09 · The ladder

Agents can't plan against data they can't query

You've consolidated systems before and regretted some of them. This page is the argument for why this consolidation is different — one rung at a time, each rung paying for itself.

Right now your lifecycle is scattered by design: issues in one SaaS, decisions in a docs folder or someone's Drive, review reasoning in PR threads, incident knowledge in dashboards and postmortem docs. A senior engineer holds the connections between them in their head. An agent can't — and neither can the next engineer.

Files are readable but not queryable. An agent can open your design doc; it cannot ask what's blocked by it, what depends on it, or whether the incident from March already invalidated it. Planning against unqueryable data is improvisation with extra steps — which is exactly the failure you've watched AI coding tools repeat. The fix is not a smarter agent. It's putting the lifecycle where every seat can ask it questions: the work-graph.

One at a time

The rungs — each pays for itself

1 Sync the issue tracker

Bi-directional sync; beads becomes the working copy, your tracker keeps running.

Payoff: "what's next" is computed from real dependency edges, and the plan survives any agent losing its context.

2 Move the design docs in

Backfill proposes your ADRs, specs and git history as beads — reconciled, never blind-imported.

Payoff: planning cites recorded decisions instead of re-litigating them, and a decision has an address other work can depend on.

3 Record reviews in the graph

Review verdicts become gates — records, not vibes in a chat thread — signed by whoever held the seat.

Payoff: a bounce targets the process that produced the defect, not just the diff, so the correction sticks.

4 Hook up the telemetry

Production signal joins the same graph the planner reads.

Payoff: an incident doesn't end at mitigation — it becomes a guardrail in the next plan, the loop everything else on this site exists to close.

No leap required

Nothing is abandoned first

Rung one is a sync, not a migration. Your tracker, your docs, your dashboards all keep working; beads is simply the copy the agents plan against. What happens next is the same thing that happens to any mirrored system: one copy is where the decisions get made and the other goes quietly stale. When you retire the old tool — if you do — it's because nobody had opened it in a month, not because a landing page told you to.

The view from above

Where the ladder tops out

With the whole lifecycle in one queryable graph, the seats can run without you in the loop of every step: planners file gated plans, developers work headless in their own worktrees, reviewers and wardens hold their gates, and production reports back into the next cycle. Your seat rises to where a CTO's should be — reading the fleet at a glance, approving at the gates that matter, steering by what the graph says instead of by reading everything yourself.

The seats and who holds what
The fleet view — every hive from one seat