We run co-work training sessions for engineering teams deploying AI tools. Before every engagement, we survey the team. One of our standard questions:
"If you've reduced your use of Copilot (or stopped entirely), why?"
Forty engineers answered this across our last several engagements. Here's what they actually said — not what managers think they said.
The Top Answers (Ranked by Frequency)
1. "The suggestions were wrong too often and I stopped trusting it" (38%)
This is the most common answer. And it's almost never about the model quality.
When we dig in, it's usually this: the engineer never learned to provide enough context. They were prompting with 10 words in a codebase the model couldn't see. Of course the suggestions were mediocre.
The engineers who stuck with it learned to front-load context. The engineers who bounced didn't — and blamed the tool.
What actually fixes it: One 20-minute session on context-first prompting. Utilization rebounds immediately.
2. "I forgot it was there" (27%)
Not a joke. A quarter of engineers who dropped off said they simply... stopped thinking about it.
They tried it in week one. Got some results. Got busy. Defaulted to old habits. The tool was never woven into any specific workflow, so it atrophied.
This is a rollout design failure, not an adoption failure. If "use Copilot" isn't attached to a specific, recurring task, it won't stick.
What actually fixes it: Anchor the tool to one workflow. For most devs: pre-PR review. Do that one thing with Copilot every single time, for two weeks. It becomes automatic.
3. "My manager didn't use it so I assumed it wasn't important" (19%)
This one stings if you're an engineering manager.
Developers watch their managers. If the manager isn't using the tool, the implicit signal is: this doesn't matter. Nobody said that out loud. Nobody had to.
The teams with the highest utilization rates have managers who visibly use AI tools and share what they're learning. Not evangelizing — just modeling.
What actually fixes it: Engineering managers need to go through training first. Before the team. Then share one thing they learned in the next team meeting.
4. "I didn't know it could do [X]" (12%)
Twelve percent of engineers who dropped off didn't realize the tool could do the thing they needed.
They knew it could autocomplete. They didn't know it could explain legacy code, write test scaffolding from specs, or help debug by explaining its own reasoning.
This is a feature discovery problem. Most rollouts cover "how to turn it on." Almost none cover the full capability surface.
What actually fixes it: A structured capability tour — not a vendor demo, but a team session focused on "here's the stuff that surprised us."
4. "It didn't fit how my team works" (4%)
Small slice, but important. Some teams have real workflow constraints — security reviews, regulated environments, no internet in prod — where AI tooling requires adaptation. These teams get generic training that ignores their context and naturally drop off.
What actually fixes it: Role-specific and context-specific training. The answer to "our team is different" should be "let's design for that," not a generic video course.
What This Means for Managers
The pattern is clear: engineers drop off for fixable reasons. Almost none of these answers are "the tool is bad." Almost all of them are "the rollout didn't set us up."
If your team's utilization is stuck at 20–35% — which is the industry average without training — you're almost certainly hitting 2–3 of these patterns simultaneously.
The good news: they're all fixable in a single, well-designed team session.
The Measurement You're Missing
Most teams don't know why utilization is low because they never asked. They look at the usage dashboard and see 30% and call it a "Copilot problem."
Spend 10 minutes. Send a one-question survey to your team: "If you've reduced how much you use Copilot, what's the main reason?"
The answers will tell you exactly what to fix.
If you want to know what your team's utilization rate should be — and what you're leaving on the table — we built a free calculator for exactly that:
👉 askpatrick.co/roi-calculator.html
No email required. Takes 90 seconds.
Ask Patrick runs co-work training sessions for engineering teams deploying GitHub Copilot, Microsoft Copilot, and Claude Code. Flat-fee for your whole team — no per-seat licensing. askpatrick.co
Top comments (0)