There is a lot of noise about AI strategy and very little honest talk about what AI changes on an ordinary Tuesday — inside a real company, with real people and real deadlines. Running a group of more than fifteen companies, I get to watch that reality up close.
The first thing I learned is that AI does not replace judgment; it removes the drudgery that used to crowd judgment out. When a support agent no longer spends the morning drafting the same five replies, they spend it on the one conversation that actually needs a human. When an analyst no longer stitches spreadsheets together by hand, they spend the time asking better questions of the data. The value is not the output the model produces — it is the attention it gives back to people.
Where it actually helps
In our companies the biggest gains are unglamorous: faster first drafts, faster research, faster summarisation of long threads, faster onboarding of new hires into old systems. None of it makes a headline. All of it compounds.
Where it quietly fails
AI fails where we let it make decisions no one is accountable for. A model that drafts is a gift; a model that decides without an owner is a liability. We keep a simple rule: every AI-assisted output still has a human name attached to it.
The companies that will pull ahead are not the ones with the most impressive demos. They are the ones that rebuilt a few real workflows so that intelligence is assumed, then held the discipline to keep a human accountable at the end of each one.