In the news
FrontierCode 2.0
Cognition · Published · 3 min read
In 30 seconds
- What happened
- Cognition released FrontierCode 2.0, a harder coding benchmark with larger tasks, stricter grading, and broader repository coverage to better evaluate AI coding agents.
- Why it matters
- Engineers building or evaluating AI coding systems need this to understand how well models handle real-world code contributions beyond simple spec-following.
- Watch out
- Top model scores only 45 percent, and models often score lower with more reasoning effort, suggesting current agents struggle with scope judgment and inferring unstated constraints.
The patterns behind this
- MMAU: Massive Multitask Agent Understanding
- TheAgentCompany Benchmark
- Agentic Context Engineering (Evolving Playbook)
Each one covers how the technique works, when it earns its cost, and where it breaks.
The Agent Architect
One pattern, one tradeoff, one production failure story. A short weekly briefing for people building agentic systems.
Weekly email, one-click unsubscribe. We only use your address to send the briefing.