Skip to content

Let the model plan. Make code enforce.

Engineering field guide · Workflows and trust · Both days

Which parts of the workflow should be deterministic?

Keep approvals, execution rules, arithmetic, and source checks explicit. Compare typed plans, policy-driven loops, and verified citations.

All topic guides · Practical takeaways

Perpetual, Coinbase, and Kepler make important constraints explicit in software. They let models interpret work while keeping required execution rules visible.

Session Approach
Perpetual · Day 1 Compiles a typed plan into an execution graph with explicit waits and approval states.
Coinbase · Day 2 Keeps the outer process policy-driven while allowing adaptive loops inside individual steps.
Kepler · Day 2 Uses source identifiers and code to retrieve facts and calculate numbers. It checks provenance as part of producing the answer.

Perpetual · Day 1 · Coinbase · Day 2 · Kepler · Day 2

Deterministic execution does not make an incorrect plan correct. A valid source identifier can still refer to the wrong period or document. Planning, retrieval, and policy checks each need their own evaluation.

  1. Choose a workflow with one factual lookup and one action requiring approval.
  2. Make the model produce a typed plan. Have code validate it, fetch facts, and enforce approval.
  3. Test a missing source, an old source, a rejected approval, and a retry after interruption.

Count unsupported facts, actions taken without approval, duplicate actions, and successful recovery. Inspect the cases the system refuses as well as those it completes.

The comparison and experiment are our editorial synthesis. They are not a reported benchmark or the speakers’ prescribed procedure.