Engineering for the Next Agent, Drawn

Agent engineering builds on itself. Skills, instruction files, tools and evals written for last year's model still work with this year's, and often work better. Prompt workarounds and step-by-step scaffolding get deleted. The model you build for today is usually replaced within months, so this explainer is about engineering for the agent that is coming: what to invest in, what to keep thin, how to read the capability trends without overtrusting them, and how to upgrade without breaking production.

Step through each diagram with the buttons or your arrow keys, then use the 30-second answer, the gotchas and the quiz. Every number is dated and linked to the lab, vendor or researcher that published it; forecasts are shown with their caveats.