Gatien retweeted
The thing that separates the top 2% of founders from everyone else isn't visible from the outside
It's the conversation they have with themselves every single week
- Am I working on the most important thing?
- Am I talking to enough users?
- Am I being honest about what the data says?"
Three questions. Weekly. Without flinching.
That practice, sustained for 12 months builds a company no hack can replicate
Goat team
We've increased CodeRabbit's engineering output roughly 20x in six months with the same sized engineering team. Marketing, sales, ops. All need to move at the same speed.
We're turning how we run the company into resources agents can read and work with:
• Agents launch and update campaigns, with changes reviewed before they go live.
• Codex builds slide decks and customer documents in our content studio.
• CodeRabbit Slack bot investigates incidents and support tickets, then proposes fixes as PRs.
We're moving more of the company's knowledge and workflows out of documents and into code. Plans, processes, and content become things agents can act on, with changes we can review and track.
Faster code isn't enough. The whole business has to keep up. I spoke with @OpenAI about the workflows we rebuilt and the bottlenecks we had to fix.
Full story below.
24
Gatien retweeted
And... we are back! 6K PRs merged this week! After struggling with costs ($13M/yr!!) and reliability issues, we built our own CI platform within a week. And it's so good we are finding it hard to resist launching it as an external product.
Some mind-blowing stats:
- PR run: about 4–5 min, down from 8–12 on vendor CI
- In the last 7 days : about 36,000 PR and merge-queue runs, and 1.7M+ jobs
- 300k+ jobs on a busy weekday, and up to about 30k in a single hour
- Each run fans out into about 40 jobs on average, spread across 18 build clusters
Small, agent-pilled teams of less than 50 engineers can now ship at the level of sophistication that large tech companies were doing a decade ago!
This week we didn’t see an uptick in shipping velocity. We stalled due to CI capacity issues with our provider, and to make things worse, even our cloud providers have run out of capacity to host our runners. VCs are getting excited about funding faster/cheaper runners while the reality is that constantly optimizing and rewiring CI/CD has become the name of the game. Once you solve the agentic validation and code-review bottlenecks, you still have plenty of engineering and janitorial work to keep the factory humming.
Pi Goat
I burned a shitload of compute so you don't have to. Releasing the Harness Leaderboard.
Which harness should you run your open model in? I ran 28,849 trials to find out.
The leaderboard compares nine coding-agent harnesses on open models. In each comparison only the harness changes: the model, provider and reasoning effort stay fixed, and every task gets three attempts.
The trials ran GLM-5.3-Flash, DeepSeek-V4.1-Flash and Flint on Terminal-Bench 2.1 and DeepSWE, for $4,746 of inference.
Averaged over the three models, mini-swe-agent (@KLieret), a bash-only agent, solves 51.4% of tasks and @openclaw 27.3%. Four others are statistically tied with mini-swe-agent.
One of them is Pi (@pidotdev), at 49.4% for $0.29 per solved task, about 45% less than mini-swe-agent's $0.53.
The best harness depends on the model: mini-swe-agent on GLM-5.3-Flash, Hermes Agent (@NousResearch) on DeepSeek-V4.1-Flash and Kimi Code (@Kimi_Moonshot) on Flint.
@opencode costs $3.35 per solved task, 11.7 times as much as Pi.
See Harness Hub's Harness Lens for what the nine harnesses do differently. It documents the prompts, tools and compaction of 34 coding-agent harnesses and links every claim to its source.
Leaderboard: huggingface.co/spaces/joelni…
Harness Lens: huggingface.co/spaces/harnes…
33
Gatien retweeted
3 weeks of runway = pure Default Dead territory.
YC’s advice is clear: aim for 18–24 months after a raise so you’re never negotiating (or answering “good”) from this position.
Once you’re under 2–3 months, the options get ugly fast.
Gatien retweeted
Dopamine does not spike highest for guaranteed rewards. It spikes highest at reward prediction error. If you want to have fun in life, you should stop avoiding risk and uncertainty.
We’re renaming our Max plans:
Max 5x → Max
Max 20x → Ultra
Max is still $100/month, and Ultra is still $200. If you’re a subscriber, there’s nothing you need to do.
We’re also removing the 5-hour limit for verified Ultra accounts, starting today with most existing subscribers. Newer accounts keep a temporary limit until they’re verified, which helps us prevent fraud and abuse. You'll get an email when it's removed.
Weekly limits remain unchanged. View our plans here: claude.com/pricing
I think the same is true of creative fields, like I see artists really worrying about AI
but I wish so badly that I could sketch animation ideas and character concepts for my agent to flesh out instead of trying to find examples
I don't have the language!
Gatien retweeted
the skillset I spent the past 13 years of my life mastering is essentially obsolete.
once I came to terms with that, I was able to see a new world of opportunities for what they were. writing code was always just a means to an end anyway.
also, incredible writing by @felixrieseberg, I aspire for this level of craft.
The coding interview is dead, at least for my team at Anthropic. I've seen enough to know that it no longer provides signal. I removed it from loops this month.
I have a spicy, optimistic take - I think there's an opportunity for junior engineers: felixrieseberg.com/hunger-be…
Gatien retweeted
there are only 2 durable moats left that no AI model will ever replace:
1) obsession, hunger and execution. the ability to iterate and build new moats faster than the old ones get copied.
2) distribution and attention, earned through reputation, trust and even regulation. the thing that makes people choose you and stay with you over the copycats.
2) only matters if you have 1)
Gatien retweeted
FP8 unlock for 30 series and 40 series hahaha can't make this up! LOL
github.com/Zagitta/gpu-unloc…
Gatien retweeted
The 5 words that kill YC applications:
- "Proprietary algorithm" (what does it do?)
- "First-mover advantage" (meaningless)
- "Viral growth" (how?)
- "No competition" (impossible)
- "Huge market" (show your own math)
Replace each with something specific. Your application will be unrecognizable in the best way.


































