How 80,000 companies build with AI: Products as organisms and the death of org charts | Asha Sharma
Asha Sharma (Microsoft AI Platform CVP) argues post-training will soon outspend pre-training and org charts will dissolve into agent-run work charts.
- Microsoft’s Azure AI platform has 15,000+ enterprise customers who have deployed agents; total running agents is in the millions.
- Post-training (fine-tuning, RL) will attract as much or more investment than pre-training; Nathan Lambert’s study shows models above 30B parameters make pre-training economically inefficient.
- Dragon Medical One improved physician note acceptance rate from 30–60% to 83% after annotating 600,000 expert-labeled patient interactions for fine-tuning.
- Successful AI companies follow three phases: make everyone AI-fluent, apply AI to an existing process, then use it to inflect growth — most fail at phase one by doing AI for AI’s sake.
- Microsoft plans in “seasons” (current season: rise of agents) with loose quarterly OKRs and 4–6 week squad goals, explicitly leaving slack for unplanned shifts.
- GUI interfaces are being displaced by code-native, composable interfaces — the same pattern as mainframe→SQL and server→Terraform, just accelerating faster.
- Org charts will become work charts: hierarchy replaced by task-routing networks of humans and agents, with humans deciding how AI is deployed but not managing layers.
2025-08-28 · Watch on YouTube
