Edge Bites - Episode 14
Fri Dec 19 2025
The final Edge Bites of 2025 — and what a week to end on.
This episode: the rules of business are being rewritten across marketing, assets, and operations all at once. Amazon and Meta are reorganising their AI divisions (again). The Chief AI Officer is having its moment — but will it end like the Chief Digital Officer experiment? UK banks are trialling AI agents that move your money autonomously. And the gap between "companies citing AI layoffs" and "economists seeing no macro impact" is widening.
Case Study: Klarna fired 700 people, replaced them with AI, bragged about it — then quietly started hiring humans back. Gary Marcus calls it "The Klarna Effect." The lesson? They found the boundary between AI capability and customer expectation by crashing into it.
Cutting Edge: Google, DeepMind, and MIT tested 180 multi-agent configurations. The verdict? "More agents is all you need" is wrong. On sequential tasks, every multi-agent system made things worse — some by 70%.
Tool of the Week: Arthur Bench — an open-source framework to benchmark AI models on your actual data. Stop vibe-checking. Start measuring.
Thanks for listening in 2025. We're back mid-January 2026.
More
The final Edge Bites of 2025 — and what a week to end on. This episode: the rules of business are being rewritten across marketing, assets, and operations all at once. Amazon and Meta are reorganising their AI divisions (again). The Chief AI Officer is having its moment — but will it end like the Chief Digital Officer experiment? UK banks are trialling AI agents that move your money autonomously. And the gap between "companies citing AI layoffs" and "economists seeing no macro impact" is widening. Case Study: Klarna fired 700 people, replaced them with AI, bragged about it — then quietly started hiring humans back. Gary Marcus calls it "The Klarna Effect." The lesson? They found the boundary between AI capability and customer expectation by crashing into it. Cutting Edge: Google, DeepMind, and MIT tested 180 multi-agent configurations. The verdict? "More agents is all you need" is wrong. On sequential tasks, every multi-agent system made things worse — some by 70%. Tool of the Week: Arthur Bench — an open-source framework to benchmark AI models on your actual data. Stop vibe-checking. Start measuring. Thanks for listening in 2025. We're back mid-January 2026.