Show Notes
Knowing where not to use AI matters as much as knowing where to use it – and the person saying so builds AI systems for agencies for a living. In this episode, Gareth puts Callum in the hot seat on why he trusts AI less the more he understands it, why the work agencies hand to AI first is exactly the work that used to turn juniors into seniors, and why a model that has swallowed an entire framework still can’t do the one thing an agency owner is actually paid for: read the room and make the call.
In this episode
- Why WPP and S4 cutting headcount doesn’t mean AI is taking agency jobs – and why Callum’s aim is more output from the same people, not fewer people
- Why Callum trusts AI less the more he learns about it: it’s a prediction machine that is built to guess rather than say “I don’t know”, and Gareth’s take that it doesn’t hallucinate at all – it can’t think, so it just makes mistakes
- Why anything short of 100% reliability can’t be left to run on its own (the credit card chip that barely changed in 40 years) – and what that means for agentic tools
- The apprenticeship paradox: junior engine-room work is what AI does best, so agencies hand it over first – but it’s also how judgment has been built for as long as there have been trades, so where does the next senior come from?
- Learn through AI, not from AI: why the senior people with the judgment need to be the ones configuring the systems – which means scoping workflows that currently live in people’s heads in painful detail
- Why the felt half of a decision can’t be delegated: 200,000 years of reading other humans, mirror neurons, and why AI can quote every psychology study but has never felt stress, fear or warmth
- The calculator test, and what it says about the STANDOUT system: AI can recite Gareth’s entire framework, but the value was never the framework – it’s the 30 years of decisions that produced it, and the person applying it now
