03 writeup

How to Actually Get Your Team to Use AI

Sam Parr asked a question that's been rattling around my head for weeks: "How is everyone getting team adoption for Claude? How do you get your team to actually use it effectively without spending all their time on Twitter and learning?"

It's the right question. Most companies are stuck here. They've bought the licenses, done the training, sent the memos. And still, adoption flatlines. I just joined Every as head of tech consulting, and I've been spending my days inside companies wrestling with exactly this problem. Here's what I've learned so far.

---

Sam's frustration is familiar. You're on Twitter, you see all the AI stuff popping up, you're testing the fringe ideas. You know what's possible. But your team? They're not spending their days on Twitter. They shouldn't be. So how do you bridge that gap?

After a bunch of consulting engagements, I've landed on seven principles. They're not theoretical — we've been running them inside Every Consulting, and they're starting to work.

1. Buy the model direct, not third-party tools

People love to evaluate tools that wrap Claude or Codex or Gemini. And look, Cursor is a good product org. They're doing great work. But I don't know how you compete with Anthropic offering $5,000 worth of tokens a month versus a $200 subscription.

The problem with third-party tools is they come with someone else's opinions baked in. Their product decisions, their hang-ups, their prioritization. When you buy direct, you get flexibility. You can schedule around model releases, train on the cutting edge, configure your own approach. The core model companies know what's coming — they build their internal tools to handle those releases. If you're one layer removed, you're always catching up.

There's a real move toward build, not buy. It's cheaper, faster, and more flexible.

2. Raise the ceiling, not the floor

Most companies approach adoption like a mandate: "Everyone must use AI now. We bought you the tools. Use them."

This doesn't work. Even on pain of death, people won't adopt AI if they're not ready. It's a self-preservation thing. The stick backfires every time.

Instead, use the carrot. Find the people who are already AI-forward — the ones using Codex in their spare time, the ones who are genuinely curious. Give them support. Give them access. Show the rest of the org that these people are the ones getting promoted, getting facetime with senior management, getting the interesting projects.

Someone who's really AI-pilled can do five to ten times more than someone who hasn't seen the magic yet. You can't force the magic. But you can make it visible.

3. Workshops with 50%+ build time

Nobody wants to sit through slides about AI theory. The way people actually learn this stuff is by doing.

Our workshops are at least half build time. We give people synthetic data, working connectors, and a real problem to solve. The facilitator is there to unblock, not to lecture. And the biggest barrier we hear from teams is simple: they don't have time. They don't have space in their day to experiment.

A workshop gives them that space. Two hours where they're expected to build something new, go outside their normal domain, and actually touch the tool. That's where the aha moment happens. That's where it clicks.

4. Assign impossible tasks

If your goal is to write one blog post a week, you can do that manually. If your goal is to write one a day, you're going to need AI — for research, for editing, for structuring.

Boris Cheney (one of the founders of Codex) said something similar: slightly under-resource your teams, and they'll figure out that AI is the only way through. I like to make it more explicit. Choose goals that are literally impossible without AI. Don't frame it as "starting today, you must use AI." Frame it as "our goal is to get to ten times the output — how do we get there?"

People are strategic. When they know the destination, they'll figure out the route.

5. Mandatory AI note-taking + MCP connector

This is the single biggest unlock we've found. Everyone on the consulting team uses Granola with the Granola MCP. Every meeting gets recorded, every note gets structured, everything is queryable.

The reason this matters: 80-90% of AI's value right now is extracting information from unstructured data and structuring it in a useful way. How many times have you sat down to a task and realized you need context from a meeting three weeks ago? With an MCP connector, that context is at your fingertips. You can generate update emails, pull context for proposals, build curriculum — all from the record of conversations you were already having.

It seems small. It's not. It's the foundation everything else builds on.

6. Map workflows and systematically automate

We have a process for discovery calls: talk about what someone does day-to-day, what tools they use, what their pain points are. Turn that into a structured spreadsheet. Then systematically work down the list, building a skill for each type of task.

The assumption is bold: we're trying to get 100% of the routine work off the humans. If a skill can do a first pass on a task — even an imperfect one — the human can operate at five to ten times their current throughput.

What's interesting is what happens next. Nobody gets let go. Instead, teams either invest way more in each individual task (we went from one group exercise per course to a personalized project per participant), or they expand their output without hiring. The constraint shifts from headcount to ambition.

7. Train people to be managers — of agents

Here's the thing that's crept up on everyone: every individual contributor is now a manager. They're managing AI tools, and they don't have management training. They don't know how to context-switch. They don't know how to set up systems and rules. They don't know how to evaluate work they didn't personally produce.

The funny twist? Managers actually take to this faster than ICs. Managers have long since stopped caring about how a problem is solved — they just want it solved to spec. And then it gets weirder: managers start becoming individual contributors again, because it's often faster to vibe-code a task than to brief a human, wait for them to brief their Claude, and wait for the response to come back a couple of days later.

We need to teach people how to manage agents. How to do evals. How to develop taste for what to work on. How to systematically train a skill — "write a really good PowerPoint deck" or "produce a daily update report." That's the management training nobody's offering, and it's the one that matters most.

---

I don't have this all figured out. Every consulting engagement teaches me something new. But these seven principles are the ones I keep coming back to. If you're trying to get your team to actually use AI — not just buy licenses and hope — start here.

I'm at @hammer_mt on X. I'd love to hear what's working for you.