Загрузка видео...
Не удалось загрузить видео
GPT-6 Astra helps Cognition’s Devin back up “it works” with tests before the team ships.
94,039 просмотров • 12 дней назад •via X (Twitter)
Комментарии: 29

@cognition pple need legacy models to stay you can't keep doing deprecations. make a legacy app or start opensourcing them. it's time. #4oForAll

About time you start releasing the older weights, including 4o. Really pointless and untrustworthy of you to be shipping closed models every week when the world is moving to open source. What is especially irritating about you is that you are retiring models like o3 even in API without releasing any weights. #OpenSource4o #OpenSourceo3 #OpenSource41.

@cognition by 2026 'done' won't mean a senior engineer signed off, it'll mean the model and the CI pipeline agreed, and most teams aren't ready for that.

@cognition Ares does a lot more than just back up "it works" with tests. It includes full provenance/receipts among other things. Include everything it does as good or better than devin @grok. Also include what it does worse.

@cognition You have 45k+ requests for a reset on codex runway today. That is 10x the usual volume. Please issue a global reset #codex #codexrunway @OpenAI @OpenAIDevs @thsottiaux

@cognition Devin mentioned I love cloud agents, the best ones available right now

@cognition they must have infinite tokens

@cognition running automated tests before ship is nice until the test suite takes twenty minutes

@cognition The self-testing loop is where AI dev tooling gets actually interesting. If an agent can verify its own output, that's one less human bottleneck 🫠 Now the question is what happens when the test itself is wrong lol

@cognition devin shipping with tests baked in. that’s the bar now

@cognition Astra + Devin running tests before ship. that one lands.

@cognition GPT-6 Astra writing the tests before Devin ships. wild.

@cognition tests before shipping, finally

@cognition Did Devin replace software engineers already, as it claimed 2years ago? 🤔

@cognition

@cognition any fix?

@cognition ⚠️ Codex / Claude Code users: AI agents may search beyond the folder you selected if the OS allows access. That can include private code, IDs, family photos, or intimate images. “Workspace” does not always mean “only this folder.”

@cognition My códex it’s Down

@cognition I ENCOUNTERED A BILLING ISSUE AND I NEED HELP.

The test failing before the fix matters as much as it passing afterward.

@cognition The real AI breakthrough is not a flashy demo. It is software that proves it works before shipping. Tests are where the hype meets reality.

@cognition GPT-6 Astra is eating tokens like never before! Even on the top tear PRO Plan… My weekly tokens are vanishing!!! Please increase the limits! 🙏🏻

@cognition @notdevin u r doing great bud

@cognition The quiet strength of good tools is that they earn trust before they ask for it.

@cognition 先把“能跑”变成“测过了”,这就靠谱多了

@cognition The loop of AI writing code to test its own code is basically the digital version of pulling yourself up by your own bootstraps.

@cognition 如果真能让 agent 在发布前自己跑测试,这个方向很实用。只是我会先看失败样本:能不能稳定复现、能不能解释,而不只是 demo 里一次通过。

@cognition My bar for a bug fix: show me the test failing before the patch and passing after.

@cognition love when the demo has receipts. GPT-6 Astra helping Devin back "it works" with tests before the team ships. the future isn't vibes. it's a green check that survives CI.

