We're automating the apprenticeship out of work. As agents take on execution, people shift towards reviewing outputs and exercising judgement. But judgement comes from years of doing the work, making mistakes and learning what good looks like. If we remove entry-level roles, where will tomorrow's experts come from? This is why junior hiring matters more, not less, in an AI-enabled organisation. Companies risk trading long-term expertise for short-term productivity. The answer isn't to preserve junior tasks unchanged, but to redesign them so people learn alongside agents. Otherwise, we're building organisations that depend on expert judgement while dismantling the pipeline that produces it.

Three in four CEOs are spending more on technology, while almost half can't show a return on what they've already spent. More budget won't fix that on its own. Gartner's answer matches what we see with clients. Redesign workflows around intelligence, measure outcomes over activity, and govern digital labour and workforce change together. For me, that last point matters most. In our experience, change without the build rarely sticks, and the build without the people change rarely pays back. Done together, each step makes the next one easier and the value adds up. If your board report on AI still counts licences and pilots, I'd switch it to unit costs, decision speed and customer outcomes.

Harnesses are evolving. They used to mean coding tools and desktop assistants. Now they’re where personal, team and managed agents do the business’s work, so every serious use case will run through them, from supply chain to finance. We’ve been building our own for internal use, and Agora looked familiar: an orchestrator working for the team, personal agents, your choice of model and data kept in the EU. A small team built that, so expect the same patterns to show up across many products. For clients, the question becomes whether to build, buy or wait.

The story is being read as “AI cuts 5,000 jobs”. The chief executive’s own words say something else: “AI is the icing on the cake, not the cake.” Three years of rebuilding the core came first, and the savings rest on that groundwork. Fair enough, but the rebuild is exactly where AI earns its keep. Untangling legacy processes and migrating systems nobody fully understands is now work AI does in months rather than years. Use it to build the cake faster, then put the icing on.