How to get your agency team to actually use AI

You bought the licences. Nobody's using them.

This is the most common AI story in small agencies right now. The founder gets excited after a weekend of playing with a new model, buys a stack of AI subscriptions, fires off a Loom and a "go play with this, it's incredible" Slack message — and three weeks later the only person using any of it is the founder. The seats are paid for. The dashboards show two logins. And the team has quietly gone back to doing everything the way they did it before.

If that's you, the problem isn't your team and it isn't the tools. The problem is that you treated adoption as an access problem ("give everyone a login") when it's actually a behaviour problem ("change how busy people do their daily work"). Access is necessary but nowhere near sufficient. People don't change a workflow they're comfortable with because a tool became available; they change it when the fear is removed, the friction is low, the first experience is a clear win, and someone follows up to make it stick.

This is the playbook we use at Fame, our own podcast agency, and that we see work again and again across small agencies. None of it requires a transformation budget, a consultant, or an all-hands "AI strategy" offsite. It requires picking the right first task, talking to your team like adults, and doing a handful of unglamorous things in the right order.

1. Start with one painful, repetitive task — not the whole toolbox

The fastest way to kill adoption is to introduce five AI tools at once. People get overwhelmed, nothing sticks, every tool gets a shallow try and a shrug, and now your team has learned that "AI" means confusion. When you give someone ten new capabilities simultaneously, you've given them a research project on top of their actual job — and their actual job will always win.

Instead, pick one task that is repetitive, time-consuming, and low-stakes — and importantly, not your most important or most visible work. For most agencies that's something like: drafting first-pass show notes, turning a call transcript into a recap, writing the first version of a social caption, cleaning up meeting notes into action items, or summarising a pile of client feedback into themes. The criteria are simple. It should happen often (so the habit has many chances to form), eat real time (so the win is obvious), and carry low risk (so a bad output costs nothing but a redo).

Make that one task the entire rollout for the first two weeks. One tool, one workflow, one win. Resist the urge to say "and you can also use it for X, Y and Z" — you can tell them that in week three. For now, you want the whole team to have the same single, repeatable success. Shared success on one workflow is what gives people the confidence — and the permission — to go explore the rest on their own.

A useful test before you pick: could a new hire screw this task up with an AI draft and no real harm be done? If yes, it's a great first workflow. If a mistake would reach a client or damage trust, save it for later.

2. Name the fear out loud

The number one driver of resistance to AI isn't laziness, and it isn't that your team is "behind." It's fear. Fear of looking stupid when they don't get a good result. Fear of "is this the first step toward replacing me?" Fear of getting it wrong in front of a client and being blamed. Most of this fear is never said out loud — which is exactly why it's so effective at quietly killing adoption. People won't tell you they're scared; they'll just not use the thing.

You cannot train your way past fear with a feature walkthrough, because the fear isn't about features. You address it the only way fear is ever addressed: by naming it directly and honestly. In a team setting, say the quiet parts out loud:

• "Nobody here is being replaced by this. The point is to take the boring 60% off your plate so you have more time for the 40% only you can do — the judgement, the craft, the client relationships."
• "You will never get in trouble for a bad AI draft. The only way to get in trouble is to send AI output to a client without a human checking it."
• "I'm still learning this too. Here are three places it's been confidently wrong for me this week."

That last point matters more than people expect. When the founder admits their own fumbles, it gives everyone else permission to be a beginner. When the founder presents AI as a flawless mandate from on high, the team reads it as pressure, and pressure produces compliance theatre — people who say they're using it and quietly aren't. Vulnerability from the top buys you honesty from the team, and honesty is what lets you actually find and fix the friction.

3. Do a 30-minute live walkthrough on real work

A recorded video and a written guide will not move the needle, no matter how good they are. People learn workflows by watching someone do them on real work, and then immediately trying it themselves. Passive learning — reading a doc, watching a webinar — produces recognition ("I've seen this") but not capability ("I can do this"). Capability is what drives adoption.

So run a live session, 30 minutes, no slides. Take an actual task from your agency — a real transcript, a real client brief, a real messy set of notes — and do the whole thing live, narrating your thinking as you go. Crucially, include the parts where the AI gets it wrong and show how you notice and correct it. Watching you catch a hallucination and fix it teaches more than any "best practices" slide, because it models the single most important skill: treating AI output as a draft to be checked, not an answer to be trusted.

Then — and this is the part most people skip — have everyone do the same task on their own example before the call ends. Not "try it later." Now, while you're there to help. The goal is that every person leaves the session having personally produced one good result. That first personal win is the hook. People who get a win in the room come back; people who are told to "try it on your own time" mostly don't.

