jonnifyagile agents
jonnify / Agile Agent Workflow / A3 · The roadmap layer / A3.2 · Extreme Programming for small batches

A3.2 · The roadmap layer

Small batches

Small releases, incremental design, and continuous feedback — the three Extreme Programming practices that make a release cheap to ship, review, and undo, re-cast for an Author/Operator pair.

A smaller batch lowers risk and feedback latency and raises learning per cycle. Shrink the dial on the right across the Portal's real nine-rung F6 ladder and watch all three move. A Claude agent that ships one well-specified rung at a time is what makes a small batch the cheap default rather than a discipline you must force.

Shrink the batch · risk and latency fall, learning rises

3
F6 ladder · 9 rungs, split into releases of the chosen batch size each dashed divider ends one release and starts the next · a review fires at every release boundary
Batch = 3 rungs per release · 3 releases across 9 rungs. Risk index 33 · feedback latency 3 rungs · learning per cycle 33. · A smaller batch lowers risk and latency and raises learning per cycle.

One dial proves the module: as the batch shrinks toward a single rung, risk and feedback latency fall and learning per cycle rises.

The batch argument

A batch is the work shipped between two reviews. Extreme Programming’s small-batch practices all push the batch toward a single increment, because a smaller batch is cheaper on three axes at once. Less is at stake if a release is wrong, so risk falls. A reviewer meets the work sooner, so feedback latency falls. And the same total work passes through more reviews, so learning per cycle rises. The dial above moves all three over the Portal’s nine-rung F6 ladder; this module unpacks the three practices that hold the batch small.

The new argument this module adds: a small batch used to be a discipline a team had to enforce against the pull of “ship it all at once.” With a Claude Author that turns one rung into a spec triad, a slice, and a green harness, a batch of one is cheap to produce — so small batches become the default the workflow runs at, not a cost it pays.

The principle

Keep the batch small — small releases, incremental design, continuous feedback — so each release is cheap to ship, cheap to review, and cheap to undo.

On the Portal

The phoenix.roadmap.md is nine rungs, each a PR-sized slice shipped and demoed before the next is planned — the agent makes each batch of one cheap to produce.

Small batches are an XP idea; an agent that ships one well-specified rung at a time makes them the cheap default rather than a discipline you must force.

The three practices

Three dives carry the module — one per XP practice, in order. Each is re-cast for an Author/Operator pair and grounded on the real F6 ladder: small releases (the batch of one), incremental design (the design grown rung by rung), and continuous feedback (the loop closed between rungs).

The F6 cadence the dives build on

All three dives stand on one real artifact: the Portal’s phoenix.roadmap.md. It climbs nine rungs f6.1…f6.9 across three milestones — ship the catalog, make it live, ship to users — one shippable capability per rung. Each rung is a PR-sized increment: a spec triad, the slice, a green harness, a demo, and a feedback note. The roadmap states the per-rung loop verbatim: sharpen → build → ship → demo → review → feedback → adapt. That cadence is the three XP practices running together.

small release
F6 ships one rung, demos it, and takes feedback before the next is planned — a batch of one (A3.2.1).
incremental design
Each rung adds only the web surface its slice needs, behind the unchanged Portal facade; PubSub, auth, and the dashboard are deferred to the rung that needs them (A3.2.2).
continuous feedback
Every rung asks one feedback question and runs the loop; feedback edits the spec, the single source of truth, never the shipped page (A3.2.3).

The first dive opens the batch argument in full: A3.2.1 · Small releases.

References

The sources behind small batches and the loop, and where this module connects in the course and to the real F6 chapter it is grounded on.

Sources

Related in this course