The AI Coworker Readiness Checklist

Picture a ten-person agency on a Tuesday afternoon. The ops lead signs up for an AI platform, connects Gmail and HubSpot in twenty minutes, and posts in Slack: "We have an AI coworker now, use it for anything." By Friday, three people have tried it once. By the following Friday, nobody has touched it. The tool was fine. The team was not ready, and nobody had run an AI readiness checklist before spending the money.

This happens constantly, and the postmortem is almost always the same. The process the AI was supposed to run existed only in one person's head. The CRM it was supposed to read was full of duplicate contacts and dead deals. Nobody owned the rollout, so nobody answered questions in week one. And nobody reviewed the output, so the first mediocre draft became the story everyone told about "the AI thing that didn't work."

None of those are AI problems. They are operational problems that an AI coworker exposes faster than a human hire would, because a human quietly works around them and an AI does not.

This guide is the checklist I wish more teams ran before signup. It covers four pillars: documented processes, clean tool access, a named owner, and a review habit. Score yourself honestly at the end. A low score does not mean "don't do it." It means "here is exactly what to fix first, and it will take you two weeks, not two quarters."

Why an AI Readiness Checklist Beats Enthusiasm

The failure rate of AI coworker rollouts has very little to do with model quality and a lot to do with the ground the tool lands on. We wrote a whole piece on why AI employees fail, and the short version is: they fail for the same reasons a contractor hired on day one with no onboarding fails. Vague instructions, locked doors, no manager, no feedback.

Think about what you would do before hiring a junior operations person. You would write down what the job actually is. You would get them accounts on the systems they need. You would assign them a manager. You would review their first month of work closely. Nobody skips those steps for a human hire and expects it to go well.

Teams skip all four for AI because the signup flow takes five minutes and the marketing implies the tool figures everything out. It does not. An AI coworker is leverage on top of your existing operations. If the operations are undocumented, inconsistent, and unowned, you get leverage on chaos.

The checklist below exists to make that visible before you spend money and, more importantly, before you spend your team's willingness to try. You usually get one clean shot at introducing an AI coworker. If the first attempt lands badly, the second attempt fights skepticism the whole way.

Pillar 1: Documented Processes

The single strongest predictor of a good rollout is whether the work you want to hand over is written down anywhere.

Not documented to ISO standards. Documented at the level of "a smart new hire could follow this without asking questions." For a weekly pipeline report, that means: which HubSpot pipeline, which deal stages count as active, what "stalled" means in days, who receives it, and what format they expect. For inbox triage, it means: what counts as urgent, which senders always get flagged, what never gets touched.

Here is the honest test. Pick the first task you want an AI coworker to do. Now write the instructions as if you were emailing them to a temp who starts tomorrow. If you can do it in fifteen minutes, that process is ready. If you keep writing "it depends" and "ask Sarah about this part," it is not, and the AI will hit exactly the same walls the temp would.

Common failure modes at this pillar:

• The process lives in one person's head. When that person is out, the process stops. An AI cannot read minds any better than a new hire can. This is also a business risk on its own, which we cover in the AI bus factor.
• Three people do the same task three different ways. The AI will pick one way, and two people will call it wrong. Agree on one way first.
• The process changes weekly but the documentation does not. Stale instructions are worse than none, because the output looks confident and is confidently outdated.

You do not need to document everything. You need to document the two or three tasks you are actually delegating first. If you want a structure for this, an AI runbook is the format that works: task, trigger, steps, inputs, outputs, edge cases, escalation.

One more honest point: writing tasks down well is a skill in itself, and it transfers directly to how well the AI performs. There is a real craft to writing tasks an AI nails, and teams that invest an hour in it get dramatically better output than teams that type one vague sentence and hope.

Pillar 2: Clean Tool Access

An AI coworker is only as useful as what it can see. If it can read your HubSpot, Gmail, Stripe, and Jira, it can answer "which deals went quiet after we sent the proposal" in one question. If it can see none of those, it is a chatbot with opinions.

Clean tool access breaks into three questions.

First: do you know which tools matter? Most teams run fifteen to thirty SaaS products, but three to five of them carry the real signal. For a services business it is usually the CRM, email, the project tracker, and the invoicing tool. For ecommerce it is Shopify, Stripe, the support inbox, and the ad platforms. Connect the tools where work actually lives, not everything with an OAuth button. We wrote a practical guide on which integrations to connect first if you want the reasoning task by task.

Second: can you actually grant access? This is where rollouts stall for a week in the most boring way possible. Who has admin on the HubSpot account? Is the Stripe account owned by a founder who left? Does IT need to approve OAuth grants? Find out before day one, not during it. A rollout that stalls on permissions in week one loses momentum it never recovers.

