Insights on AI-assisted software development from the Fabric team
Warehouse robots are ready now. Dark factories are coming this year. But residential plumbing? Roofing? Electrical work? Still 5+ years out. Here's the realistic timeline for physical automation.
15% white-collar unemployment triggers mass protest. 3.5% of the population protesting topples governments. Human-centric cartels will form. The policy window is 18-24 months. Here's the timeline.
For most business tasks, AI models are already 'smart enough.' The real blocker is governance: approval workflows, audit trails, compliance mapping. Here's the infrastructure that's actually missing.
The tech industry thinks humans will be 'exception handlers' for AI. Wrong. Liability structures, tacit knowledge, and professional self-interest will keep humans in the loop for reasons that have nothing to do with AI capability.
Multi-agent AI demos look magical. Reality is messier: 3 agents at 95% reliability each = 85% system reliability. Current 'multi-agent' is mostly nested workflows in disguise. Here's when it actually makes sense.
Everyone's putting AI agents in charge of precise, deterministic systems. They're making a huge mistake. Agents thrive in unstructured chaos and struggle with order. Here's how to deploy them correctly.
Everyone expects AI costs to keep falling. I'm putting 40% odds they rise instead. Here's the evidence from RAM shortages, chip constraints, and energy limits that the consensus is ignoring.
There's a problem with how AI writes code today—and it's not the one you might think. After hundreds of conversations with developers, one theme emerged: the accountability gap. AI makes the changes, but you bear responsibility for the bugs.
