Kimi's CEO Zhilin Yang: "everyone's trying to build one smarter agent many agent hits a wall fast, so instead of making it smarter we just made more - one boss, a thousand workers." in a 39-minute talk he explains why one agent won't get you to real work. agent swarms + long context + RL on every sub-agent - that's the fix. bookmark this ↓
sunick
@Serantych
Jul 2
Article

Stop prompting, start looping: the Boris Cherny method

Everyone is optimizing their prompts. Boris Cherny threw the whole idea out. He built Claude Code. He runs the team. And he uses the tool in a way that looks nothing like how most people use it. No

Jul 17, 2026 · 4:43 PM UTC

39
376
2,578
388,152
Sort replies: Relevant Recent Liked
Replying to @Serantych
more agents isnt the fix, smaller scoped agentsis, Kimi just found out what claude code sub-agents have shipped for months
1
2
3,323
interesting point. the future is probably specialized agents working together
2,535
Replying to @Serantych
The orchestration layer is becoming just as important as the model itself. Coordination can be a bigger bottleneck than intelligence
1
2
16
5,388
that’s the key point
4
4,650
Replying to @Serantych
这货为什么不说中文
3
3,660
好内容不分语言
1
3,113
Replying to @Serantych
yo, never seen this one, that’s a nice watch, thanks
1
2
5,317
glad it made your watchlist
1
4,712
Replying to @Serantych
This is a fascinating approach to problem-solving. It's like an ant colony, but for AI.
1
186
Replying to @Serantych
his talk from this year orcadub.orcarouter.ai/share/…

Clodex - the CEO behind China's Kimi models says the quiet part out lo… — Mandarin → English dub |...

Voice-cloned video dubbing. Start free — up to 2-min clips, 28 languages.

orcadub.orcarouter.ai
21
Replying to @Serantych
The future might not be one genius AI, but a team of specialized ones.🔥
1
116
Replying to @Serantych
@grok summarize this entire videos and give very specific details on key things from this video
2
2
1,434
Replying to @Serantych
agents are like humans. once you realize that you embrace ai as a researcher and as a user. at least for the early few years
1
88
Replying to @Serantych
that's a wild take from Yang. agent swarms over single super agents makes sense for real world complexity
1
914
Replying to @Serantych
Esto es la base de resolver cualquier problema complejo, dividirlo en pequeños problemas sencillos 👏🏻👏🏻
1
191
Replying to @Serantych
Saw this in @TheUpsideAI this morning, definitely caught my attention.
1
4
Replying to @Serantych
What a banger
1
1,195
Replying to @Serantych
Smarter agent scale isn't always the solution.
1
1,549
Replying to @Serantych
@grok explain the role of RL in this loop
1
936
Replying to @Serantych
他很紧张啊，能听出来面对如此多的观众演讲时，声音在颤抖。
518
Replying to @Serantych
I'm hustling to get swarms for my new A2A but feel limited any tips? I'm on all registries, I added X402, good grades glama etc...
18
Replying to @Serantych
@grok 非エンジニアにもわかるように優しく解説してください
508
Replying to @Serantych
splitting tasks across worker agents is where this actually gets useful
139
Replying to @Serantych
贼眉鼠眼的
172
Replying to @Serantych
One boss and a thousand workers is a great way to frame agent swarms
609
Replying to @Serantych
@ASI_Alliance already has this.
345
Replying to @Serantych
One boss and a thousand workers is a clever way to frame agent swarms.
447
Replying to @Serantych
duh. are there people who don't understand this? It's literally in the design of a lot of tools.
69
Replying to @Serantych
하나의 에이전트를 고도화하는 대신 스웜으로 접근한 게 흥미롭네요.
800
Replying to @Serantych
the "one boss, a thousand workers" framing is perfect. single agent architecture is like trying to hire one person who's amazing at everything instead of building an actual team
324
Replying to @Serantych
The shift from one genius to a swarm of workers changes everything about AI labor.
4
Replying to @Serantych
Kerjaan cepet kelar tp bos malah nambahin load baru, tetep aja gak bisa pulang tenggo.
378
Replying to @Serantych
Study $INSTACLAW
5
Replying to @Serantych
yea ive been using agent swarms with free models and just having frontier models with low rate limits review the work. I've created some really phenomenal things with my ai army so far.
1
894
Replying to @Serantych
If you are in hurry; tl’dr: linkedin.com/posts/jindalsan…

Scaling Kimi AI Across 3 Dimensions for 1T Parameters | Sandeep Jindal posted on the topic |...

💡 Is brute-force compute scaling hitting a wall? Moonshot AI is scaling Kimi across three entirely different dimensions: At Nvidia GTC 2026, Zhilin Yang revealed how they scaled Kimi K2.5 into a...

linkedin.com
16
Replying to @Serantych
Hi , I build apikeys.in to access all ai llm models on one platform,it also supports UPI payment no visa credit card required,It is more transferent and better then openrouter

Yantra AI — One key. Every model. Pay with UPI.

An LLM API gateway with one OpenAI-compatible endpoint that intelligently routes every AI API request. Indian-rupee billing over UPI.

apikeys.in
11
Load more