Uploaded November 2025 | Updated September 2026, 1 hour ago
Moonshot AI just dropped their new OK Computer agent mode inside Kimi doesn’t just generate text, it actually acts. And I got early access to try it on a real problem: fixing our messy B2B slide deck at Towards AI.
I fed it the raw files. It stored everything, planned the work, wrote its own to-dos, iterated through versions, and rebuilt the entire deck to look clean and professional. Then I asked it to optimize the deck for conversions, with human approval steps in the loop so I could guide the direction.
And here’s the crazy part: it gave me a broken file. I said “I can’t open it,” and without any extra instructions it fixed itself and regenerated a working HTML version. That’s the new wave of agentic systems: understand, reason, correct, and execute with just enough human feedback to stay aligned. Honestly… equal parts impressive and scary.
If you want me to test OK Computer on a tougher workflow or something from your industry, tell me and I’ll push it further.
I’m Louis-François, PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀
Comment for the link to try it!
kimi.com/?utm_campaign=TR_zwkjw3Tw&utm_content=&utm_medium=tiktok&utm_source=CH_O3fWqjAg&utm_term=
#moonshotai #kimi #agenticai #short
Moonshot AI just dropped their new OK Computer agent mode inside Kimi doesn’t just generate text, it actually acts. And I got early access to try it on a real problem: fixing our messy B2B slide deck at Towards AI.
I fed it the raw files. It stored everything, planned the work, wrote its own to-dos, iterated through versions, and rebuilt the entire deck to look clean and professional. Then I asked it to optimize the deck for conversions, with human approval steps in the loop so I could guide the direction.
And here’s the crazy part: it gave me a broken file. I said “I can’t open it,” and without any extra instructions it fixed itself and regenerated a working HTML version. That’s the new wave of agentic systems: understand, reason, correct, and execute with just enough human feedback to stay aligned. Honestly… equal parts impressive and scary.
If you want me to test OK Computer on a tougher workflow or something from your industry, tell me and I’ll push it further.
I’m Louis-François, PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀
Comment for the link to try it!
kimi.com/?utm_campaign=TR_zwkjw3Tw&utm_content=&utm_medium=tiktok&utm_source=CH_O3fWqjAg&utm_term=
#moonshotai #kimi #agenticai #short
