Our CTO @tbinetruy built 𝗣𝗮𝘀𝗲𝗼 𝗥𝗼𝗼𝗺𝘀: a collaborative coding agent 🚪
Share your live agent session as a link. A teammate or client joins the same conversation and answers the agent's questions, while you keep working.
🔓 Fully open source, end-to-end encrypted.
https://t.co/bJMs9lh7y9
📰 We sat down with the @SCMPNews to talk about why we invested in @Kimi_Moonshot.
Our reasoning is the one we keep coming back to: as intelligence becomes cheap, value shifts to whoever keeps the power to act.
Not whoever rents the most capable black box, but whoever can inspect, adapt, and run the model on infrastructure they control. 🔑
🔓 That is why open weight matters to us. We were using Kimi, and we tested self-hosting before we invested. For most real-world use cases, companies do not need the most capable model in the world.
They need one that is good enough, cost-effective to run, and genuinely theirs.⚡
Full article 👇
https://t.co/vpJE5o1ZKN
Un agent IA qui prévient avant que le problème arrive. 💡
Gros prélèvement URSSAF à venir, compte presque à sec. Sans qu'on demande rien, notre agent a fait le calcul et alerté @bilalelalamy : « il restera ~1 000 €, pensez à réapprovisionner. » ⚠️
🎥 L'extrait ci-dessous 👇
───
Our coding agents opened 1,834 merge requests in 12 weeks 😅
We're a team of 5 developers. That's far more than we could review by hand.
So every one of them went through Code Voucher, the review tool we build. Here are the numbers (Aug 13 – Nov 5, 2026):
→ 1,834 merge requests reviewed, across 18 repositories
→ ~36 merge requests a day over the last 4 weeks
→ 2.7 review rounds per merge request on average, because every "fixed it 🙂" from the agent gets checked again
→ 50 reviews in a single hour from one developer, at the peak
→ 197 merges stopped by a blocking finding
Each review looks at two things: code quality and security. It runs on the Codex subscription we already pay for, so there are no per-review fees.
This is our own data from our own team, so it's a first-party case study.
👉The full numbers and how we counted them are here: https://t.co/qHUtoigzOe
« Je ne sais pas écrire une ligne de code… et pourtant j'ouvre des merge requests. »
Pendant notre live, @bilalelalamy montre sa méthode : le bon skill, un seul morceau du problème, et challenger l'agent jusqu'à la merge request.
L'humain reste le sparring partner. ✅👇
How many versions of leads_FINAL.xlsx do you have? 📊
Rainmaker works differently.
Paste your website → it finds who actually buys what you sell, pulls the right people, and writes the emails. ✍️
👉 https://t.co/YCSkQE0t1X
Demain, votre seule interface sera un agent.
Pendant notre live, @bilalelalamy montre l'UI générative : plus de menus, on parle à un agent et l'écran se construit pour répondre. Gmail, Qonto, le Drive... on ne les ouvre plus.
Quel outil as-tu déjà arrêté d'ouvrir ? 👇
Reading the code isn't the same as using the app. 👀
A one-line change can pass every review and every test, and still break checkout.
A discount code takes a €4 order to −€1.00. The diff looks clean. The bug only shows when you click through.
Code Voucher now catches that. 🧾
2:14am. Still fixing Meta and Google ads that aren't even live.
That's how most founders run their first ads.
AdsGroom does it for you: brief, creatives, landing pages, Meta + Google campaigns. Paused until you approve.
👉 Try it out for free: https://t.co/vgJTeo3Mp3
one failed request would have blanked our whole Settings page. not one panel. the whole page.
Code Voucher caught it before the merge. the fix: keep the page rendering when one section can't load.
real bug, caught before the merge: a change would have sent our private invite links to Google Analytics.
none of our tests checked what ends up in analytics. the review did.
another real one from our own code: a change for our daily Slack report would have written our secret Slack key into the logs.
tests passed. Code Voucher blocked the merge.
secrets in logs are exactly what a tired reviewer scrolls past.
I expected our AI reviewer to mostly flag security issues.
the top category in our own code was correctness: 517 findings. then state & lifecycle and docs drifting from the code, 196 each.
the boring bugs are the ones that ship.
at our peak, one developer triggered 50 reviews in a single hour.
that's not sloppy work. that's what building with coding agents looks like: push, review, fix, push again.
no human reviewer keeps that pace. Code Voucher is built for it.
a bug Code Voucher caught in our own code: a billing change would have let trial teams raise their seats to 1,000 without paying.
tests passed. build passed. the review blocked the merge.
we review our own code with it every day, and it still catches us.
for 12 weeks, our 5 developers ran their merge requests through Code Voucher, our own AI code reviewer.
1,834 merge requests across 18 repos. 5,000 review runs. 197 merges stopped by a blocking finding.
we wrote it all up, with the numbers:
https://t.co/Vy0fGJWtnf