Pre-flight checklist — what to do before you turn Cowork on

TL;DR. Before anyone clicks Enable Cowork, there are nine small decisions that take half an hour to make and save weeks of mess later. This is the page you print, tick, and sign off. We run this checklist in the first 60 minutes of every Tinkso engagement.

The nine pre-flight decisions#

Each item takes a couple of minutes to decide. The rationale matters more than the wording — if you're going to deviate, deviate knowingly.

• Sponsor named. Someone with budget authority and four hours a week for the next six. Without a sponsor, deployments stall in week three.
• Pilot function picked. Not "the company" — one function. Finance close, marketing proposals, supplier comparison. Pick the function with the most repetitive document work and a clear artifact at the end of every week.
• Pilot owner named. A different person from the sponsor; the operator who will be hands-on inside Cowork every day. The sponsor unblocks; the owner ships.
• Workspace folder agreed. A single path on the laptop or in a synced shared drive. Not Desktop. Not Documents root. One folder per function, with a structure (see The workspace pattern).
• Plan tier chosen for the pilot. Default: Pro for most users, Max for the operator who runs Cowork all day. Re-evaluate at week six once you have actual usage data.
• Security review status known. Either approved, in-progress, or N/A. If review is in-progress, run the pilot on non-confidential data only. Do not let the pilot wait on procurement; do not pretend it's done either.
• Data classification rule set. A three-tier rule: what data is allowed in Cowork (e.g., public, internal, restricted-no). Two pages would be overkill; a one-pager that names the three tiers is exactly enough.
• Rollback plan documented. What does "we turn this off" look like? Mostly: revoke folder access, uninstall the app, archive the workspace folder. Even if the path is obvious, write it down — the act of writing it surfaces edge cases.
• Success metrics agreed. Two metrics. No more. We cover the choice in Measuring ROI.

If you cannot tick all nine after one meeting, the gap is information, not energy — find out what you don't know, then come back.

The 30-minute meeting agenda#

A copy-paste agenda for the single kickoff meeting that decides all nine items. Attendees: sponsor, pilot owner, IT lead, optional security partner.

• 0–5 min — what is Cowork. The pilot owner demos one thing. Not a sales walkthrough; one workflow they will actually run.
• 5–15 min — sponsor and owner confirmed; pilot function and metrics agreed. Two metrics, named.
• 15–25 min — workspace folder, data classification, plan tier. Pick a folder, agree the three tiers, default to Pro / Max.
• 25–30 min — rollback and security review status. Document the off-switch; record where review stands.

End the meeting with a slack message to the team announcing the pilot and the named owner. Public commitment shortens week one.

What IT needs to know#

• Cowork installs as a desktop app on macOS or Windows. It is not browser-based.
• It only sees the folder you grant. There is no system-wide access; no "now Claude can read your whole drive" surprise.
• It does not require admin or root install for the user. It does require IT-level decisions on data classification.
• It runs under the same Claude account as Chat and Code; for organisations, SSO via the Team or Enterprise plans.
• VPN incompatibility is a known issue — see VPN & network. Test on the production VPN before pilot day.

What security needs to know#

• Data is processed on Anthropic infrastructure under the active subscription's data terms.
• Default training opt-out is configurable in the account; we set it explicitly during setup.
• Files Cowork creates or edits stay local. Cowork does not push files anywhere unless an MCP connector is configured to do so.
• For deeper detail send them to the Security model and Data residency & retention pages — they are written in language a security reviewer will recognise.

What HR or Comms might need#

A Cowork rollout deserves a small internal note: "we are piloting an AI tool that helps with file work in Finance, owned by [name], for the next six weeks." Avoid overselling. Avoid the word transformation. The point is to set expectations, not generate excitement that pilot results then have to live up to.

Common pre-flight mistakes#

• Skipping the sponsor decision because "it's just a tool." Cowork edits real files. Treat it like a system, not a Chrome extension.
• Granting Cowork access to the entire user home directory. Always a single workspace folder. Always.
• Picking a pilot function with no measurable artifact. "General productivity" is not measurable. "Number of weekly invoices processed" is.
• Trying to roll out to four functions at once. One pilot, then scale. The temptation to broaden is the surest way to dilute every pilot to inconclusive.

Tinkso's take#

We run this checklist in the first 60 minutes of every engagement. The checklist plus a shadow session is what we call Sprint 0 — the Observe beat of Observe → Co-design → Ship.

Mid-market teams sometimes resist the "this is a real rollout" framing because the tool feels lightweight. The lightweight tool is exactly why governance has to be set up early — Cowork is usually the team's first agentic AI deployment, and the patterns set in week one stick for the next year.

Try this#

Print the nine-decision checklist. Tape it to a wall. Cross items off as you go. Do not enable Cowork until they are all crossed.