How I use AI as the co-CEO of a remote company

I'm the co-CEO of a completely remote company. 

Sam and I have never shared an office, and most of the time we're not even on the same continent; our team works across more time zones than I can reliably hold in my head.

We run the company almost entirely in writing: product discussions, support escalations, roadmap debates, executive decisions, and the long tail of context that builds up when async work is your operating system.

This is how I use AI to manage in a fully remote environment - without cutting the human from the loop…

The bargain

You make a bargain when you build a company around written communication. Decisions don't disappear into meeting notes or long-forgotten Kanban cards; someone can search for the topic from three years ago where people wrote down the original reasoning, and the company develops a verifiable shared memory.

That's what you get - and it's an incredible benefit.

But that memory also gets harder to manage. In a traditional organisation, leaders gather context by sitting in calls and asking someone to repeat what they've already explained twice. We read and write instead - which can take longer.

For a while, I was losing to that bottleneck.

I'd come in on a Monday, spend an hour catching up on roadmap topics and feature debates, and still not know whether any of it actually needed me. Then someone would ping me for a decision, and I'd read backwards through a tangled thread, hunting for the actual question, arriving at it already mentally exhausted.

People communicate like people. They don't naturally write, "These are the relevant facts, this is the decision I need, and this is my recommendation." They work through the problem in public - which is the right way to do the work. But when the issue gets escalated to me, I have to separate the working conversation from the decision point.

What I actually use it for

Where AI first became useful: it lets me gather context without costing other people their time.

I run two Monday queries. One gives me the product and engineering picture: what shipped, what's moving on the roadmap, and which decisions may need executive attention. The other gives me the customer picture: which enterprise accounts are healthy, and where renewals or legal negotiations are getting complicated. If we still need a call after that, we spend it going one level deeper.

I have the most community management experience at Discourse, and people want my judgement on which features matter most to community managers, but don't always know how (or when) to ask. Now I see those discussions while the idea is still forming, rather than after the decision has gathered momentum.

Sam knows as much about running Discourse as I do, but much more about engineering and product; for a long time that imbalance bothered me. When he needed something from me, he had to explain the whole chain of reasoning before I could understand the question. I felt like a hindrance in my own company. Now I come in with more of the context already loaded - and we get to the hard part faster.

An index, not an authority

I don't make decisions based on AI summaries. Ever. If a bot flags an account as urgent, I open the topic, read what the customer actually wrote, and talk to the customer success manager. An index tells me where to look, but it doesn't tell me what to think.

AI has a terrible habit of sounding more certain than it has any right to be, stating assumptions as findings, with a real fondness for vague psychological theatre. "Cautious optimism." "Emerging tension." "Strategic alignment." The language alone makes me grind my teeth. I don't want opinions, I want facts: what happened, where it happened, and a link to the underlying conversation.

AI is a lossy summariser - like a JPEG. It introduces artifacts, but you get the picture. You just have to know that going in.

People are off limits

A hard rule: I don't use AI to assess people. A manager can ask for a summary of what deals someone closed; a real person still has to verify the work. Asking AI whether someone is strategic, resistant, checked out, or secretly hostile is automated mind-reading. I don't trust it, and I don't think it helps an organisation.

I have used it on myself, with my coach, to look at my own communication patterns. It gives me a mirror, and my coach helps me argue with the mirror. But I wouldn't point it at a colleague who hadn't consented.

The fear around AI at work makes sense to me. The worry is often specific: that working in public becomes raw material for a hidden assessment, that every sentence you write can feed a confident little paragraph about your attitude. Summarising a public product topic is one thing; running sentiment analysis over a team's internal discussion is another; asking a model to assess an individual goes further again. The closer the tool gets to a person, the more careful managers need to be.

We tested that line

A few years ago, I became interested in the difference between organisational culture and climate. Culture is the deeper system: values, norms, habits, stories. Climate is more immediate. How does it feel here right now? Are people stressed? Defensive? Afraid?

Around the same time, anonymous survey feedback arrived sharp enough ("Hawk and Sam aren't real CEOs") to make me wonder what we were missing. I wanted earlier signals, and I proposed AI sentiment analysis to find them. The team pushed back because it felt like surveillance; they were right. A leader can know that people are having a contentious discussion and go read it. A leader does something else when they tell the company that a model thinks the organisation is "stressed." Can an organisation handle hearing that? Maybe. Can leadership know it and pretend otherwise? Probably not.

AutoCAD, again

My current position is practical rather than ideological. AI helps when it finds the conversation I should read, or flags three customer accounts that need attention. It becomes dangerous when managers turn it into a personality test, a performance review, a motive generator, or a substitute for managing. Use AI, but don't let it launder your judgement...

The popular language around AI - either salvation or extinction - doesn't help, when the more immediate question is, can this tool make work better?

I graduated architecture school around the time AutoCAD reached mainstream adoption, and people panicked then too. They said it would take architects' jobs, that clients would make their own plans and nobody would need us. Then the architects actually used it. It saved hundreds of late nights drafting by hand and made iteration cheaper. Some crafts changed, and new crafts appeared. AI feels like an extension of that experience; it makes some things easier, some sloppier, some newly possible - and we have to work out which is which without losing our minds.

I don't want AI in every corner of my life. I use it when I have a problem, and I close the tab when I don't.

At work, the problem is context: there's just more of it than I can read, than any human can read - and AI helps me find the parts that need my attention without asking other people to spend their morning catching me up.

But the responsibility stays with me.

The machine can point; I still have to read.

It can surface a pattern; I still have to ask whether the pattern is real, and whether using it will help or harm the people doing the work. And it will never - and should never - become a hidden manager, a psychological profiler, or a substitute for trust. 

Not at Discourse, and not anywhere else.