"Five days" is short enough to sound like a marketing claim. The reason it works is mechanical: we built the sprint backwards from five concrete outputs the team takes home. Not five lessons. Five things they can use on Monday morning.
Every output exists because operating teams asked for it twice. The shape is locked. The content is custom. Below is the day-by-day, the deliverable, and how to verify each one is real before we leave the building.
01The five outputs
Before the day-by-day, the deliverable list. Each item below is a thing the team owns when we walk out: their repo, their account, their stack. We do not host anything we do not have to.
- Shared skills. A library of reusable AI workflows, scoped to the team's actual jobs. Not "prompt templates." Actual versioned skills with triggers, instructions, and acceptance tests. Lives in a git repo the team controls.
- Skills hosting. The skills run somewhere. By default a Claude Projects setup or a self-hosted Claude Code config. The team picks one. We wire it.
- Tool creation and hosting. Two to four custom internal tools, written and shipped during the sprint. Examples: a client-report generator, a Shopify section builder, a Klaviyo flow auditor. Hosted on the team's existing infra.
- Secure MCPs. One or two Model Context Protocol servers wired to the team's actual data sources (Shopify Admin, Klaviyo, GA4, internal Postgres). OAuth done right. No keys in plain text.
- Shared knowledge base. A living document layer that captures every decision, exception, and customer fact. The thing that keeps the new hire from asking the same question twice.
02Day-by-day
The sprint runs Monday to Friday. 4 to 5 hours of pair-building per day, plus async homework. We aim for "useful by Tuesday afternoon" and "shippable by Friday morning." Day 5 is hand-off, not building.
Audit and shock
We sit with the team and watch them work. Not interview. Watch. Inside 90 minutes we surface five workflows where AI plus the right plumbing collapses the work by a factor of 5 to 20. Then we do one of them live, end to end, while they watch. That is the "wait, what?" moment that breaks the typewriter mindset.
First skill, first MCP
We pick the heaviest-payload workflow from Day 1 and rebuild it as a versioned skill. Then we wire one MCP server to the data source it needs. The team does the typing. We do the architecture. By 4pm the team has shipped one custom tool that runs against their real data. They built it. They can read every line.
Stack the tools
Now we go wide. Two to three more internal tools, built with the team. The patterns from Day 2 repeat, faster. By end of day we have a small library of skills, MCPs talking to two more data sources, and the first tool the team built without us touching the keyboard. That is the inflection point.
Knowledge base, hosting, security pass
Half the day is the shared knowledge base, set up so every skill, decision, and customer fact has a home. The other half is the security pass: token rotation, OAuth scopes, secret storage, who-has-what. We sign off on a deploy plan the team can run without us. No staging webhook left flapping.
Hand-off and bury the agency
We do not build on Day 5. The team walks us through every artifact: skills, MCPs, tools, knowledge base. They explain what each one does and how to extend it. If they cannot, we go back and fix the gap. End of day, we sign off and leave. No retainer, no Slack channel, no monthly call. The team owns it.
Median across 6 sprints, measured 30 days post hand-off. Hours saved on workflows that existed before the sprint. New work the team picked up after the sprint is excluded from this number.
03What we do not ship
The list is shorter than the deliverable list, and the omissions are deliberate.
- No retainer. If you need ongoing help, we recommend the team ship more sprints with you, not us.
- No Slack channel after Day 5. The hand-off is the hand-off. If something breaks in week six, the team fixes it. We left the runbook.
- No black-box automation. If you cannot read the code, we will not ship it. Every skill, every MCP, every tool is in your repo, in plain English, with comments.
- No proprietary platform. We do not host your skills on a "Wedge platform." That would make us your dependency, which is exactly the trap we pulled you out of.
04How to verify it actually shipped
The end-of-sprint receipts doc lists every deliverable with a verification step. Not "we built a thing." A check the team runs themselves. Examples from a recent sprint:
- Run skill
generate-monday-reportfor client X. Confirm Drive folder receives the deck inside 90 seconds. - Hit MCP endpoint
shopify-admin-productswith curl. Confirm 200 response and product list returned. - Open knowledge base. Search "refund policy exception, Brand Y." Confirm the August 2025 decision shows up first.
If a check fails, we fix it before we leave. The point is the team has receipts they can run without us, on day 6, and on day 60.
We walk in. We rewire. We walk out. Wedge field notes
05What it costs
Two prices. €700 for the first three clients (intro pricing, two slots left as of this note). €1,200 standard after that. Pricing is locked through 2026 unless something breaks.
If the math on €700 versus the agency invoice on your desk is uncomfortable, that is on purpose. The point of the sprint is to make the agency invoice optional. Read Your team uses AI like a typewriter for the why, then book a sprint when the why lands.