1st Edition

🎧 In a rush? Listen in 3 minutes, in my own voice.

I hesitate even writing a newsletter. I tend not to take part in the hype of creating content, honestly because I just don't enjoy it. But in this case, there's great power when we share knowledge and learnings and advance our thinking together. So I'm giving this a whirl for a year.

Once a month, I'll share three things with executives leading AI adoption:

When there's big news from the AI labs, I'll start with that and what it means for you.

I'll keep clients and people anonymous. I don't think names matter for the learning, and I tend to think you just want the brass tacks of it.

So let's get into it.

What it is. This week Anthropic opened a preview of Claude Duet, an agent that works in the background. Each morning it sends you a short brief on what matters across your connected apps. During the day it messages you only when something needs you, and it keeps working on tasks after you step away, from scheduling meetings to building dashboards and forecasts.

Before it sends, pays, signs or deletes anything, it shows you exactly what it's about to do and waits for your OK. It's early: Pro and Max plans only, a waitlist, and no Microsoft 365 connection yet. Anthropic itself says it "might misread a thread or miscalibrate what's urgent."

Why it matters. This is the shift from AI you go to, to AI that comes to you, and every lab is building it. OpenAI has Dots. Microsoft's version, Autopilot, is in private preview. Meta says more than 10 million people in the US and Canada set up its Muse agent in about six weeks.

Your people will bring that expectation to work. That makes "what can an agent read and do on someone's behalf?" a policy decision, not a tooling one. And since Duet runs on individual plans, your builders will try it on personal accounts first.

Also this month

The models will eventually match to similar capabilities. Whoever is ahead this month probably won't be in six months, so picking the "best" model is the wrong question.

The better question: who are your "builders" and who are your "daily users"?

Daily users are most of your company. Email, meetings, documents. They should be on the tool with the best connectors for your business. Most of our clients are on the Microsoft stack, so for them that's Copilot. The interface is a pain, and you may need to train your people more to get value out of it, but at least it's inside your 365 environment, connected to where the work is being done.

Builders are a much smaller group already pushing these tools hard, like the estimator in the Learn section below. They need the most capable tool, and right now that's often Claude or ChatGPT. Give them a sanctioned seat with data controls instead of making them pay for it themselves.

Do you switch? Not the whole company every time a new model tops a leaderboard. These tools are going to keep changing. What carries over is your people knowing how to work with AI, and a company-wide switch resets that. For builders, switching is cheap, so let them follow the capability. Revisit the split once a quarter.

We surveyed 107 senior leaders across Canadian construction, engineering and architecture firms this year. AI maturity didn't track with headcount, sub-sector or seniority. It tracked with whether employees were paying for ChatGPT out of their own pocket.

Firms where people pay their own $20 a month scored 5.5 out of 10 on maturity. Firms where nobody does scored 3.95.

At a large employee-owned general contractor, we ran two days of workshops with 77 staff. Eleven of them described themselves as building with AI, not just using it. Nobody had asked them to.

One is a chief estimator. On his own time he's built about fifteen working tools: a bid go/no-go scorer, a contract review agent, and a takeoff-to-estimate pipeline that gets to an early estimate in about ten minutes at roughly 75% accuracy.

He built them in Claude, then moved the outputs into Copilot to stay inside policy.

He isn't alone. A project manager there pays for Claude himself because it handles documents and Excel better than the tool he was given.

Meanwhile the company had about 42 Copilot licences, and its own usage survey got 12 responses.

Sanctioned use is lower than your licence count. Real use is far higher than sanctioned. Both numbers are wrong.

84% of the leaders who named privacy as their top AI concern use AI personally anyway. Privacy fear isn't stopping AI. It's only stopping the governed version.

And the people doing this are often your best operators, and they're nervous. That estimator told us younger staff hide their AI use because they're afraid of looking replaceable. That's a retention problem, not just a tooling one.

Every company has a recurring status meeting. Ours is Monday at 11.

My prep used to be Sunday evening, four tabs open, scrolling my own sent mail trying to remember what I promised to whom. I'd still walk in with half of it.

So we gave the meeting an assistant. Here's the recipe in plain language, so you can hand it to whoever runs your tooling.

Not the technology. Writing those rules, and tightening a first version that kept claiming things that weren't mine.

It hasn't made the meeting shorter. It's made it better, because I show up knowing what I owe and what's stuck instead of finding out live.

It's the same idea as the Claude Duet morning brief above, just narrower and with stricter rules. Narrow and strict is why it works.

AI makes it cheap for me to write this letter three ways, so I'm going to. Starting next month, pick the length that fits how you actually read, and that's the one you'll get. Change it any time.

Hit reply if you have any thoughts or debate! I read every response.