The human side of rolling out AI at your company

Most AI rollout post-mortems say the same thing: the technology worked, the people didn't change. Not because they were resistant or incompetent — because nobody thought carefully about change management. The configuration was fine. The communication and support structure wasn't.

This is the part that determines whether your rollout actually delivers value.

Why people don't change even when the tool is good

The barrier to adopting a new tool is almost never capability. It is habit, uncertainty, and the feeling that the new way is more work than the old way — at least at first.

When someone opens Claude for the first time, they face several questions simultaneously: What should I type? How will I know if the output is good? What happens if it's wrong? Is this the kind of thing I'm supposed to use this for? If any of those questions don't have clear answers, the path of least resistance is to close the tab and keep doing things the way they've always done them.

Your job in change management is to answer those questions before people have to ask them.

What actually works: the first-use moment

The most important intervention is not training. It is the first-use moment — the first time someone opens Claude with a real task in mind.

People need to leave that first session feeling that it was worth it. One successful experience is worth ten slides of onboarding material.

Design for this:

• Give people a specific task to try first, not an open invitation to "explore." Something from their actual work that will take 20-30 minutes and where the output is obviously useful.
• Make sure their Project is configured before they log in. They should not have to figure out settings.
• Have someone available to help for the first 30 minutes — a Slack message is fine. Just someone who will answer "why did it do that?" questions quickly.

One good first session creates an advocate. One confusing first session creates a skeptic who will tell three colleagues.

The early adopter strategy

Do not roll out to the whole team at once. Find the two or three people per team who are most curious about new tools. Roll out to them first. Give them two weeks to develop their own practice — prompts that work, workflows they have changed, outputs they have shared.

Then have them present to their team. Not as an IT demo — as a peer saying "here's what I actually changed in my week." Peer credibility is exponentially more powerful than admin credibility. The same information about how Claude works lands completely differently when it comes from a colleague who has done it.

This approach also gives you time to find the problems before they happen at scale. The early adopters will encounter the edge cases, the workflow fits that don't quite work, the prompts that need refining. Better to discover those in a small group than across the whole team simultaneously.

The question you need to answer for every team

For each team, the change management question is: "What is the specific workflow change you want them to make?" Not "use Claude more" — a specific behaviour change.

For a CS team: "Before you draft a response to a complex ticket, open Claude and paste the ticket in. Use the draft Claude produces as your starting point."

For a marketing team: "When you have a content brief, use Claude to produce three angle options before you start writing."

For an operations team: "When you're documenting a new process, talk it through with Claude first and use its structure as the starting point for your SOP."

These are specific, testable, and small enough to try once and evaluate. Vague encouragement to "use AI" produces vague adoption. Specific behaviour targets produce measurable change.

Handling resistance

Most resistance is not ideological. It is practical: people are busy, learning a new tool is an investment, and the payoff is not always immediate. Respect this.

The people most resistant to Claude are often the most experienced — they have workflows that work, and they are right that disrupting a working workflow for marginal gain is a bad trade. Do not try to change their whole workflow at once. Find the one task where the gain is obvious and undeniable, and start there.

The people who say "AI is going to replace us" need a different conversation. Acknowledge the concern honestly. Share your company's position. Be clear about what Claude is being used for and what it is not. Uncertainty about job security is legitimate — pretending it isn't makes it worse.

The ongoing work

Change management does not end at launch. At 30 days, do a structured check-in (four questions: what are you using it for, what's faster, what stopped, what's not working). At 90 days, identify which teams have meaningfully changed how they work and which haven't — and find out why.

The teams that are not changing are telling you something. Maybe the configuration is wrong. Maybe the use cases don't fit. Maybe the team lead is not modelling the behaviour. Each of these has a different solution. Ask before assuming.

The rollouts that work are the ones where someone treats change management as a continuous process — not a launch event. The configuration is a quarter of the job. The rest is helping people actually change.