In the early days of Airspeed, one of our advisors, Pablo Mendes, introduced me to Georg Bauser. Georg had built Accoleo, sold it to Airbnb, and then helped build Airbnb’s international expansion.

He gave me a piece of advice that I have returned to many times since: treat every hour of your time as if it costs $1,000.

If something can be delegated for less than that, delegate it.

It can sound like one of those neat pieces of ivory-tower advice that is much easier to repeat than to follow. Early-stage founders cannot simply delegate everything. You are short on money, the company depends on knowledge that mostly lives in your head, and many jobs are difficult to hand over without spending more time explaining them than doing them yourself.

But over the long term, I have found Georg’s advice pretty spot on.

The most expensive thing a founder can do is spend most of the week on work that someone, or now something, else could do. Chasing people for updates. Preparing for meetings. Looking through Slack for the one conversation that matters. Checking whether a promise was followed up. Pulling information together from five systems. Reconstructing why a decision was made three months ago. Individually, these tasks feel small. Collectively, they consume a surprising amount of the founder’s attention.

This is why I find the current conversation about AI harnesses so interesting. A harness is the layer around a general-purpose AI model that gives it context, memory, rules, tools, and recurring jobs. And it is now remarkably easy to build one. You do not need a large engineering team. A Mac mini and a ChatGPT or Claude subscription will get you surprisingly far. In my experience, ChatGPT/Codex is better suited to this kind of always-on, scheduled workflow, so that is what I would recommend.

I adapted Diogo Neves’s open-source Glide into my own system, which I call Glider. The company workspace inside it is Airspeed HQ. At its core, Airspeed HQ is a private repository of simple Markdown files containing the context an executive assistant needs to be useful: our company strategy, metrics, customers, current priorities, my working preferences, previous decisions, open questions, contradictions, and promises that should not disappear. It organizes this across the main areas of the company, including product, GTM, finance, hiring, operations, and investors. It can also read from connected systems such as Airspeed (obviously!), email, Slack, the CRM, the calendar, and company documents. Reading and drafting are allowed, while anything that changes the outside world still requires my approval.

Around that context sit a set of recurring routines. Airspeed HQ runs daily founder check-ins, monitors follow-through, reviews operations, scans relevant market developments, prepares a weekly CEO review, and flags when our behavior is drifting away from our stated priorities. It does not try to give me another dashboard containing everything. The point is to inspect a lot and surface very little: the decision I am avoiding, the commitment at risk, the contradiction in our plan, or the one question that deserves my attention today. I also program it through normal conversation. When it gets something wrong, I correct it, and that feedback can become a durable instruction, checklist, or piece of company context. Over time, the system becomes less like a chatbot and more like an assistant that understands how I and the company operate.

It is far from perfect. Connectors fail. Context becomes stale. Agents occasionally misunderstand instructions. Giving software access to company systems also requires clear boundaries and a healthy amount of caution.

But it already handles a surprising range of small and large tasks. More importantly, it preserves my attention for the work where the difference between me doing it and someone else doing it is genuinely significant.

Georg’s $1,000 rule used to imply hiring or contracting. AI changes the economics. Founders can now delegate work that previously would have been too fragmented, too context-heavy, or too expensive to hand to another person.

If you’re a founder, try it. Clone Diogo’s repo, connect your tools, and adapt it to how you run your company.

It won’t make every hour of your time worth $1,000. But it might stop you spending those hours on $20 tasks.

And if I can help you set it up, let me know.