A 7-day AI employee sprint

What is a 7-day AI employee sprint?

It is a week with a finish line. Not a discovery workshop that produces a deck. Not a pilot that lives in a sandbox until someone loses the password. You point at work that still happens by hand. I take that work through pick, map, build, and handover. On day 7 it is running in the same tools your team already opens.

I use this shape because companies do not fail at AI for lack of models. They fail because nobody owns a single process end to end. Seven days forces a choice. One pile. One owner. One path for the cases a person still has to see. If you cannot name those things, you are not ready to build, and a longer calendar will not fix that.

The jobs I take on are boring on purpose. The same support questions. The first pass on a stack of CVs. Follow-up notes that slip. Fields copied between two systems. If your team already knows the pile, that is the job. If they do not, we are not starting a sprint. We are still arguing about what work is.

What happens on day 1?

Day 1 is the pick. We sit with the person who still does the work, not only the person who approved the week. I want the hours, the tools, the volume, and the cost of a miss. I want the cases they already escalate without calling it escalation. I want the sentence they use when they dump a ticket on a colleague.

A good first process is frequent, narrow, and already written down in someone's head. A bad first process is the one that would impress a board. Strategy work, brand-new products, and anything that needs a six-month data cleanup are not week-one jobs. I have said no on day 1. That is cheaper than a pretty build nobody trusts.

• What is the pile, in hours per week?
• Which tools already hold the truth?
• What does a miss cost in money, time, or a customer?
• Who still owns the weird cases after we ship?

What do you map on days 2 and 3?

Days 2 and 3 are the map. Every step. Every field. Every tool. Every exit. If a person currently opens a ticket, searches a doc, writes a reply, and sometimes pings a lead, that is four steps, not one. If they skip the search when they already know the answer, that is a branch, not a vibe.

The map also names the cases a person still owns. Refunds over a line. Legal language. A customer who is already angry. A candidate who is a referral. Those are not failures of the AI employee. They are the job you are buying: take the first pass, leave the judgment. If you skip this, day 6 becomes a fight about whether the model is 'smart enough.' It is not a smartness problem. It is an unmapped exception.

I write the map so a stranger can follow it. That is the same habit I used when a clinic could not open a loan path on a Monday. Pretty diagrams do not help at 2am. A sequence with owners does.

What gets built on days 4 to 6?

Days 4 to 6 are the build. I connect the workflow to the tools you already pay for. Help desk. CRM. Inbox. Spreadsheet you will not admit is a system of record. Calendar. The HR pile. No new login if we can help it. If the work needs a small knowledge store, it is a store of your documents, not a second wiki your team will ignore.

This is also when testing starts, not when the UI looks done. We run real tickets, real CVs, real follow-ups through the path. We score misses. We tighten the cases that should never auto-close. We wire the escalation so a person gets a working state and a decision brief, not a raw transcript they have to re-read.

If a connection is blocked by a missing API key or a person who 'will get to permissions next week,' the sprint pauses. I will not pretend a screenshot is a live employee. The week is short because the work is scoped, not because we skip the unglamorous parts.

What does handover look like on day 7?

Day 7 is live, with a write-up and a call. The write-up is short: what it does, what it must not do, where the logs are, who to ping, and how a case comes back after a person decides. If your team will not open a 20-page playbook, I will not write one. I write the page they will actually open when something looks wrong.

The call is not a product tour. We walk one happy path and two ugly ones. We show the monitor. We agree who watches it in week two, because week two is when the cases you did not map show up. The $7,500 week includes 30 days of support for that reason. Shipping on Friday and disappearing on Monday is how these things die.

What does the week cost, and when is it the wrong shape?

One workflow is $5,000: pick, one live process, tools you already have, testing, a write-up, and a handover call. The week is $7,500: that, plus a plan for what to automate next, a knowledge store if the work needs one, proper API connections, monitoring, and 30 days of support. Several workflows start at $15,000 when you already know this is more than one process and you want someone on the architecture.

A week is the wrong shape when you want a platform, a new chatbot brand, or a replacement for a team you have not defined. It is also wrong when nobody can name the hours. I would rather refund the conversation than take money for a sprint that will produce a demo. The point is to take work off people. If we cannot point at the work, we are not building yet.