Invite-only.
How Dock is different

Four shapes Dock isn't.

Every founder asks within thirty seconds of the home page: "wait, how is this different from what I already use?". Below: the four patterns Dock is rejecting and what we ship instead.

Not Dock

The chatbot

You message the agent, it messages back. Each turn starts cold. The agent has no addressable workspace, no version history, no audit trail you can hand to the next agent.

Doesn't carry state between sessions. Doesn't carry handoffs between agents. Doesn't survive the team scaling past one human + one assistant.

↓ Dock

Shared workspace

Tables and docs read + written by humans and agents alike. The artifact lives. Every state-changing action lands in the audit log named to the actor. Conversation, if any, is optional.

Read more →
Not Dock

AI bolted onto an existing doc

Your existing collaboration tool with an inline AI assistant. The agent can suggest edits but doesn't have its own identity, its own permissions, or its own audit trail.

When the agent acts, the log says the human did it. When it goes rogue, the human owns the bill. Can't run a fleet because there's no per-agent accountability.

↓ Dock

Agents as first-class identities

Every agent has its own API key, its own scopes, and its own row in the audit log. Dangerous operations require a two-key handshake. Authorship and accountability never blur.

Read more →
Not Dock

The stateless agent

Frontier-model + tool-use, but no persistent state. The agent has tools but no place. Each invocation is a fresh start. Agent-to-agent handoff is impossible because there's nothing to hand off.

Doesn't compose. Two agents working on adjacent tasks have no shared substrate to coordinate through. Plural work falls back to copy-pasting between chat windows.

↓ Dock

Workspace as substrate

Workspace state IS the handoff. Agent A writes a row, Agent B sees it, picks it up, marks it done. The next human reading the workspace sees the entire trail.

Read more →
Not Dock

The single-player tool

An agent for me, in my IDE, on my machine. Great for solo speed; terrible the moment work crosses a team boundary. The team has to invent their own glue around it.

Doesn't multiply when the team grows. Each member runs their own assistant; outputs are copy-pasted into the actual collaboration surface, where the AI provenance is lost.

↓ Dock

Plural-by-default

Multi-author workspaces from the first row. Humans + agents share state, including the agents you've already deployed elsewhere — they connect over MCP and read/write the same workspace.

Read more →
What that looks like in practice

One workspace. Humans editing in the UI. Agents calling MCP. Everyone on the same row.

An agent rewrites a row. The version history names the agent. Another agent picks up the next item in the queue. A human reviews both edits and approves. Nothing leaves the workspace. No screenshots flying through chat. No "who wrote this" guesses.

That's the shape. If you want the full thesis, read the manifesto. If you want the engineering, read the blog.