EPISODE 8: Where NOT to use AI in your agency.

Ask anyone selling AI into agencies whether it’s coming for people’s jobs, and you’ll get the same reassuring line every time. Ask them what AI still can’t do - even after swallowing an entire book, framework and philosophy whole - and the answers get a lot more interesting.In episode 8, Gareth puts Callum in the hot seat to unpack why 300,000 years of human wiring can’t be replicated from a training set, why the right analogy for AI isn’t a robot takeover but a calculator, and the paradox sitting underneath every agency quietly handing its junior-level work to a machine: if AI does the reps that used to build an expert, where does the next expert actually come from?It’s a candid, occasionally self-implicating conversation (not least because Callum promotes AI adoption for a living, and still doesn’t want to be replaced either).In this episodeWhy Callum trusts AI less the more he learns about it. He’s a genuine advocate, blown away by its utility, and yet the more he understands how it actually works, the less he trusts it unsupervised. The simplest illustration: AI hallucinates because it’s mathematically predicting the next most likely word, not looking anything up — and it’s trained to guess rather than admit it doesn’t know. Even 99% certainty is a bad number for a business to run on.The apprenticeship paradox. Judgment has always been built through reps - the boring calls, the repetitive spreadsheet trawling, the meetings nobody wanted to sit in. That’s precisely the category of work AI is best at, and precisely why agencies hand it over first. Which raises the real structural question: if a machine is doing the reps that used to turn a junior into a senior, what replaces the apprenticeship as the way judgment gets built at all?Through AI, not from AI. Gareth and Callum’s answer isn’t to protect the boring work - it’s to make sure senior team leaders build AI systems that actually contain their knowledge and workflow, so juniors are learning the system a senior built, filtered through AI, rather than learning from AI directly with nobody senior in the loop.Why the felt half of a decision can’t be delegated. AI can read the facts of a situation but not the room. Two people can sense tension, know when to soften something, feel what’s coming.Calculators didn’t retire mathematicians. Callum’s pushback to the obvious “but shouldn’t AI keep learning too?” challenge: a calculator changed what a mathematician spent their time on, it didn’t replace the maths. But you still need to know the maths to use the calculator well - which is exactly why schools taught mental arithmetic and calculator use as two separate skills. Same split now applies to specialism and AI.The STANDOUT or Die test, turned on its own author. Feed the entire STANDOUT framework into an AI system and it will recite it back fluently - word for word, on demand. It has never sat in a room, watched a face go quiet, or felt a deal nearly die. A framework in an AI system is a starting point for consistency. It is never a substitute for someone who has actually been in the room.The line they land on. AI does the arithmetic - finds the hours, calculates the cost, spots the pattern. That part isn’t up for debate; it’s faster and better at it than any person. Humans still do the judgment: what that number means for this client, this relationship, this risk, right now.What Callum wants listeners to do this week. Write down what your business believes and doesn’t, what it does and doesn’t do - then use AI to pressure-test decisions against that. Never to make the decision itself. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit thestandoutagencypodcast.substack.com