⌁boltmob

▸ an AI-DLC workbench · self-hosted

Agents draft.
Your team decides.

boltmob is a workbench for the AI-driven development lifecycle. Work runs as units, delivered in bolts — days, not sprints. AI agents draft the plan, the code, and the docs. They raise questions before they assume. Your team answers together, steers, and signs every artifact before it ships.

▸ the record — every unit leaves one

who asked, who answered, who signed — captured as the work happens, not reconstructed after.

▸ the loop

How a bolt runs

A bolt is one short burst of work: from intent to signed artifact. Four moves, in order, every time.

01

Frame

Describe the unit of work in plain language, in chat. Agents turn it into a draft — the plan, the code, the docs, as one piece.

02

Interrogate

Agents raise questions before they assume. Ambiguity surfaces at the start, where it's cheap — and the mob answers it together.

03

Sign

Nothing ships unsigned. A person reviews the artifact, requests revisions, and puts their judgment — and their name — on it.

04

Remember

The record of who decided what, and why, stays with the work. It compounds into team memory that sharpens the next bolt.

▸ the product

What that actually looks like

Three views of one unit of work, from the running app. No mockups.

01

The agent asks, in your thread

A boltmob message in Slack titled 'Question raised — SSO Migration Spec'. It shows the ambiguous line from the spec, the unit name, the phase, who raised it, and the question — with answer buttons: Internal only, Both internal + external, Assume and flag, Open Unit in boltmob.
An agent hit an ambiguity in the spec and stopped instead of guessing. The question lands where your team already talks, carrying the exact line that was unclear. The answers are buttons. What gets picked is recorded against the unit.
02

Every step is a bolt, and one of them is a person

The artifact graph for a unit in boltmob: an inception lane holding the approved v1 draft with its answered questions, a dashed line marking the transition into construction, then three bolts chained left to right — Implement the change (developer), Sweep docs for drift (tech_writer), Sign approval (human_signer).
The unit's graph, left to right. The approved draft and the questions that shaped it, the line where inception becomes construction, then the bolts: developer, tech writer, human signer. Click any one of them for that agent's full trajectory.
03

And then it stops, for you

The boltmob console showing a unit in construction. The left rail lists Units, Questions, Workflows, Substrate, Enforcement and Memory. The unit min-signers-on-open is marked active:construction, and its Construction card reads AWAITING_HUMAN with an amber 'WAITING ON YOU — HUMAN SIGN GATE' panel offering Sign approval, or Reject with a required reason.
The run reaches the sign step and will not continue on its own. Approve it and your name goes on the audit trail. Reject it and the reason is required, and recorded against the bolt that failed.

▸ the principle

We refuse to grade the people doing the work — we surface the work itself.

AI is a tool for good when the right people steer it. boltmob is the surface where humans tune agents to their team, sign their judgment on every artifact, and capture what they learned together. Humans aren't a constraint we tolerate in the loop. They're the thing that makes the output worth shipping. They always will be.

▸ what you run

self-hosted

Yours, on your infrastructure

Ships as a Docker deployment you run yourself. Your repos, your keys, your record — none of it leaves your network.

where you talk

Meets your team in chat

The work moves out of ticket queues and into the conversation you're already having. The who, what, and when is captured as you go.

bolts, not sprints

Cycles in days, not weeks

Built on the AI-DLC method: short bursts of elaboration and delivery, with the whole mob steering. Momentum stays high; drift stays low.

▸ the method

What is AI-DLC?

AI-DLC — the AI-Driven Development Lifecycle, also written AIDLC, AI-SDLC, or AISDLC — is a software delivery method introduced in AWS's AI-DLC paper. AI agents draft the plan, the code, and the docs; they raise their questions up front instead of assuming; and the team answers together, steers, and approves the result. Work is scoped into units and delivered in bolts — cycles measured in days, not sprints.

How is AI-DLC different from the traditional SDLC?

The traditional SDLC treats AI as an assistant inside human-paced phases. AI-DLC inverts that: agents carry the drafting continuously, and humans concentrate where their judgment matters most — framing intent in plain language, resolving ambiguity when agents ask, and signing every artifact before it ships.

What are bolts, units, and mobs in AI-DLC?

A unit is one piece of work, framed in plain language. A bolt is the short burst — days, not weeks — that takes a unit from intent to signed artifact. The mob is the whole team steering together: answering agents' questions, reviewing drafts, and putting their names on what ships.

What is boltmob?

boltmob is a self-hosted workbench that runs the AI-DLC method end to end: agents draft, the mob steers in chat, and every unit leaves a signed record of who asked, who answered, and who approved. The name is the method — bolt plus mob. Currently in private beta.

Does AI-DLC remove humans from the loop?

No — it moves humans to the decisions that matter. Agents draft and ask; people frame the work, resolve ambiguity, and sign every artifact before it ships. In AI-DLC, nothing ships unsigned.

▸ status

In private beta. First conversations under way.

boltmob is being built and battle-tested now, and it ships when it's ready to be signed for. Leave an email above, or write to us — we read everything.

jason@blackbandanas.com