All of my long-form thoughts on AI, programming, product development, and more, collected in chronological order.
Hilton Dubai Jumeirah’s AI waste bin pilot with UNEP West Asia shows a quieter product lesson: sometimes AI should not make the original decision, but create a reliable memory of where that decision failed. The useful product is not the forecast; it is the feedback loop operators actually trust.
Synthetic customers will be useful only if businesses can keep them from contaminating bookings, ledgers, staff records, CRM histories, and the trust of the people who have to serve real demand.
AI will not just help sellers write better proposals. It will give ordinary buyers a cheap procurement department that audits every line item, asks awkward questions, and changes what a trustworthy quote looks like.
The next useful AI product may not look like a copilot. It will watch for wasted capacity — a cancelled appointment, an unanswered call, an empty return truck, a dead hour in the diary — and quietly turn it back into revenue.
Many post-purchase promises are priced as if customers will forget, lose the receipt, or give up. Cheap AI memory changes that: the claim rate becomes a product and operations problem, not just a support problem.
As AI vision and reasoning get cheap, the valuable automation may not be doing the work but proving the work happened. The next quiet product surface is the proof layer: photos, timestamps, signatures, comparisons, and exception trails that decide invoices, refunds, claims, and blame.
Many businesses quietly rely on customer effort as a filter: the people who wait, chase, call back, or argue are the ones who get resolution. AI agents change the economics by making persistence cheap, forcing companies to design for clear delegated resolution instead of human exhaustion.
When AI agents choose where to buy, book, return, cancel, or substitute, the policy page stops being legal furniture. Returns, delivery cutoffs, cancellation windows, warranties, and exception rules become part of the product itself.
As voice and browser agents get cheap, businesses will stop testing only whether systems work and start testing whether promises survive contact with real workflows. The next useful AI layer may be a quiet patrol that catches the gap between the website, the chatbot, the receptionist, the checkout, and the policy binder.
The gap between the boardroom view of AI deployment and the daily reality on the floor is creating a new category of enterprise failure: tools that look adopted but deliver nothing.