Capstone

Agentic Workflows · ~10 min

The Workflow Decision Table

Eleven lessons, one reflex: read the situation, reach for the move. This is the whole course as a lookup table — and a mixed review to prove it stuck.

Why this, for you: a workflow is a habit, not a checklist. When a task lands or an agent misbehaves, the instinct is to wing it. This table retrains that instinct — most situations map to a specific, named move you already learned.

The through-line of the whole course: structure the work before you run it, and put the human where judgment beats persistence. Diagnose the situation, pick the move, and let automation carry the rest. Here's the diagnostic map.

1 Situation → move

SituationMoveFrom
Non-trivial change in code the agent may misreadMake it summarize the subsystem read-only, correct it, then plan before any editL1
Wrong assumptions are expensive to surfaceResearch → plan → implement; sandwich reasoning high at plan/verify, low at executeL2
You keep fixing the same class of bad output by handMove onto the loop — fix the harness that produced it, not the artifactL3
An irreversible vs. a reversible actionGate by reversibility; review the decision, not the executionL3
One agent is the bottleneck on independent workFan out the factory model — parallel sessions, automated feedback, you orchestrateL4
Parallel agents would collide on shared filesWorktree isolation — one branch, one PR each; single-branch + guards past ~10L5
You want an agent to run on every PR, unattendedHeadless in CI — cap --max-turns, shift permissions off promptsL6
A task's final tenth needs human taste or unreachable infraBackground→foreground handoff at the judgment threshold, with a distilled summaryL7
"Get this unfamiliar repo running" keeps failing silentlySnapshot before state changes; verify with prosecutor-judge, not exit codesL8
A monolithic agent prototype fails silently in productionThe five-step refactor: sequence, schema, dynamic RAG, tracing, circuit breakersL9
The codebase rots between commitsEntropy reduction agents — encode taste once, scan on a cadence, one-violation PRsL10
"Done" for an agent feature is undefined or slipperyWrite the eval first; baseline before building; the low pass rate is the surfaceL11

2 The two rules under all of it

First: structure beats improvisation — phases, plans, schemas, and evals make the expensive failure (wrong assumptions becoming code) cheap to catch. Second: put the human on the loop, gated by reversibility — fix the harness for recurring failures, review decisions not execution, and reserve your attention for the judgment the automation can't supply.

Part 1 was the inner loop you run on one task: plan first, decompose into phases, and decide where you stand relative to the agent. Part 2 scaled that out: parallel sessions move the bottleneck to your review bandwidth, worktrees and single-branch keep swarms from colliding, headless agents bound the pipeline, and handoffs move ownership cleanly. Part 3 kept the codebase healthy across all of it: reversible setup, the prototype-to-production refactor, scheduled entropy reduction, and eval-first development. Each is the same move — structure the work, place the human well — applied at a different layer.

Every move had a "when not to"

The reflex is to apply the strongest move by default — and every lesson warned against it. Plan-first is overhead on one-sentence changes; research doesn't compound in well-mapped code; the factory model needs deterministic verification or it inflates the 41–87% multi-agent failure band; single-branch is riskier than worktrees without its preconditions; entropy agents can't substitute for fixing a slop factory; evals-first misleads on novel or subjective work. Match the move to the cost of being wrong — heavy structure for durable, expensive work; light touch for throwaway spikes.

↪ Your win: a workflow reflex

Mixed review — across all eleven lessons

Question 1 · from L2The reasoning sandwich spends the most reasoning at…

Question 2 · from L3"On the loop" means that, faced with a bad output, you…

Question 3 · from L6For an unattended CI run, the flag you must set is…

Question 4 · from L8Repo-setup agents verify with prosecutor-judge because…

Question 5 · from L11Writing an eval before the feature means a low baseline pass rate is…

You finished the course. Ask me to apply the decision table to a real task of yours — draft a plan-first loop for a feature, wire a headless PR-review workflow, set up an entropy-reduction cron, or write the first eval suite for an agent feature. Or revisit any lesson — the moves compound when you use them together.
✎ Feedback