Anne Cantera owns a t-shirt that says sprinkle some AI on it is not an AI strategy. She was not joking when she brought it up, and by the end of our conversation I understood why she needs it in wearable form. She builds multi-agent systems, trains businesses on AI, and left enterprise on principle. Her diagnosis of why projects fail is the most useful thing I have heard on the subject.
The question itself is wrong
I asked her why projects collapse before they start. She did not hesitate. You are asking the wrong question, and you are starting in the middle instead of at the beginning.
The right opening question is not how do we add AI to this. It is what problem are we solving. Her example is deliberately unglamorous: if you answer the phone four hundred times a day to tell people your opening hours, that is a problem with an obvious AI shape to it. Start there, then match the problem to the solution.
What happens instead is the reverse. Somebody has a technology they want to play with, so that becomes the starting point. As she put it, we have abandoned design. We stopped digging into the problem before choosing the answer, and it is happening everywhere.
"I have a hammer, everything looks like a nail. That's not that and that's what's happening, and that's why things are not working out well. We've just abandoned design."Anne Cantera
Three beliefs that break the budget
She named three assumptions that leadership tends to hold simultaneously, and all three are wrong.
It is not cheap. It is not easy. And it is not set it and forget it. You have to monitor and improve it over time, which means somebody owns it after launch, which means it is a running cost and not a project cost.
She is watching the bill arrive in public. Post after post from teams blowing their token budgets, and leadership discovering that AI can be more expensive than the humans it was supposed to replace. Her image for the underlying problem stayed with me: leadership handed people a Ferrari and nobody has a driving licence.
Her example was a design team that burned through tokens meant to last until Friday by Tuesday. When she asked whether they were doing spec driven development or just prompting, the answer was just prompting. The tool was not the problem.
Readiness is more layers than people expect
Her advice to small and medium businesses is four words: do not DIY your AI. She has been brought onto enterprise projects that were designed by people with no AI experience, and it does not go well.
Readiness has layers. People. Tools. Processes. Legacy technology. She calls it a laundry list, and the point is that if you are not experienced in the space you will miss several of them, and your project will either stall or ship something worse than nothing.
Her economics on hiring help are blunt and, I think, correct. It is far cheaper to pay a consultant to put you on the right path than to skip the fee and go off the rails. Spend it now, save it later.
Governance on day one, not after the incident
She joined a startup recently and by day two was arguing that governance had to be one of the first things they took on. She says this gets her eye rolls. Then she brings up the truck.
A developer talked a Chevrolet dealership chatbot into agreeing to sell him a fifty thousand dollar truck for a dollar. His point was made. Her question is the one that matters commercially: who is the human on the hook when that happens? Naming that person in advance is itself part of governance.
Governance is not one thing either. It covers model behaviour, and it covers what your employees are allowed to do, and it should sit across teams rather than on the founder alone. It also has to keep changing. If something goes sideways and your policy never addresses it, the policy needs updating.
What I would take from this
Write down the problem before anyone names a tool. Budget for the running cost, not just the build. Get someone experienced to check your readiness before you commit, because the expensive mistakes are made early and discovered late. Decide who is accountable when the system does something you did not intend, and write that down too.
And treat training as part of the purchase rather than something people pick up on their own. Handing capable people a powerful tool with no instruction is how you get a token bill on Tuesday and nothing to show for it.
We also got into emotional debt, agents, and what she thinks the workplace will look like in five years, which is on Zaptime.