In partnership with
We missed a week, so you get two guests — and by accident they've handed us the sharpest argument of the year.
Ask an Oxford economist why big projects fail and he'll tell you it's bias: humans are hopeless at estimating, and AI can correct us. Ask the man advising the UN on AI policy and he'll tell you the models doing the correcting carry biases of their own — and almost nobody in the room got to vote on them. Both are right. That's the problem.
🎧 Hit play on Atif Ansar — Oxford Expert Reveals Why Projects Fail & How AI Changes Everything (Ep 120) and Dr Craig Ramlal & Daniel Goitia — UN AI Advisor Reveals the Future of Artificial Intelligence (Ep 119). Back to back, they're about 90 minutes well spent.
Atif Ansar has spent a career at Oxford working out why projects blow their budgets, and his answer is uncomfortable: it usually isn't the engineering. It's the psychology and the politics. Optimism bias in the estimate, strategic misrepresentation in the business case, and an organisational culture that punishes the person who says the number is wrong. His fix is reference class forecasting — stop asking "what will this project cost?" and start asking "what did the last hundred like it actually cost?" — and he thinks AI is very good at exactly that kind of de-biasing decision support. His work at Foresight Works is built on it.
He's also blunt about the limits. AI can pattern-match a schedule; it cannot hold the implicit knowledge of the person who has built the thing before. Codify what you can, he argues, but don't kid yourself that a complex delivery system can be fully written down. His bet for 2034 involves a lot more autonomy on site and a lot smaller teams running it — pods, not pyramids.
Which is where Dr Craig Ramlal and Daniel Goitia of the University of the West Indies come in with the awkward follow-up question. If the model is the referee, who trained the referee? They make the case for data sovereignty — that a region which doesn't own its data, or shape the standards, ends up importing someone else's assumptions and calling them objective. The Caribbean is their example; it is not remotely the only one. Their week is spent on the harder, less glamorous work: harmonising policy across jurisdictions, building local capacity, and arguing that AI safety and interoperability are governance problems long before they're technical ones.
Put the two together and the through-line is hard to miss. Ansar wants AI to strip the bias out of your project decisions. Ramlal and Goitia want you to remember that AI is not a neutral instrument — it is a set of choices made by people who were not in your room and did not have your interests in mind. Use it to de-bias, by all means. Just don't mistake it for the absence of bias.
Why it matters: the profession is rushing to let models grade its forecasts, and the governance of those models is being settled right now — mostly without us. If you adopt the tool and ignore the question of who built it, you haven't removed bias from your projects. You've outsourced it.
👉 One thing to do this week: take your next estimate and ask what a reference class of similar completed projects actually says. Then ask which dataset your AI tool is quietly using to answer that same question.
Links and Stuff
Ep 120 — Atif Ansar (Oxford; Foresight Works) — LinkedIn · foresight.works
Ep 119 — Dr Craig Ramlal (University of the West Indies) — LinkedIn
Ep 119 — Daniel Goitia (University of the West Indies) — LinkedIn
Listen on
Subscribe to the Project Flux newsletter — the week's hottest AI and project-delivery news in your inbox every Monday.
Reply to everything. Edit nothing.
Your inbox is full. Slack is piling up. Client messages need a response yesterday. Typing thoughtful replies to all of it takes hours you don't have.
Wispr Flow turns your voice into clean, professional text you can send the moment you stop talking. Speak like you would to a colleague — tangents and all — and get polished output. Emails, Slack, LinkedIn, WhatsApp, whatever's open.
89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Works on Mac, Windows, and iPhone.
Project Flux is independently owned, funded, and operated by its founder. Any external references or features are for community benefit only and do not imply endorsement, control, or ownership.