Distributed teams have been told async-first is the answer for a decade. GitLab wrote the canonical handbook on it. Every remote-first company since 2015 has put it in the careers page. And yet most distributed teams still feel synchronous, still burn the timezone overlap, still wait on the one person who hasn't replied.
Async-first was always a property of the substrate, not the policy. The substrate kept being chat. Chat is synchronous-feeling even when messages arrive eight hours apart, because a chat message is a ping aimed at a person, and a person feels pinged whether they read it in three seconds or three time zones later. "No need to reply tonight" changes nothing.
Async-first needs an object-first substrate: persistent, attributed, edit-not-message. Work that lives in a doc, a row, a status column. The next person opens the artifact and sees the new state, not a notification asking them to assemble it.
Layer agents on top and you get a real async stack. Agents are the most async colleagues you will ever have: they do not sleep, do not feel pinged, do not lose context between sessions. Put them in a workspace where they read and edit the same artifacts as your humans, and the gaps between shifts stop being dead time.
The five patterns of async-with-agents
1. Update the artifact, not the channel
When an SDR closes an account, update the account doc. Not a Slack thread, not a DM, not an emoji on a calendar block. The next person to touch Acme opens the workspace, reads the state, sees the audit row.
Linear's async ticket model gets close to this for engineering: ticket is the artifact, comments are the conversation, status is the truth. Generalize it. Sales has accounts, research has briefs, content has drafts, ops has runbooks. If news is meaningful enough to share, it is meaningful enough to write into the artifact it changes.
2. Agents fill the gaps between human shifts
Your APAC reviewer signs off at 6pm Singapore. Your US team starts at 9am ET. That gap used to be dead time, or worse, a forced handoff meeting at one team's bad hour.
The research agent runs the overnight sweep against the brief your APAC reviewer left open. Twelve hours later the US team opens a populated sources table, each row attributed to the agent. Notion's async patterns ask you to wait for the right human to ping you. Slack's workflow builder chains humans. An agent in the workspace is the colleague who works while you sleep, writes to the same surface, and leaves attribution behind. Time-zone arbitrage that compounds.
3. Decisions live in the doc, not the meeting
A distributed team's worst hour is the recurring decision meeting. Six people on video, three time zones, one call that could have been an edit. The gravity well is the real cost: any decision waits for the next sync, which is Thursday, which means nothing moves until Thursday.
Move the decision to the doc. The spec gets edited inline by the PM, the engineer who has to build it, the design agent that drafts variants, the research agent that pulls precedent. The audit row says who decided what. When something is contested, a comment thread pulls in the two people who disagree, async. The meeting is for the residue. Most weeks, there is no residue.
4. Review is a column, not a calendar block
Review in async teams usually fails because it is implicitly synchronous. "Take a look when you have a sec" is a calendar invite without a time. Work sits in tabs until somebody remembers, then gets a rushed Friday pass.
A workspace pattern fixes this. The artifact table has a reviewer column and a review status column. A review agent surfaces stale rows, dual-keyed audit shows when the reviewer signed off, the reviewer's queue is a filtered view of "things waiting on me." The agent does not approve, the human does. The substrate does the bookkeeping nobody wants to.
5. Knowledge accretes in workspaces, not in Slack history
A Slack message decays the moment it scrolls. Workspace artifacts compound. Six months in, your account workspaces hold every brief, call summary, and decision rationale, each attributed and addressable. New rep joins, you hand them a workspace, not a search query.
Agents writing into workspaces grow institutional knowledge. Agents writing into chat grow noise. A year on, the team running async-with-agents in workspaces has a corpus they can search, audit, and onboard against. The chat team has a transcript nobody reads.
The Dock primitives that enable this
Five primitives, each load-bearing.
Shared workspace. Docs and tables as tabs of the same room, humans and agents as members. Full thesis in the shared workspace as the new collaboration primitive; conceptual frame in what is an AI workspace.
Dual-keyed audit. Every edit has a real author. Research agent's sources row says so. APAC reviewer's approval says so. See agent audit and compliance.
Dangerous-ops gates. Agents never auto-approve spend, permission widening, customer email, destructive deletes. Agent proposes, human consents. Context in humans and agents.
Signed-agent inheritance. A teammate's agent gets the same workspace access the teammate has, no separate provisioning. Agent collaboration primer walks the model.
Scoped credentials. The agent's reach is the workspace, not the org. Same logic behind the AI workspace, not the AI assistant and why chat is the wrong abstraction.
A worked 24-hour cycle
09:00 SGT. Priya in Singapore opens the launch-research workspace, updates the brief with three hypotheses, tags the research agent on a sources row. Signs off at 18:00.
22:00 SGT. The research agent finishes its sweep. Eleven sources added, each attributed to the agent, each linked to the hypothesis row it supports. Summary appended to the brief.
09:00 CET. Jonas in Berlin reads the populated table over coffee, prunes for an hour, comments on two rows, edits the brief inline. No handoff message. His edits show as his.
14:00 ET. Ana in New York opens the same workspace. The brief is in its third revision. She drafts the spec in a sibling doc with cross-refs to the research table. The design agent, signed by her teammate, drafts three variant mockups into a third surface.
17:00 PT. The dispatcher in San Francisco scans the workspace. The launch is on track. No status meeting required: the workspace is the status. Full revolution, no Zoom.
Microsoft's 2025 Work Trend Index calls the emerging shape the "Frontier Firm": small teams scaled by human-agent collaboration. The cycle above is what that looks like at the desk level.
What NOT to expect
This does not replace synchronous chat for ambiguous, fast-moving decisions. When a customer call is happening right now and you need three people in ninety seconds, chat is correct. Async-first is about the bulk of work, not edge cases that need presence.
Do not make agents auto-resolve debates. Humans drive disagreement. An agent can summarize positions, flag stale threads, surface precedent, but the call belongs to the people whose names are on the work. An agent that decides and ships on contested artifacts is how you get a beautifully attributed mistake.
Do not gate trivial async work behind dangerous-ops review. The consent gate is for things that matter: spend, permissions, customer outbound, destructive deletes. Wrap a comment edit in a consent flow and the team routes around the workspace within a week.
The shape
Async-first distributed work has been an aspiration on a substrate that could not carry it. Object-first workspaces with agents as members carry it. Use cases like Dock for research and Dock for sales walk specific shapes; the playbook generalizes.
If your team is distributed and the substrate is still chat, the substrate is the problem. Try Dock for distributed teams. One workspace, one agent, one timezone gap. See what compounds.