Resources

Six stages. One Bolt.

A Bolt is the unit of work in BoltOps — a 4–8 hour AI development work cycle. The Bolt Lifecycle takes it from intent to ship across six stages with AI-powered transitions and human gates where they matter.

Core concept

What is a Bolt?

Story points, t-shirt sizes, and 2-week sprints were designed for a world where humans wrote every line of code. A Bolt is the AI-first replacement: a contained 4–8 hour cycle where one or more AI agents (and a human reviewer) take a clear intent through to production. Fast enough that you can ship multiple per day. Big enough that it’s a real piece of work.

4–8 hours, end-to-end

Sized for what an AI agent and a human can finish in a single sitting. No multi-week tickets that decay into stale context.

📝

Anchored by an Intent Document

Every Bolt starts with a structured spec: title, priority, user stories, acceptance criteria, edge cases. Claude Assist fills in the details.

📄

Governed by CLAUDE.md

The CLAUDE.md for the project is the spec the agent is held to. Managed, versioned, and synced — not a stale file rotting in your repo.

The Bolt Lifecycle

Six stages, AI-native

Every Bolt moves through six stages on a Kanban board. Drag-and-drop, with AI-powered transitions and human gates at the points that matter most.

1

Intent

A PM (or Intent Architect) authors a structured Intent Document. Title, priority, user stories, acceptance criteria, edge cases, dependencies. Claude Assist generates the first draft from your title and project context.

2

Generate

Dragging the Bolt into Generate auto-dispatches a Claude Code session wired to your CLAUDE.md. Multiple Paperclip agents can work in parallel — one writes code, one writes tests, one updates docs — with budget caps and approval gates.

3

Verify

The verification engine runs lint, type check, test suite, security scan, and CLAUDE.md compliance. Every check must pass before the Bolt can move to Review.

4

Review

A human reviewer (or external stakeholder via a Magic Link) reviews the artifact and submits severity-rated feedback. No login required for external reviewers.

5

Iterate

Stakeholder feedback flows back into another agent loop. Most Bolts pass on the first iteration — the average is 1.3 cycles.

6

Ship

Shipped Bolts get coordinated into a Release Train. The Release Captain holds the scope gate. PR description, test results, and change summary are auto-generated. Merge and you’re live.

4 hrs
Median Intent → Ship
1.3
Average iteration cycles
94%
First-pass accuracy
0
Sprint ceremonies

Ready to ship at Bolt speed?

Join engineering teams that have replaced sprint ceremonies with AI-orchestrated workflows.

Get Started Free