4. Use a champion, not a mandate

In every team there's at least one early adopter who's already curious and already poking at these tools on their own. Find that person and make them the go-to, not you. There's a simple reason this works better than a top-down mandate: when the message is "here's how I use it for our exact workflow," coming from a peer who does the same job, it lands as helpful. When the same message comes from the boss, it lands as pressure.

Give your champion a small amount of status for the role — "Maya's our go-to for AI editing questions" — and, importantly, a little bit of time to actually help others. If you make someone the champion but give them no slack to do it, you've set them up to fail. Peer proof is the most powerful adoption tool you have, and it's nearly free; the only cost is a few hours of one enthusiastic person's time and a bit of public credit.

A second-order benefit: champions surface the real use cases. Left to their own curiosity, they'll find applications you would never have scripted from the top, and those organic discoveries — shared with the team — are far more credible than anything in an official rollout plan.

5. Measure outcomes, not tool usage

Here's a trap that feels productive and quietly poisons everything: rewarding "did you use the AI tool." If you measure and praise usage, people will perform usage — they'll run a token prompt to show up in the stats and learn nothing real. You'll get a dashboard that looks healthy and a team that hasn't actually changed how it works.

Measure the outcome instead. Show notes shipped faster. More episodes out the door per week. Fewer revision rounds with the client. The effective hours saved on a recurring task. When the outcome is what matters, the right usage emerges on its own — and people find their own use cases in pursuit of the result, because the result is what they're actually being recognised for.

And when a win happens, make it visible. "This used to take Sam two hours every Monday; this week it took twenty minutes." Small, specific, shared wins are the real engine of adoption. They do something no training session can: they make the person sitting next to Sam quietly think, "wait, I want that too." That self-motivated pull is worth ten top-down pushes.

6. Survive the two-week cliff

There's a predictable moment where adoption goes to die, and it lands about two weeks after launch. The initial buzz has faded, the novelty is gone, and then someone hits a genuinely confusing edge case — a result that's wrong in a way they don't know how to fix, or a workflow that breaks on a weird input. With no support at that exact moment, the rational move is to quietly revert to the old way that always worked. Multiply that by a few people and your rollout is dead, not with a bang but with a shrug.

You beat the cliff by planning for it before you launch. Put a 14-day check-in on the calendar on day one, not as a vague "let's see how it's going" but as a real session with three questions: what's working, what's confusing, and what's the one thing we'll change. That single scheduled follow-up does two things. It catches the friction before it hardens into "yeah, we tried AI, it didn't really land." And it signals that this matters enough to come back to — which itself drives people to keep at it, because they know they'll be asked.

7. Write the one-page rules so people feel safe

A short, plain AI usage policy is not bureaucracy — it's permission. Counterintuitively, clear rules increase experimentation, because the thing holding cautious people back is uncertainty about what's allowed. When someone knows exactly what's approved, what must never go in (client data, credentials, anything under NDA), and who to ask when unsure, they stop hovering and start trying things. Vagueness makes people cautious; clarity makes them confident.

Keep it to a single page: approved tools, what never goes into any AI tool, the rule that AI output is always a first draft a human reviews, and the name of the person to ask when in doubt. We've written a free AI usage policy template you can adapt in ten minutes — introduce it in the same live session where you teach the first workflow, so safety and capability arrive together.

A realistic timeline

Put together, the whole thing is about a six-week arc, not a six-month programme. Weeks one and two: pick the workflow, name the fear, run the live walkthrough, ship the policy. Weeks two to three: let the champion field questions and capture the first wins. Around week two, hold the check-in and fix the friction you find. Weeks four onward: once the first workflow is a genuine habit — not just "tried it," but the default way that task gets done — add the next one. The pattern repeats, one workflow at a time, and that drumbeat is what turns "we have AI tools" into "this is how we work."

Where this connects to your tools

Getting the team to use AI is really a special case of a much bigger problem: getting your team to adopt new tools at all. Every principle above — start small, kill the fear, win first, follow up, measure outcomes — applies just as well to the internal tools you roll out to run the agency: a team portal, a time tracker, a client dashboard. The difference with a purpose-built internal tool is that adoption is dramatically easier when the tool fits your exact workflow instead of being a generic app your team has to bend around. Friction is the enemy of adoption, and the biggest source of friction is a tool that wasn't shaped to how you actually work.

That's the thesis behind Forge. We build internal tools shaped to how your specific agency works, host them so there's nothing for you to maintain, and — crucially — show you who's actually using them, so adoption stops being a guess and becomes something you can see and improve. See how it works →