Adopting AI coding tools is easy. Changing how your team actually ships is the hard part.
I ran GitHub Copilot's rollout at WestJet from a 60-engineer pilot to more than 200, and across thirteen years I have built across the stack: enterprise platforms, internal tools, web apps, automations and agentic AI systems, and real client work, as both an engineer and a lead. Getting a team to genuinely change how it ships is what I do.
For engineering leaders, from a handful of developers to a couple hundred. Remote across Canada and the US.
I ran this at scale. I did not advise on it from the outside.
The tools are bought. The change isn't happening.
Adoption goes flat
Seats are assigned, a few champions lean in, and everyone else drifts back to how they worked before.
Thirty different styles
With no shared patterns, every engineer uses the tools differently, and code review gets harder, not easier.
Security says no
Without guardrails for data, IP, and review, the people responsible for risk freeze the rollout, and they are right to.
Nobody is measuring
No baseline and no signal on whether any of this is helping, so the spend cannot be defended at budget time.
Assess, pilot, scale.
A way of thinking about the rollout, not a fixed package. The exact shape depends on your team, your stack, and where you already are.
Where the team really is, what will stall it, and the baseline you measure before anything changes.
A cohort gets hands-on. Shared patterns, and the guardrails that get security to yes, take shape against real work, not slides.
Expand across the org with the patterns that worked, track adoption honestly, and keep what is actually paying off.
Every team is different. Treat this as the arc, not a calendar.
What good looks like.
- A team that reaches for the tools by default, not just the champions.
- Shared patterns instead of thirty personal styles.
- Guardrails clear enough that security can say yes.
- Adoption you can actually see in the numbers.
- Licenses that earn what you are paying for them.
Growing seniors in an AI-first team.
Here is the harder problem, the one I am still working out with the teams I talk to. Standard AI use does not build skill. The judgment that catches a wrong answer fast, the kind a senior has and a junior does not, is exactly what gets skipped when an agent writes the first draft. And if the junior tasks that used to build that judgment are the first ones you automate, where do your next seniors come from? Left alone, output homogenizes and the bench thins out.
I do not have a packaged program to sell you here. I have a point of view, drawn from years of leading and mentoring engineers, including rebuilding a team after near-total turnover and coaching it across the full stack. A few of the practices it starts from:
- Juniors explain and defend what the agent wrote, in review, before it merges, so the judgment muscle still gets built.
- Some work stays hand-coded on purpose, chosen so people still learn the parts the agent would otherwise hide.
- You watch one signal: can a junior catch the agent's wrong answer without being told it is wrong. You can see that grow, or not.
The founding cohort is a few teams shaping this together. A first call is concrete, not a pitch: you leave with a review practice your team can run next sprint, and a clear signal to start watching, so over the next few sprints you can tell which of your juniors are actually leveling up.
New, founding cohortTwo proven ways to start.
Both are work I have run before: turning licenses into shipped code, and the hands-on method behind it. Start where it fits.
Adopt
The full rollout this page describes: assess, pilot, scale, with the patterns and guardrails that make adoption stick.
Book a working session →Train
A hands-on workshop where the team learns the method, not the buttons: porting specs, tests, review, and context discipline onto agentic coding.
Book a working session →The talent pipeline is the newer, less-charted work, and it has its own section above. Talk about the talent problem →
Engagements are quoted to scope.
Someone senior who has actually run this.
I am Safi. Thirteen years building software, most recently a senior engineer and tech lead at WestJet, where I ran our GitHub Copilot rollout from a 60-engineer pilot to more than 200 and shipped FlightClub to 14,000 employees. Before that I led teams of up to twelve and spent years mentoring engineers, which is the experience the talent work draws on. I run this hands-on with your team, through your own people, not as advice from the sidelines.
More about me →Bring me the hard part.
Thirty minutes on your AI rollout or your talent pipeline, and you will leave knowing where it is stuck and what to do first. No deck required.
Book a working session →