Product · Ship
Closed betaWorkspace for AI-assisted product delivery.
Repo, tracker, policies, knowledge, automation, and evidence — one workspace. Owners see what is moving, what is blocked, who decided, and which rules bound the work before an agent touches the code.
The problem
Most product companies run on processes nobody can read.
And then they hand AI agents the keys.
Before Ship
Invisible processes drift.
The process exists — just not in writing. It lives in the muscle memory of three senior engineers, in a Slack thread from last quarter, in a wiki page nobody updates. When two people disagree, neither version wins. When agents start running, they amplify the disagreement at machine speed.
With Ship
Written-down processes can be improved.
A process you can read is a process you can argue about — and then change. Ship gives every workspace a place to write the process down: states, transitions, owners, the routines that fire along the way. Once the process is legible, the work that follows it is trackable, and the work that breaks it is visible.
The model
Process on specialists.
Specialists on executors.
Five layers. Swap your executor tomorrow and nothing else changes.
01
Workspace
A graph of processes — Development, Marketing, Support. The workspace is the boundary: everything inside it shares the same policies, knowledge, and evidence trail.
02
Process
A sequence of named states with allowed transitions. Work cannot skip review or stay in-progress for ninety days without flagging.
03
Routine
A named, recurring job — security scan, daily digest, architecture review. Each routine always runs inside the process so context is never wrong.
04
Specialist
A versioned role definition — developer, QA, architect, marketing operator. Same routine + different specialist = different output.
05
Executor
The AI agent — Cursor, Claude Code, Codex, Copilot. Swap tomorrow and the routine, specialist, and process stay unchanged. That separation is the whole product.
Capabilities
Why teams buy control, not another AI demo.
Legible process
A process you can read is a process you can improve
Name every stage, owner, and transition. Work that breaks the process is visible the moment it breaks — not during the post-mortem.
Bounded agents
Agents inside fences, not in the open field
Specialists, routines, and executors run against policy and retrieval. Policies are injected into every system prompt — workspace-wide, no opt-out.
Evidence by default
Audit trail as a side-effect, not a project
Decisions, retrievals, and outcomes are captured as artefacts. Every action points to a ticket, branch, PR, or knowledge article.
Knowledge engine
Retrieval that knows the workspace
Docs, policies, and codebase ingested into knowledge buckets. The distiller promotes signal to atomic claims. Agents retrieve facts, don't hallucinate.
Human inbox
Decisions stay with humans
Clarifications, approvals, and proposals land in a structured Inbox. Every disposition — accept, reject, defer — is a record.
Zero migration
Connects to what you already use
Linear, GitHub, Jira, Cursor, Claude, Codex. Ship is additive — existing automations stay exactly as they were.
Development process
The SDLC, fully drawn.
Eight named states in three phases, eight scheduled routines, fifteen versioned specialists — all diffable, reviewable, and running in production.
Phase 01 · Requirements
Intake
Intake specialist
Requirements
Business analyst
Phase 02 · Implementation
Architecture plan
Technical architect
QA plan
QA engineer
Implementation
Developer
Phase 03 · Review
Manual QA
QA engineer
Automation QA
QA automation
PR review
Code reviewer
8 states · 3 sub-processes · continuous flow
States
Routines
Scans dependencies and secrets policy.
Consolidated summary of in-flight work and blockers.
Async standup nudge with state and blocker summary.
Architecture drift and design consistency check.
Reconciles CI, workflows, and guardrails after failed runs.
Lightweight retro prompts and follow-up actions.
Triages and sizes technical-debt work for upcoming cycles.
Recurring check on test architecture and coverage.
Specialists
Fifteen versioned roles, ready to run.
A specialist is a role definition, not a person. Versioned, diffable, swappable without touching the process.
Engineering
- Technical architect
- Developer
- Code reviewer
- QA engineer
- QA automation
- DevOps / platform
- Security engineer
- Data / ML engineer
Product
- Intake specialist
- Business analyst
- Product manager
- Designer
Operations
- Support / success
- Technical writer
- Marketing operator
How it works
From workspace to evidence in five steps.
01
Create a workspace
Start with the founder, product owner, or product area that owns the outcome. Ship keeps that scope visible before any automation runs.
Workspace · members · policy
02
Connect the repo
Install the GitHub App, activate the repositories Ship should observe, and let the console detect what is already wired.
Repo · GitHub App · bundle
03
Bind the tracker
Tie work back to Linear, GitHub Issues, or the tracker your team already uses so intent stays human-owned.
Tracker · states · owners
04
Set policies and knowledge
Give agents and reviewers the product facts and boundaries they need: brand, code style, review rules, and repo context.
Knowledge · policies · secrets
05
Review decisions
The dashboard and Inbox show blockers, clarifications, improvements, shipped work, and the evidence behind each action.
Dashboard · Inbox · PR evidence
Ship ships Ship
Zero lines of human-typed code.
This entire workspace — backend, console, landing, CLI, docs, blog — was written by AI agents under Ship's own process. Humans reviewed, decided, merged. No one typed code by hand. The product is the proof of the workflow.
0
Lines hand-typed by humans
Every production line was written by an AI executor. Humans set policy, reviewed PRs, and merged.
236k+
Lines of code shipped
137k Python · 73k TypeScript · 26k Markdown — all AI-authored under Ship's own process.
608
Commits in 30 days
From extraction to today. Every commit on main, public, with the AI executor named in the trail.
7
Team members, none typing code
Leadership and engineering review diffs, name routines, set policy. The product proves the workflow scales.
From the canon
496 docs ingested → 708 atomic claims → 98 auto-rendered topics. Same workflow.
/knowledge — the workspace's own knowledge base
From the catalogue
25 specialists, 9 routines, 1 production process — running on themselves.
— catalogue v0.13
From the changelog
Claim-graph P0 → P5 shipped in 36 hours. The same agents wrote, reviewed, and merged it.
— 2026-05-05 → 2026-05-06
Reference deployment · ElMundi
From scattered Slack pings to one Inbox.
The team replaced ad-hoc agent prompting with a reviewable product workspace: every ticket traced from Linear to branch, PR, Playwright evidence, and the Inbox decision that needed a human.
"Tickets close in fewer hops because the agent and the human share a vocabulary. Code review effort moved from setup-and-context-rebuild to the part that actually needs judgement."
"I can show a buyer the production site and the docs page that explains how it got built. There is no demo gap — the deck and the repository tell the same story."
"Every agent action is a Linear state transition, a labelled branch, an Actions log line, and a PR. The chain of custody is the existing GitHub + Linear chain — nothing new to audit."
Closed beta
Ready to make your process legible?
Ship is onboarding founder workspaces by invite, cohort by cohort. North star: from sign-up to your first closed ticket in one day.