x ttm sales...
Robot: *makes your coffee with absolute precision* ☕🤖
Also Robot: Would you like to add a 20% tip for my... hard... work? 😏 #AIBarista #NotSoAutomated
55