Fifty per cent right is the starting point

Thanks for coming back for issue two. The last fortnight has been a busy mix of one-on-one coaching, a demo with a law firm, and interviews with a client whose accreditation process we’re automating. A few things from those conversations have stuck with me.

The first is a business owner I coach one-on-one. A few months ago he hadn’t used any AI tools. He’s now set up his whole business around Claude: a separate Claude Project for each business pitch, loaded with the context of that pitch. This fortnight we got Claude filing his meeting transcripts into the right pitch folder in OneDrive, so by the time he’s back at his laptop the filing is already done. He records the meeting in Otter, and Claude works out which pitch it belongs to. Watching someone go from never-used-it to that in a few months has been one of the best parts of my year.

The second came from an interview I ran with a coordinator whose accreditation process we’re automating. She’d spent two hours talking to Claude about how the process works before our call. Her words: “within the last five minutes I’ve had a nine page document generated from me going blah blah blah blah blah.” The conversation is the work: the document is just Claude tidying it up. I’ve started treating transcripts the same way. A recorded conversation is data, and I can ask Claude to go back through it wearing different hats: what would a customer notice here, where’s the revenue opportunity, what did we miss.

The third is something I said in that same session and have been chewing on since. When Claude first attempts a complex task, I expect it to be 50 to 60 per cent right. That used to frustrate me. Now it’s just how I work: the skill isn’t writing the perfect prompt, it’s knowing what a good output looks like so I can say “these three things are wrong, go again”. It’s the same as reviewing a draft from someone new on the team. I wouldn’t hand back their first draft and declare the whole thing a failure.

One more thought before the takeaways. I keep hearing that teams can’t start with AI until their data is cleaned up. That’s the wrong frame. A messy spreadsheet Claude can find is more useful than a beautifully formatted one it doesn’t know exists. Knowing where things are beats having them perfect.

Three things to try

• Brief Claude like you’d brief someone new on the team: the role, the stage you’re at, and what the output should look like. A framework without that context gets used badly, by people and by AI.

• Next complex task, expect 50 to 60 per cent on the first pass. Instead of starting again, tell Claude exactly which parts are wrong and what you expected. The second pass will show improvement.

• Talk your thinking into Claude for ten minutes: a process you run, a decision you’re weighing up. Then ask for a structured document. Compare it with what you’d have produced typing.

What’s new in AI (worth knowing)

Claude can now do work inside Microsoft 365

Until now Claude could only search your Microsoft files. It can now draft and send email in Outlook, manage your calendar, and create and update files in OneDrive and SharePoint. If your organisation runs on Microsoft, this is the moment Claude stops being a separate window and starts working where your work already lives. Your admin needs to enable the Microsoft 365 connector with write tools turned on: worth asking about.

ChatGPT’s voice mode grew up

OpenAI released GPT-Live, new voice models that listen and speak at the same time instead of taking turns. It will go “mhmm” while you’re mid-thought, and stays quiet when you pause to think. It’s rolling out now in the ChatGPT app: open voice mode and you may already have it. I find voice the easiest way to use AI between meetings, so this one matters more than it sounds (ahahah pun intended).

Chat and “AI doing the work” are becoming one window

Claude now puts Chat and Cowork side by side in one home, and Microsoft made Copilot’s equivalent generally available in the same fortnight. Two of the biggest players making the same move within days tells you where the next year is heading: AI that does the work alongside you, in the same place you type. The question is shifting from “should we use AI chat” to “what work are we ready to hand over”.

If anything from the above resonates with where you are, reply to this email. I read every one.

Gareth

P.S. The insights in this issue came from workshops and one-on-one coaching sessions I run on AI adoption. If any of it sounds useful for your team, or you’d like to work through your own AI adoption, reply to this email and we’ll work out what makes sense.