An AI adoption journey: somebody still has to read it
An AI-assisted pull request can take an hour to write. Reading it properly can take someone else the afternoon.
Most teams I’ve seen go through AI adoption feel this before they name it. The person shipping feels fast. The person reviewing feels tired, and isn’t sure they’re allowed to say so.
So I wrote it as a story. A made-up company called Fjordly. Six people building booking software for small clinics. They just got Claude Code, and a senior engineer called Marit goes all in first.
It’s an AI augmentation journey in four minutes. What it takes to go from one fast person to an AI-augmented team: trust between the person who sends and the person who reads, and who ends up paying for the speed.
Watch the story (4 min, narrated, with music, MP4 9.4 MB).
Transcript
This is a four-minute story about AI adoption, and who pays when one person gets fast. You shipped it in an hour. Someone else spent the afternoon reading it. Maybe you’re the one who shipped. Maybe you’re the one who read. Either way, somebody paid. Meet Fjordly. A made-up company. Six people. They build booking software for small clinics. They just got Claude Code. Pick the person you identify with the most, and follow them.
January, 2026. Everyone has the same tool. Most people use it at the edges. Marit goes all in. She doesn’t trust it yet, so she reads every line. She ships very little.
Spring. Marit connects the AI to the team’s logs, metrics and tickets. When something breaks, she knows why in ten minutes, and she can check every answer against the logs. Writing new code is different. There’s nothing to check it against. But the trust she built fixing problems starts leaking into the code she writes. And Amir starts sending unread work too.
Late summer. The bill arrives. Look at the colours. Marit feels fine. Everyone around her is paying. Every line she didn’t read, someone else had to. And Marit starts to wonder if they think she’s a fanatic. Nobody’s said it. It’s in her head.
October. The turn. Marit doesn’t say, the AI fooled me. She says, I didn’t read it. The AI sounded sure, and Marit let its confidence stand in for her own.
So the order changes. Agents review first. Then Marit checks what they found, and reads the work herself. Nothing reaches the team until Marit understands it. Her reviewers start using AI too, as an assistant when they review. And one rule. The agents’ verdicts stay with Marit. You don’t paste a model’s opinion into someone else’s review.
Early 2027. From here, this is a guess. The rest of the team starts using agents. Marit has already made the mistakes, so she saves them the trouble. Reviews get fast. Then a seventh teammate joins. Ada, an AI. Nobody had to leave to make room for her.
After that, honestly, nobody knows. So every guess here wears a label. Further out is fog. But in the near term, AI likely becomes part of every job. And as the lines between jobs blur, the work becomes figuring out what to build, and why. Letting AI build it. And making sure it’s right.
So, what did the team learn? There will be friction. You will get it wrong. Own it, fix it, try again. Shape the workflow around your people, not the tool. And AI isn’t always the answer. The team went looking for an AI voice tool. Then they realised the dictation built into their laptops already did the job.
So, one question. What did you send last week, without reading?
Made by Achint Satsangi at Claude Impact Lab, Oslo.
Prefer your own pace? The same story is below as a deck you can click through. Pick the character you identify with the most, and follow them to the end.
What’s real here
Fjordly, Marit and the rest of the team are fiction. Nobody in it is a real colleague, and the company doesn’t exist.
The numbers on screen are fictional too, inspired by real experience across teams. Read them as a shape, not as data.
Everything dated 2027 is guesswork, and the slides say so.
Narrated in my own voice, cloned with an open-source model on my laptop. I wrote every word it says.
Music: “Upbeat Corporate” by The_Mountain, via Pixabay. It’s an AI-generated track under the Pixabay Content License, so the story’s CC license doesn’t cover it.
Made at Claude Impact Lab, Oslo, 21 October 2026. Story, deck and video are licensed CC BY-NC-ND 4.0.
If your team has already made this work, with agents reviewing first and people reading what they send each other, what did you have to agree on before your reviewers trusted it?