In December 2025 we bought AI licenses for our engineers and handed them out. We expected the usage dashboard to climb over the following weeks. It stayed flat. A small group used the tools every day, a larger group opened them once or twice, and most of the company kept working the way it had in November.
If you run technology for a company and your AI tool usage sits under 30%, you know this dashboard. You approved the budget, security signed off, and the seats are there. Your engineers still write code the way they did last year. We are a nearshore engineering company that sells AI-assisted delivery, so the flat line stung us more than it would sting most. This is our account of what we tried, what failed, and what moved the number.
Training was our first answer, and it failed by session two
We did what most companies do next and scheduled training. We ran Friday sessions with our own people and closed each one with a surprise exam to check that the material had landed.
Attendance held for the first Friday. By the second session people had started to drop out. Project deadlines gave them a polite excuse, and the exams gave them a reason. An engineer with ten years of delivery behind her does not enjoy a pop quiz on a tool she has yet to find a use for. We had turned adoption into a compliance exercise, and our engineers treated it like one: they did the minimum or found somewhere else to be.
We had diagnosed developer AI adoption as a knowledge gap. Our engineers understood the tools. They lacked a reason of their own to change how they worked.
The question that changed the plan
In a leadership conversation, David Espinal asked the room what would make people get on board.
His question moved our attention from the tool to the person. Licenses and classes are things a company does to engineers. Getting on board is a decision each engineer makes alone, on a real ticket, with a deadline close. We began to run the rollout as AI adoption change management, with the rigor we give a client’s delivery plan, and that work became the People axis of our MOA.
One curve, several interventions
Everett Rogers mapped how a new practice moves through a population: a few innovators first, then early adopters, then the two majorities, then the holdouts. Our Friday trainings ignored that map. We gave one intervention to the whole company, and it fit almost none of it. The innovators were bored and the majority felt examined.
The MOA assigns a different intervention to each segment of the curve.
Innovators and early adopters got a job in the AI champions network. Champions are engineers who were using the tools well on real projects before we asked. We gave them time and standing to help teammates inside the team, on the ticket the teammate has open that day. A colleague who shows you a faster path through your own pull request persuades you in a way a Friday slide deck cannot.
The curious majority got AI at Night, evening sessions with no exam and no attendance sheet. Engineers bring something they want to build and try it with others around. The format removes the two things that sank the Friday trainings, the grade and the obligation.
Badges gave progress a public shape. An engineer earns one by demonstrating a skill on working software, and teammates can see it. Engineers value recognition from peers above a certificate from HR.
For the late majority we skipped persuasion and showed them agents their own colleagues had built.
The marketplace: engineers adopt what they helped build
Our strongest mechanism came from the Ceiba Method itself. The Method runs on specialized agents that work alongside our engineers across the development lifecycle. For a long time a small group built those agents and everyone else received them.
We opened the supply side. An engineer who meets a recurring problem on a project can build an agent for it and submit it to our internal marketplace. Reviewers check the agent before anyone else can use it. Twelve agents have cleared that review so far.
Twelve is a modest number. Each of the twelve has an author who sits in a delivery team, uses the agent on client work, and hears from colleagues the day it breaks. In the marketplace an engineer holds a different position than a license gives. With a license, you consume a tool somebody else chose. With a published agent, your name is on it and your teammates depend on it. Authors use their own work, and they recruit users for it.
The review also answers the question your security team will ask. An open door for home-built agents sounds like shadow IT. With a review gate you get a governed catalog: you know who wrote each agent and which checks it passed.
If your usage sits under 30%
Read your usage number by segment before you act on it. A flat 25% can hide a group of heavy users who could serve as your champions and a majority that opened the tool twice. Those two groups need different things from you.
Drop the exams. Measure AI tool usage on real work, in pull requests and delivered stories, and leave classroom metrics alone.
Give your early adopters a role and the hours to perform it. A champion with no time allocation is a volunteer, and volunteers stop the week the sprint gets tight.
Build a path from user to author. Your engineers will adopt what they helped build, and a reviewed internal catalog lets them build without opening a hole in your governance.
A nearshore partner that has been through it
Change management takes longer than procurement, and your roadmap will not wait for it. A nearshore partner whose engineers went through this curve can shorten yours. Our teams work in your time zone, inside your standups and code reviews, so your developers see these practices on their own tickets during their own working day. A Ceiba champion can pair with your engineer at 10 a.m. your time, on your codebase, and that hour does more than a recorded webinar from twelve time zones away.
We still have engineers at the far end of the curve, and we expect to. Since December we have learned who they are and which intervention we owe them.
If your licenses sit idle, bring your usage numbers to a session with our engineers. We will open the marketplace, show you the review an agent passes, and walk you through how champions work inside a delivery team.
Let’s Talk
