Issue Planner · Case study

From backlog to executable plan

Human-in-the-loop AI planning that turns a vague issue into a repo-correct, PR-sized plan a person can trust.

The challenge

Most wasted effort is decided before the first commit.

The gap between an accepted issue and the first line of code is where scope creep, rework, and stalled tickets collect. Three problems sat underneath it.

Tap a point on the path to see each one.

My role

I led the design end to end.

A capable planning pipeline already existed. My work was the experience on top of it, and the way we'd know it worked.

Toggle to see what I owned versus where I partnered.

Strategy

A product line, not a feature.

Reviewing code is downstream; we arrive once work is half done. Planning moves the product upstream, to where work is shaped.

Switch the surface to see what changes.

Discovery

One job, three people who need it.

“A plan good enough to start from, without the work of writing it.” It read differently for each.

The design

Five decisions, one rule: the person decides.

The AI drafts a full plan, but the default action is review. Edit in place, replan in chat, revert without risk, hand off cleanly.

Try the tabs.

Measurement

One metric that is hard to fake.

The north star sits at the end of the loop: plans that become a merged PR, linked back to the issue. Beneath it, a funnel shows where plans drop off.

Hover a stage. The dashed one is the part worth instrumenting.

Reflection

What I would carry forward.

The decisions that mattered were about doubt and control, not layout.