Third: is the data inside those tools trustworthy? This is the uncomfortable one. If your CRM has 400 deals and 250 of them are zombies nobody closed out, every pipeline answer the AI gives will be technically correct and practically useless. The AI does not know deal #2841 is dead unless something in the data says so. You do not need pristine data. You need the fields you will actually query to mean what they say: deal stages that reflect reality, an inbox where the important folders are the important folders, a Jira board where "In Progress" means in progress.

A practical note on how this looks in a real product: when a team connects tools to Skopx, every answer cites the specific records it drew from, which turns out to be a data-quality audit in disguise. Ask "what is stuck in the pipeline," click through the citations, and you find out in minutes whether your CRM reflects reality. Teams often spend their first week doing exactly this: asking questions, following citations, and cleaning what they find.

Pillar 3: A Named Owner

Every AI coworker rollout that works has one person's name on it. Every one that quietly dies was "a team thing."

The owner is not necessarily the most senior person or the most technical one. The best owners are the people closest to the operational pain: the ops manager who builds the reports, the chief of staff who chases status updates, the founder in a five-person company because there is nobody else. What matters is that they have three things: enough process knowledge to write good instructions, enough authority to get tool access granted, and enough calendar space to spend two to three hours a week on it for the first month.

What the owner actually does:

• Picks the first two or three tasks and writes them down (Pillar 1).
• Gets the integrations connected and unblocks permissions (Pillar 2).
• Reviews output daily in week one, then on a declining schedule (Pillar 4).
• Answers "how do I ask it for X" questions from teammates so early friction does not become early abandonment.
• Kills tasks that are not working instead of letting them limp along and poison trust.

If you cannot name this person right now, you are not ready, and no tool choice fixes that. The full argument for how to pick, and the failure modes of picking wrong, is in who should manage the AI. And once you are past the first month, the ongoing job looks a lot like managing a capable junior teammate, which we cover in managing AI like a team member.

One caution from watching many rollouts: do not assign ownership to someone as a side quest they did not ask for. Reluctant owners produce dead rollouts. Pick someone who is annoyed enough by the manual work to want this to succeed.

Pillar 4: A Review Habit

The last pillar is the one teams most want to skip: someone has to actually read the output, especially early.

Not forever, and not everything. But in the first two weeks, every report, every draft, every summary should get human eyes before anyone acts on it. This is not because the AI is unusually error-prone. It is because the AI does not yet know your context: which client is sensitive right now, which metric your CEO actually watches, which Jira project is a graveyard. Review is how that context gets transferred, one correction at a time.

A review habit that works in practice:

• Week one: review everything, same day. The owner reads each output, notes what is wrong or off-tone, and refines the task instructions. Expect to edit instructions three or four times. That is the process working, not failing.
• Weeks two to four: review outputs that leave the building. Internal summaries can go direct. Anything client-facing or money-adjacent still gets a human pass.
• Month two onward: spot-check on a schedule. Weekly for high-stakes outputs, monthly for routine ones. Never zero, because processes drift and data drifts.

The graduation question, when to let something run without a human look, deserves its own decision framework, and we wrote one: when to let AI act without review. The short version is that review earns its way down, task by task, based on demonstrated accuracy, not on vibes or week counts.

Tooling can make the habit cheaper. In Skopx, actions inside your tools happen on your instruction with your approval, and monitoring follow-ups are approval-gated, so the review step is built into the mechanics rather than depending on someone remembering. A morning briefing that reports what moved across your tools also gives the owner a natural daily checkpoint instead of one more thing to remember. But no tool replaces the habit itself: someone with judgment, actually reading, at a declared cadence.

The AI Readiness Checklist: Score Yourself

Ten questions, one point each. Answer honestly. The failure mode of this exercise is optimism.

1. Can you name the first three tasks you would delegate, specifically enough that a temp could start on them tomorrow?
2. Is at least one of those tasks written down, end to end, including edge cases and who gets the output?
3. Do the people who do these tasks today agree on how they should be done?
4. Do you know which three to five tools hold the data these tasks need?
5. Can you grant access to those tools this week, without a permissions scavenger hunt?
6. Would you trust the data in those tools to answer a question your CEO asks? (Deal stages current, inbox organized, board statuses true.)
7. Can you name one specific person who owns this rollout?
8. Does that person have two to three hours a week for the first month?
9. Is there a declared plan for who reviews output in the first two weeks, at what cadence?
10. Has your team agreed on what success looks like at 30 days? (Hours saved, a report that ships on time every week, an inbox triaged by 9am. Something checkable.)

