a builder's codex
codex · patterns · Agents mapped 1:1 to JTBD with named human checkpoints

Agents mapped 1:1 to JTBD with named human checkpoints

Convergence

Five operators converge on the same agent-design pattern: don't build an "AI-augmented" abstract layer. Enumerate the jobs-to-be-done. Map each to a named agent (or agent class). Insert named human checkpoints at the alignment-critical moments. The mapping is concrete, not generic.

Operators

Variation

Implication

Before building any agent, write the JTBD list. Before deploying, name the human checkpoint and the trust tier. Before scaling, classify the agent (Spiegel/Farooq taxonomy) and decide its loop (CASH/CCCD). If you can't draw the wiring diagram on a whiteboard, you're not ready to build.

Sources

Open the interactive view →