jefford

problem hacker #10

The Platform Paradox: When AI Eats Its Own Lunch

Mark Jefford ·

The everyday headache

Buying today’s AI tools feels like buying a washing machine that gets replaced the very next week. You have barely plugged it in before the shop calls to say a quieter, faster model just arrived. Un-install, re-install, new plumbing, more money. Figure out how to work the thing. All in all, fairly unnecessary pain.

That is how big companies feel right now. They sign a contract, then the supplier shows a newer “beta” version, and the first deal already looks out of date.

A quick story

Maya runs technology at a large insurer.

  1. Last month she bought an AI system that reads claim forms.
  2. Last week a rival vendor launched a shiny assistant that not only reads but also writes the customer email reply. Maya’s an early adopter, and she always needs the best tech, so she jumped onboard fast.
  3. Yesterday the CFO asked why they were now paying for both. “Do we cancel the first contract?”.

Maya’s answer: “Give me the weekend. Everything might change by Monday.”

Why the chaos keeps coming

At Snowflake Summit, OpenAI’s Sam Altman called today’s AI an intern that still needs supervision and claimed next year the same software could “discover new knowledge” on its own.

The very people selling the tools admit the roadmap is a moving target, so customers cling to the biggest, richest brand because it feels safe. That is the platform paradox: constant upgrades nudge us toward the one name we already know.

The Problem Hack: How to keep your options open and remain ahead of the curve

  • Separate the front and back. Let the user interface stay the same even if you swap the engine underneath.
  • Short contracts only. Treat AI deals like renting a flat, not buying a house. A one-year lease hurts less if you need to move.
  • A small “try anything” pot. Ring-fence ten percent of your tech budget for experiments that need no committee approval.
  • Ask about goodbye costs up front. If a vendor makes it hard to leave, push back or walk away.

A 30-day plan anyone can follow

  1. List every place you rely on one AI supplier.
  2. Spin up a sandbox. Give a small team permission to play with two alternative tools.
  3. Measure both “time to value” and “time to quit.” How fast does each tool show results, and how hard is it to unplug?
  4. Show finance the numbers in plain pounds and pence. Compare the cost of staying stuck with the cost of switching.
  5. Add a renewal kill switch. If a vendor slips by more than three months, the budget moves automatically to your sandbox.

Three prompts worth copying

  1. “Compare prices for Anthropic, OpenAI and Cohere to process five million documents. Show where each breaks even.”
  2. “Rewrite this supplier’s roadmap as a risk list for our audit committee. Rank by chance and impact.”
  3. “Draft a two-page board update explaining why we are running two pilots instead of signing a three-year deal.”

Buying AI tools today is like ordering clothes off the internet. You don’t know if they’ll fit, or if they look anything like the photo. The smart move? Keep the tag on and check the returns policy before you commit.

Thoughts, horror stories or survival tips? Hit reply or tag me. Let’s swap notes and stay sane.