How to build a company with AI from the ground up

Key Takeaways

• Treat AI as part of the operating system of the company, not as a collection of productivity tools. The larger opportunity comes from redesigning workflows around what AI can do rather than simply making existing tasks faster.
• Move from open loops to closed loops. Instead of making a decision, executing it and losing what was learned, capture the work and its outcome so the system can use that feedback to improve the next cycle.
• Make the organisation queryable. Meetings, decisions, project activity, customer feedback and other operating information need to be accessible to the AI; otherwise agents cannot build an accurate picture of what is happening or coordinate useful action.
• Humans increasingly define the goal and judge the result while AI handles more of the implementation. Hu uses software development as an example: humans can specify what should be built and how success will be tested, while agents iterate on the implementation until those tests pass.
• AI-native organisations may need different structures, not merely fewer people. If AI can route information, summarise activity and coordinate work, some functions traditionally performed through layers of management can be redesigned around individual contributors, clear owners and AI-supported coordination.