Practical, candid writing on operating-model design, engineered operations, and how high-performing teams stay efficient at scale. No fluff — just the operational thinking we use every day with clients.
When demand grows you can hire more people, or redesign how the work gets done so the same team handles more — up to the point where cutting too far breaks quality. A plain-language guide to getting more done per person: a worked example, an interactive model in real units, and a five-question self-check, grounded in OECD, Stanford HAI and NBER evidence.
Every operation gets mapped by cost and by the org chart. Neither of those maps tells you the one thing that decides your worst day: how far a single failure spreads. And the efficiency moves everyone rewards — consolidating onto one provider, one system, one source of truth, and stripping out the slack — quietly maximise that spread. This is a way to see the blast radius your cost model hides, to price the trade-off honestly, and to shrink it at the few nodes that matter without giving up the efficiency everywhere else.
The bottom of the career ladder is thinning — early-career hiring is falling fastest exactly where AI is strongest. The usual response is to defend the entry-level job. That defends the wrong thing. The rungs were never the point; they were a slow way to deliver something else — exposure to real decisions. Keep that, deliver it deliberately, and the path to expertise can be rebuilt faster than the ladder ever managed.
As we hand work to agents, the bottleneck isn't the agent — it's that our handoffs were built for human eyes. The move isn't to make everything agentic; it's to reshape each handoff for its real next reader — machine or human — which is where the value, the reliability, and the token economics actually turn.
If compute keeps getting cheaper, does re-engineering your workflows stop being worth it — the way repairing electronics stopped being worth it? Sometimes, yes. But the deciding variables are volume, durability, ownership and stakes, not the token price. Cheaper inference moves the crossover point; it never removes it.
Every capability decision gets framed as build versus buy. It's the wrong question. The real variable is how much of a capability you must own versus simply access — a spectrum, not a switch — and the most valuable position is often the one the binary hides: the blend.
Automation is very good at the routine 80% of back-office work. But the routine was never where the cost, risk, or difficulty lived. As the routine disappears, everything that's left is exceptions — and the organisations that win are the ones that industrialise the exception, not the routine.
Most organisations buy tools to speed up steps, then wonder why the end-to-end result barely moves. The value doesn't leak inside the steps — it leaks in the seams between them. And because that loss compounds, the tax is far larger than any single step is slow.
For decades, the T-shaped professional was depth of execution plus breadth of knowledge. AI inverts this model: the vertical bar is now judgment, and the horizontal bar is orchestration. Execution depth is commoditised. Judgment depth is what's scarce.
Automate the easy 80% of a job and you don't get an easier job — you get a harder one. The work that remains is the work that resisted automation: judgment, exceptions, coordination. Here's why efficiency concentrates difficulty, and what it means for how you design roles and teams.
We help leaders design and operate the integrated operating model described in these articles. Start with a discovery conversation.
Schedule Discovery Call