Topic
AI, and what it’s actually doing
Most writing about AI is either a sales pitch or a eulogy. This is neither. It’s about what the technology is actually doing to industries and to the people who work in them: who wins, who gets quietly replaced, and what it means for the teams in the middle.
The future is being built by the people coding it, for the people coding it. The rest of us need to catch up before we’re the ones being automated.
Playbooks on this topic
- Half of agencies say they use AI agents. Almost none can prove it.Half of US agencies say they run AI agents. Only 3 of the 20 I studied can show a client-verified result. Part 1 of 3 on what agencies are actually doing.
- What the agencies that can prove it actually builtThe independent agencies with real results share a stack, a design pattern and one governance rule. Part 2 of 3: the 10 builds worth copying.
- The two agents I built, and the gaps nobody has filledAn audit engine that runs 120 checks on Google and Meta accounts, a Slack agent that reads release notes, and 5 agents no agency has published. Part 3 of 3.
- Where AI earns its place in ecommerce contentMost product pages on a luxury site get no video, no model and no campaign shot. How to tier the catalogue and point AI at the pages production never reaches.
Articles
IKEA retrained the people AI replaced, and they became a €1.25bn sales channel
When IKEA's AI assistant Billie took over routine calls, reskilling 8,500 call centre staff as design advisers built a €1.25bn sales channel.AI gets you most of the way. The last part is still yours
Epoch AI tested the best models on its own staff's real work. They met 65% of the standard. That's a useful target, not a reason to hand over the rest.Cheaper isn't a reason to switch AI models
OpenAI is cutting prices by up to 90% to win Anthropic's customers. For an agency, the model should be chosen by the job it does, not by this quarter's invoice.AI pays back in speed before it pays in profit
Apollo's chief economist finds AI has lifted profit only for the 7 tech giants. In agencies the payback is real, but it shows up as speed, not margin.The AI pilot that never lands
Most agencies are running AI pilots that will never become anything else. The problem isn't the technology. Nobody was ever asked to stop doing the old job.The $29 cap on the tool that was working
Accenture has capped staff Claude spend at $29 a month. Meta and Microsoft are rationing too. The firms selling AI transformation hit the bit nobody budgets.The robo boss now needs a witness
California has banned sacking people on an algorithm alone. A human has to check the work first. Most agencies already run a robo boss: the utilisation report.Questions I get asked
- Is AI going to replace marketing jobs?
- Some of them, and sooner at the junior end than most agencies admit. The work that goes first is the work nobody wrote down as a decision: reporting, first drafts, trafficking. The work that stays is choosing what to say to whom, and that was always the scarce bit.
- How should an agency roll out AI tools?
- Decide what stops being done if it works, put a date on it, and go and sit with the people using the tools most. A pilot with no definition of success becomes a hobby with a budget line. A flat cost cap before anyone has measured output stops the only experiment that was working.
- Do marketing agencies really use AI agents?
- Half of US agencies say so. Of the 20 published cases I checked, 3 can show a client-verified result, and all 3 were co-written with the vendor. The ones that work do 3 jobs: reporting, media buying and creative inside a governed loop, with a person approving before anything spends or sends.
- Should clients be told when AI was used on their work?
- Yes, before they find out. The problem was never the tool, it is the person who discovers it second-hand. Say what was generated, what was checked and by whom, and most clients care less than you fear. Hide it and it becomes the whole story.