From onboarded to value

Most organisations that rolled AI tools out to their developers are in the same spot: the tool is in place, the teams are trained, but the gains stall. This is what actually happens, and what you can do about it. With us or without us.

Onboarded is not the same as changed

You have done the hard part. Tools rolled out, teams trained, maybe the whole organisation. That is further along than most who only talk about it.

But onboarding and a changed way of working are two different things. The common pattern: the agent is used as an assistant for the last little bit. Tickets are broken down by hand. Design is done manually. That is the old way of working with a new tool glued to the edge, not agentic development. The lift is in the step from there, and it is not taken on its own.

The uncomfortable truth first

Do not count on everyone becoming twice as fast. That picture does not exist. About half of the developers adopt the new way of working, half do not. It is rarely about competence -- many chose the profession because they wanted to write code themselves, not direct a text predictor. You cannot train it away. Plan for it instead of measuring against a number that will never come.

The senior paradox

The seniors are the hardest to bring along and at the same time the only ones who can truly validate output. Juniors who push ahead create more technical debt per unit of time, not less. The solution is not to leave anyone out -- it is to move the seniors' value, not their habit. They should not write less code. They should decide how the team builds, what may be checked in, what "good enough" means. That is a leadership question as much as a technical one.

The working loop

Not magic. Discipline. Read the project first. Plan to a file. Flush the context window. Build from the plan. Review every line -- that step IS the job. Speak the problem instead of writing it. Switch models deliberately. Own your commit. If the team takes away only one thing: load context first, prompt second.

Review on confidence

Give the review a scale so it becomes a decision, not a feeling. 90%+ : a quick read-through. 70-89% : line by line. Below 70% : stop, load better context, start over. Tests being green does not mean they test the right thing.

Context beats prompt

The quality of the output does not depend on how cleverly the question is phrased, but on what the tool already knows when you ask. The scepticism you see is often self-inflicted: no context in, junk out. Messy in, crisp out -- talk freely in the specification phase, be precise in the iterations. AI suggests, the expert decides.

Voice against the codebase

A senior talks into their repo, runs the transcript against the codebase, gets a structured report against team-agreed criteria. Not AI as magic -- the world's best transcription against a codebase. It lowers technical debt, raises documentation, captures the senior's knowledge. And it gets the seniors over the threshold, because it starts in what they already know.

Find your why

The seniors have little to gain personally from doing something they already do well twice as fast. A way of working without a why that is more than "produce more code" stalls. Many already have the answer but communicate it weakly -- tie the way of working to something that matters to the people who are meant to do it.

Start narrow

Do not boil the ocean. One or two teams, a task alongside live production so no one is graded on their output, and prove the approach before you scale. The unevenness between teams is not solved with more generic training -- it is solved with coaching per team, on their own tasks.

In summary

You are further along than most. The problem is a luxury problem: from onboarded to value. Do not aim for 100% adoption -- aim for the adaptable half plus structure around the seniors. Move the seniors' value, not their habit. The lift sits in the way of working, not in the tool.

You can start this on Monday. If you want help taking the team all the way, that is exactly what a coached Sprint does -- three sessions, one team, lasting conventions the team itself owns.