Cognition: Letting Models Do the Heavy Lifting
petergyang · x · 2026-07-13
Jared shared several key takeaways after observing top engineers using AI agents at Cognition:
- Fewer rules are more effective: When building agents, too many constraints actually weaken performance; the key is letting the model "do its thing."
- Break tasks down: Split work to keep the context window of each sub-agent smaller.
- Let agents self-check: Design mechanisms for the model to verify its own outputs.
- Agents managing agents: A live demo showed how one agent can manage a team of agents.
He also mentioned that an increasing number of Devin sessions are no longer initiated directly by humans but are passed off from other Devin instances. The team's goal isn't "token max," but rather "ROI max."
Related event: Cognition Shares AI Agent Engineering Insights(3 posts)→
More from coding & agent
- Building a Secure AI Agent Gateway: Self-Hosting OAuth for Multiple SaaS Apps — Defiant_Cod_2654 · 2026-07-22
- Rowboat launches as an open-source, local-first AI coworker with memory — ycombinator · 2026-07-22
- Scoble says AI “loops” really means long-running multi-agent workspaces — Scobleizer · 2026-07-22
- Kimi Code opens a waitlist as Moonshot rolls out its coding product — Fabulous_Bonus_8981 · 2026-07-22
- Open-source runtime lets each repo define its own AI code reviewer — ibabufrik · 2026-07-22
- Indie Dev Asks: What's Actually Broken in Your AI Agent's Memory Today? — AcceptableTime7937 · 2026-07-22