Harry Chen

pm @PlaudAI — the ai notetaker with 2M user, copied by @heypocket
alum @columbia_biz, ex @lovart_ai
writing from inside ai hardware, between sf and china

[Harry Chen @harrychen404]

best blogs of the week (so insightful that i read at least twice)

Coding Agent:
How @cognition reduces Devin's token cost by harnessing: https://t.co/vJGifNjAcp
How @cursor_ai builds complex projects with low redundancy through swarm: https://t.co/a8JFjJrb1h
How @claudeai Code designs context for new model-driven agents: https://t.co/OH7p51iRi5

Robots:
@MistralAI SOTA visual navigation model: https://t.co/uCT594hmAh

Vision models:
@GoogleDeepMind unveils the source of creativity for diffusion models: https://t.co/8avXySmhwf

[Harry Chen @harrychen404]

@danshipper @OpenAI @every What a come back for @OpenAIDevs . No one can expect this months ago. From user experience codex 8 versus Claude code 6.5. Agree?

[Harry Chen @harrychen404]

@sama I believe Zhilin Yang @Kimi_Moonshot and Wenfeng Liang @deepseek_ai don’t take long weekend trips

[Harry Chen @harrychen404]

@ajs6888 @cognition @cursor_ai @claudeai Exactly

[Harry Chen @harrychen404]

best blogs of the week (so insightful that i read at least twice)

Coding Agent:
How @cognition reduces Devin's token cost by harnessing: https://t.co/vJGifNjAcp
How @cursor_ai builds complex projects with low redundancy through swarm: https://t.co/a8JFjJrb1h
How @claudeai Code designs context for new model-driven agents: https://t.co/OH7p51iRi5

Robots:
@MistralAI SOTA visual navigation model: https://t.co/uCT594hmAh

Vision models:
@GoogleDeepMind unveils the source of creativity for diffusion models: https://t.co/8avXySmhwf

[Harry Chen @harrychen404]

@DecartAI this is so fun to play with! good job @DecartAI Lucy 2.5 for real-time visual effect and video editing! haven't feel this kind of aha moment in video ai for a while.
strong recommend!!!

[Harry Chen @harrychen404]

@trq212 i think ppl neglect the 50% improvement on business automation benchmark.

[Harry Chen @harrychen404]

@danshipper @every i think ppl neglect the 50% improvement on business automation benchmark.

[Harry Chen @harrychen404]

@thsottiaux fyi @MistralAI already have vibe in their product name

[Harry Chen @harrychen404]

@claudeai tldr: sota on coding and knowledge work
@Kimi_Moonshot K3 still wins on browser use and 3D

btw why agent coding decline after $15 effort🤔

[Harry Chen @harrychen404]

@istdrc great pms are always great philosophers

[Harry Chen @harrychen404]

@sophiamyang @MistralAI @MistralAI finally get sota on a model! huge congrats!!!

[Harry Chen @harrychen404]

@danshipper i quit thinking about agi. i just view LLM as tool-caller + knowledgable worker in some area

[Harry Chen @harrychen404]

@rork Bye bye ExpoGo

[Harry Chen @harrychen404]

@danshipper is not building @every as a media+product comp, rather a professional school with tools, like @DeepLearningAI / @coursera with digital materials and assets u will use in courses.

[Harry Chen @harrychen404]

@hthieblot The second I recognize it’s AI, I swipe left

[Harry Chen @harrychen404]

@jack FYI u r copying @raft_hq

[Harry Chen @harrychen404]

@jiaying_ai @jack @Kimi_Moonshot @istdrc @raft_hq No wonder they lay off so many employees

[Harry Chen @harrychen404]

@jack Buzz accidentally “copied” idea from @Kimi_Moonshot

@istdrc used to work at Kimi and built @raft_hq months before buzz launced

raft is a good product. i use it everyday. it is unfair to see the credits flow to a wrong place

[Harry Chen @harrychen404]

@paulg It’s always switching cost versus sunk cost.

[Harry Chen @harrychen404]

@WilliamBryk Okay my agent can finally learn scientific contradict stuff from academic research