Count your points, then find your band below.

What Your Score Means

The reason the bands are ordered this way: ownership unblocks documentation, documentation unblocks useful tool access decisions, and all three make review meaningful. Fixing them out of order wastes effort. An owner with no documented process has nothing to delegate; pristine documentation with no owner sits unread.

Notice what is not on the checklist: your team's technical skill, your industry, your company size. A three-person bookkeeping firm with tight processes scores higher than a fifty-person startup with tribal knowledge, and the bookkeeping firm will get more from an AI coworker, faster. Readiness is operational, not technical.

If You Scored Low: The Two-Week Fix

A score of 3 to 6 is the most common result, and the gap is closable in two weeks of part-time effort. Here is the sequence that works.

Days 1-2: name the owner and the first task. One person, one task. The best first task is boring, frequent, and rule-based: the weekly pipeline summary, the Monday metrics email, triaging the support inbox into categories. Do not pick the hardest problem in the company.

Days 3-5: document that one task. Write it as an email to a temp. Show the draft to whoever does the task today and let them tear it apart. The arguments this surfaces ("wait, you exclude renewals from that number?") are the whole point.

Days 6-8: audit tool access for that task. List the tools the task touches. Confirm who can grant access. Open the CRM or tracker and fix the worst data problems in the specific fields the task reads. Not a full cleanup, a targeted one.

Days 9-10: declare the review plan. Owner reads every output in week one, same day. Client-facing output gets a second reader. Write down the 30-day success check.

Then start. Your first week has its own rhythm and its own traps, and we mapped it hour by hour in the first week with an AI coworker. The teams that follow a sequence like this hit their 30-day check. The teams that skip to signup usually spend the same two weeks anyway, just after the enthusiasm has drained.

On tooling cost, since it shapes how much ceremony the decision deserves: this is not an enterprise procurement exercise anymore. Skopx's Team plan is $16 per seat per month with 2.3 million AI tokens included per seat, and a Solo plan runs $5 a month with your own API key at provider rates, zero markup either way. At that price the readiness work is the expensive part, which is exactly why it deserves the two weeks.

FAQ: AI Readiness Checklist Questions

How long does it take to become ready?

For most small teams, two focused weeks, part-time, following the sequence above. The work is naming an owner, documenting two or three tasks, auditing access to a handful of tools, and declaring a review cadence. Teams that report months of preparation are usually trying to document everything or clean all their data, neither of which the first tasks require.

Do we need to document every process first?

No, and trying to is the most common way readiness projects die. Document only the two or three tasks you are delegating first, at temp-could-follow-it depth. Each new task you delegate later gets documented when you delegate it. The AI rollout becomes the forcing function for documentation you should have had anyway, one task at a time.

What if our data is messy?

Almost everyone's is, so do a targeted cleanup, not a general one. Identify the specific fields your first tasks read (deal stages, ticket statuses, invoice states) and make those truthful. Leave the rest. Cited answers help here: when every answer links to the records behind it, you find the messy spots by using the system, which is faster than auditing tables in the abstract.

Who should own the AI coworker if we have no ops person?

The person most annoyed by the manual work, provided they can write clearly and get access granted. In companies under ten people this is often a founder, and that is fine for month one. What matters is that ownership is explicit and singular. "Everyone will use it" is how tools die. Once it works, ownership can widen; there is a right way to share an AI coworker across a team without recreating the nobody-owns-it problem.

Is a small team ever really ready?

Small teams are often more ready than large ones. Fewer tools, fewer conflicting versions of each process, and the owner can be the decision-maker. What small teams lack is slack time, which is why the checklist emphasizes picking boring, frequent tasks first: they pay back the setup hours fastest and fund the patience for everything after.

Do we need IT or a developer to get ready?

Not for readiness itself. The four pillars are operational: writing, permissions, ownership, review. Modern platforms connect to tools like Gmail, HubSpot, and Jira through standard OAuth grants, and workflows are described in plain language rather than code. Where you may need IT is permissions: if your company routes OAuth approvals through an admin, involve them in week one so access does not become the bottleneck.

Score It, Then Start Small

The AI readiness checklist comes down to four questions a good manager would ask about any new hire. Is the job written down? Can they access what they need? Who do they report to? Who checks their work?

Score yourself tonight, honestly. If you are at 8 or above, start this week with your most boring, most frequent task. If you are below that, you now have a two-week plan and a specific list of gaps, and closing them improves your operations whether or not you ever add an AI coworker. That is the tell of a good readiness exercise: the preparation is valuable even if you stop there. Most teams that do the preparation do not stop there.