The problem this solves
You run a product. Your team has been told to "use AI more." So tickets get a little better, a few pull requests show up unprompted, and somebody on Slack keeps demoing a magic chat window. By the end of the quarter you can't answer two simple questions: which of these changes did we actually decide to make, and who owns the next one? The trail evaporates the moment the demo ends.
Ship is the workspace we built so the trail stops evaporating. It is where a founder or product owner directs AI-assisted delivery the same way they'd direct a small team — with priorities, written context, named approvals, and a record that survives the meeting it came out of.
What you get on day one
When you open Ship for the first time you'll see one screen, not ten — the workspace hub. It answers three questions: is the engine healthy, is anything waiting on your decision (a short "Waiting on you" strip, each item deep-linking to an approval page), and how do you connect your agent. That last card is the important one: Ship is driven from your own AI agent — Claude Code or Claude Desktop — connected to Ship over MCP. You run claude mcp add ship <url> -t http, log in through the browser, approve the workspace, and from then on you direct delivery by talking to your agent, which calls Ship's tools (create tickets and projects, kick off decomposition, review a PR, drain the inbox). The console stays deliberately thin: a hub, an approvals page, Settings, and a fallback chat for when you don't have an agent in front of you. Domain detail (tickets, projects, PRs, CI) lives in Linear and GitHub, where it always did.
Behind that home screen, Ship binds to the tools your team already uses. You point it at your code (GitHub, today), at your tracker (Linear, Jira, GitHub Issues, GitLab, or Azure DevOps), and at the channel where digests should land (Slack or Teams). Ship doesn't replace any of those. It sits in the middle and gives the AI work somewhere accountable to live.
What changes about your week
Three things change, and they're the ones you'll notice on Monday morning.
First, the spot where decisions used to scatter — chat threads, DM, the comment field on a ticket nobody re-reads — becomes a single inbox of decisions: clarifications, approvals, failures. Three to five on a normal day. You work them through your agent ("what's waiting on me?"), or for the ones that need a deliberate human signoff, through the approval page the hub links you to. You read, you decide, you move on. The system waits.
Second, the work that goes to AI carries written intent with it. When you talk to Ship about a new initiative, the conversation lands as a project description your team can read tomorrow. Tickets stay short and link to it. The work that gets done isn't done from a vibe — it's done from a sentence somebody wrote and somebody approved.
Third, every action AI takes leaves a trail. Which ticket it touched, which knowledge article it cited, which check it waited on, which human signed off. Six months later you can answer the question that usually has no answer: why did this change land?
What stays human
The fences are wide on purpose. AI in Ship does not decide what matters, does not invent priorities, does not merge code without your team. It opens pull requests, asks clarifying questions, drafts plans, and runs the checks you defined. The judgment calls — what to ship, what to park, what's worth the risk — stay with the operator who carries the outcome.
That's the whole pitch. Velocity you can explain. Movement you can defend. A workspace that goes quiet when it should and speaks up when it must. The next page is the eight words you'll see everywhere; after that we walk through a normal day.