The First Hundred Hours

Whenever I sit down with a founder who has decided to get serious about AI, the same conversation happens. They want to know what tool to buy, which use case to start with, and how quickly they can have something running.

I have started forcing a different conversation. The question is not what tool or what use case. The question is: how should the next hundred hours of your time be spent? Because the next hundred hours are the entire game. If you spend them well, the rest of the year compounds. If you spend them on the wrong things, you will be back in the same conversation in twelve months, asking the same questions with slightly different jargon.

What follows is the allocation I now recommend, and the reasoning behind each block.

Before I tell you what I think the answer is, it's worth being honest about what most founders default to.

The default first-hundred-hours plan is: forty hours of training (workshops, courses, ChatGPT tutorials), forty hours of evaluating tools (sitting through demos, comparing pricing, asking peers), and twenty hours of trying to find a use case. The last twenty are usually the worst spent of all, because they are spent looking for problems that match the tools the team has just learned about — the wrong direction entirely.

At the end of those hundred hours, the team is more aware of AI, has shortlisted two or three vendors, has built nothing, and has changed nothing about how the business operates. Net delta: zero.

Here is the allocation I think actually works. It is unusual enough that I have to defend it every time I propose it.

• Twenty hours on workflow mapping. Watch your highest-leverage people work. Identify every repeated task. Group them. Score them by frequency and pain.
• Twenty hours on building one boring thing. Pick the highest-scoring repeated task. Build the simplest possible AI assist for it. Ship it to one person.
• Twenty hours on rollout. Use it daily. Watch them use it. Adjust. Measure the actual time saved against your estimate.
• Twenty hours on writing things down. Standards. Voice. What good looks like. Acceptance criteria. The thing the AI needs to know about your business that lives in your head.
• Twenty hours on the second build. Now that you know how to deploy, build the second thing. Notice it is twice as fast as the first.

Notice what is missing from this plan. Vendor evaluation. Training courses. Strategy decks. Generic AI literacy. None of it is in the first hundred hours, because none of it produces capability.

Each block builds on the previous one, and each one produces something durable. The workflow map is a permanent artefact you'll use for years. The first build teaches the team what deployment actually looks like. The rollout reveals the gap between estimated and actual time saved — which is almost always larger than the team expected, in unexpected directions. The writing-down phase produces the taste artefacts that every future AI build will rely on. And the second build proves the team can do it again, faster.

Compare that to the default plan, which produces awareness, slides, and a sense of motion. The default plan looks productive. This plan is productive.

The thing nobody warns you about is that this plan is uncomfortable for the people involved. There is no training course at the start to make everyone feel competent. There is no vendor slide deck to make the choice feel rigorous. There is just the founder, the team, the work, and the awkward early attempts.

If you can sit through the awkwardness — and pay your team to sit through it with you — you arrive at the end of a hundred hours with something almost nobody else in your market has. Real, deployed, used, measured AI capability. Two workflows shipped. A team that knows what shipping looks like. A founder who can write a brief without hand-waving.

From there, the next hundred hours go three times faster. The hundred after that, ten times. That is the compounding. That is the whole point.