Intent Engineering framework: /goal covers only 2 of 8 parts of agent intent

PawelHuryn · x · 2026-09-09

Paweł Huryn published a long-form guide on the Intent Engineering framework for building reliable AI agents.

Core thesis: agents fail not because they can't reason, but because their objectives, outcomes, and constraints are underspecified. The fix isn't more detailed instructions—it's completing the intent. Intent determines how an agent acts when instructions run out, and organizations have codified far less than they think.
Platform shift: OpenAI shipped /goal in Codex on April 30 and Anthropic shipped /goal in Claude Code on May 12—both let you set a measurable completion condition the agent pursues autonomously. But /goal covers only 2 of the 8 parts of intent; the other six are still yours.
What's inside: the guide breaks down what intent actually means, where /goal stops, the 8-part intent structure, the workflow, a complete customer-support-agent example, and a "do this tomorrow morning" checklist.
Quotes Shopify CEO Tobi Lütke on context engineering: stating a problem so it's plausibly solvable without extra information is the fundamental AI skill—but context without intent is just noise.

More from coding & agent