A creative director opened a Slack DM at 11pm. The team had spent three weeks on a Klaviyo flow split. They had Claude open the whole time. The output was a deck. The fix was 40 lines of code and an MCP server. They never reached for it because they did not know it existed.
This is not a story about a smart team failing. The team was sharp. The director was sharper. The agency had sold the work, scoped the work, and shipped a version of the work. The reason the wedge between effort and outcome stayed wide is mechanical, and it repeats in every team we walk into.
The team was using AI like a typewriter. Faster than handwriting. Slower than what was possible by a factor of about twenty.
01The 5% everyone hits
When a team adopts Claude or ChatGPT or Cursor, the first month looks identical across companies. People paste in their drafts and ask for tighter copy. They paste in a brief and ask for an outline. They generate slide titles, rough out a Loom script, clean a meeting summary, brainstorm subject lines.
Every one of those tasks ends in a text box. The AI is a faster typewriter. Output goes into a Google Doc, a Notion page, a Slack message. It does not touch a system. It does not change a workflow. It does not move a number.
This is the 5%. It is not nothing. A senior writer drafting copy with Claude can ship 2x to 3x more work per week. That is real. It is also the ceiling for almost every team we audit, and the gap between ceiling and floor stays wide for one reason: the team has never been shown what comes next.
Audit of 11 agency teams using Claude or ChatGPT daily. Median use case: text generation and summarisation. Zero teams had wired AI into a production system. Zero teams had built a tool for themselves. Sample is small, ratio is consistent.
02What the other 95% looks like
The 95% is everything that connects AI to the systems your team already runs. Some examples from the last sprint cycle:
- Build a Shopify section. Not "draft copy for a section." Actually generate the Liquid, register the schema, ship it to the theme via the Admin API, verify the live URL returns 200. Three minutes for a junior with the right setup. Three weeks the old way.
- Reconcile Klaviyo flows against actual GA4 data. Pull both via API, diff them, surface the flows where Klaviyo says "sent" and GA4 says "0 sessions from this campaign." Find dead segments. The team did this manually once a quarter. With AI plus the right MCP, it runs daily.
- Spin up a custom client report. Not "write a template." Actually pull the numbers, format the comparison, generate the deck, drop it in the shared Drive. The work that takes a senior account manager four hours every Monday morning. Replaced by a 90-second script the team owns.
- Wire Klaviyo flows to the segment tracker the ops team built three years ago in Airtable. Stop the daily CSV export. Stop the "did anyone update the sheet" Slack thread.
Notice what these have in common. None of them are "use AI better." All of them are "wire AI into the part of the business where the work actually happens." The skill is not prompting. The skill is plumbing.
03Why the gap stays wide
If the 95% is so valuable, why does almost no operating team get there on their own? Three reasons, in order of importance.
1. Tool fluency, not AI fluency
The 95% requires Claude Code, MCPs, the Shopify Admin API, the Klaviyo API, OAuth, file system access, and the ability to read a stack trace. Those skills are normal in engineering teams and rare everywhere else. Your senior strategist has never opened a terminal. That is not a defect. It is a training problem nobody has solved.
2. The frontier is invisible
If you have only used the chat box, you do not know the chat box is not the product. You do not know that Claude Code exists, that MCPs exist, that the same AI you use to clean emails can also generate a Liquid section, run it locally, fix the bug, and push it to your store. The frontier is invisible from inside the typewriter.
3. Agencies sell the gap
This one is uncomfortable. Most agencies charge 4 to 8 thousand euros a month for the kind of work an in-house team could now do in an afternoon. The agency knows. The client does not. The information asymmetry is the business model. We are not interested in protecting that asymmetry, which is part of why this site exists.
The skill is not prompting. The skill is plumbing. Wedge field notes
04The shape of the fix
Closing the gap is not a course. It is not a Slack channel of tips. It is a compressed week of pair-building inside the team's actual stack, with the team's actual data, against the team's actual problems. The output of week one is not knowledge. The output is five tools the team owns and runs without us in the room.
That is what a Wedge sprint is. Five days, five outputs, no retainer. We walk in. We rewire. We walk out. The team is on the other side of the typewriter.
If your team has Claude open every day and your agency invoice is still 8k a month, the gap is the whole point. Look at what we ship in 5 days to see what's on the other side of it.