The machine asks, people decide.

Every company I talk to is somewhere on the same journey with AI. Some are AI-native, built around it from day one. Some are part-way through, with people using it every day while the company itself still runs the way it did five years ago. And almost all of them are wrestling with the same underlying question, even if they don't put it this way: in a company where AI can do a lot of things, what should it actually do, and what should people keep?

I've got a strong view on this, and it's the fourth belief behind Clearahead. The machine asks, people decide. AI should do the asking, the gathering, the reporting and the remembering. People should do the thinking and the deciding. I'd draw that line very firmly, and I wouldn't move it.

The machine should take the admin

Look at what fills a leader's week in a growing company, and a startling amount of it is coordination admin. Asking people for updates. Chasing the ones who didn't reply. Collating it all into something readable. Spotting what's late. Writing the summary for the board. Remembering what was agreed three months ago and why. I went through how much of this exists purely to move information around in Information is cheap now. Decisions aren't.

None of that needs judgement, and all of it is something AI is now genuinely good at. It can ask a team what's changed and turn the answers into evidence. It can keep a picture current from dozens of small updates. It can write a clear weekly summary, notice when something looks finished, and remember how the company solved a similar problem last spring. It never gets bored of chasing and it never forgets.

So I'd hand all of that over, without hesitation. Every hour a founder or a COO spends collating updates is an hour they're not spending on the decisions only they can make, and in a small company that trade is enormously expensive.

People should keep the decisions

Where I'd stop is the decision itself. What goal to chase. Which constraint matters most. Whether to stop an initiative someone cares about. Whether a piece of evidence is convincing. Whether something is really done. These are judgements about priorities, trade-offs and people, and they belong to the people who'll live with the consequences.

There's a practical reason for this as well as a principled one. A team only commits to a decision it owns. If an AI decides the priorities, the team will follow them half-heartedly at best, because nobody in the room chose them. If the team decides, with the AI having laid out the facts, it'll actually go and do the work. Alignment, which I think is the real reason any operating model succeeds, only comes from people deciding together. I made that case in Alignment beats activity.

Information, not a gate

The subtler version of this belief is the one I care about most. It's not enough for AI to avoid making decisions; it also shouldn't block them.

There's a temptation, when you build software with an AI in it, to let the AI enforce the rules. To refuse a goal that isn't specific enough, or stop someone adding a move it doesn't think fits. It feels helpful, and it's a mistake. The moment the AI becomes a gate, the team starts working around it, or worse, starts deferring to it.

So in Clearahead, AJ's view is always information, never a gate. If the team wants to add something AJ doesn't think fits, it happens. AJ says once what it would keep an eye on, and then gets out of the way. AJ will tell you a move looks done, or a constraint looks cleared, or the goal looks achieved, and offer you the one-click way to close it. But a person presses the button. Guide, don't block.

I think this matters far beyond our product. As more of how a company runs gets touched by AI, the companies that do well will be the ones that use it to make people better informed, not to take the steering wheel away from them.

What that looks like in practice

Inside Clearahead, AJ does five jobs, and you'll notice every one of them is on the "asking" side of the line:

• It shapes. It helps set a sharp goal, asks what's holding you back until you get past the symptoms, and proposes moves for the team to keep or discard.
• It gathers. It turns each person's update, in their own words, into evidence, and keeps the current reality up to date from the move all the way up to the goal.
• It reports. Every Monday it writes This week from AJ: where you stand, what moved, what you learned, and up to three things that need a decision from you.
• It notices. It says when something looks done, cleared or achieved.
• It remembers. It keeps how each constraint was cleared, so it can tell you what worked last time.

The people side is just as clear. Architects, usually the founder, MD, COO and heads of department, set the goal, confirm constraints and make the calls AJ says need them. Contributors own moves and say what's changed in a sentence. Viewers, such as a board or investors, can read everything and ask AJ questions, without anyone preparing a pack. You can see the whole thing in action on Meet AJ, and how the roles work in the team docs.

What this does to the leader's job

The most interesting effect of drawing this line isn't on the software; it's on the people running the company. When the machine takes the asking and the admin, the leader's job changes shape. A lot less chasing and collating. A lot more deciding, coaching and thinking about what's next.

For some leaders that's uncomfortable at first, because a surprising amount of management has quietly become information management. Being the person who knows what's going on is a kind of power, and when everyone can see the same picture, that power goes. What replaces it is better: being the person who makes good calls, quickly, with the whole team behind them.

Why I'd keep the line where it is

AI is going to keep getting better, and I'm sure there'll be pressure to move the line, to let the machine decide a little more each year. I think that'd be a mistake for the companies we work with. Not because the AI won't be clever enough, but because a company is a group of people who've chosen to work towards something together, and the choosing is the part that makes it theirs.

So our vision is a world where every company is aligned in real time, and our mission is to build an operating model where AI does the asking, gathering and admin, and people do the thinking and deciding. Those two halves aren't a compromise. They're the whole design.

The next belief, A loop, not a plan, is about the rhythm that keeps those decisions coming.