If recursive self-improvement is not a core part of your product AND a core part of your business, you are not going to make it.

You have the ability to make a feedback loop that automates turning customer feedback and signals into product improvements, automatically - particularly for interactive AI-powered applications.

• Agents can mine and analyze all of your data.
• Agents can turn that into product strategy and requirements in linear.
• Agents can turn that into evals that prove poorly handled situations are now handled better, without regressions in other areas.
• Agents can manipulate your codebase and your prompts to implement the changes.
• Orchestration tools can host the jobs, and run the machinery to make it automatic in your cloud environment.
• These days even open-source LLM models are good enough and run on your own (or rented) hardware to keep the data loop private.

Yes, you should probably still have humans looking at the data and rubberstamping the improvements (proved by evals) before they go out the door… But everything can be ready for a human to review in a nice friendnly UX, or in a complete ephemeral app environment, with an approval button.

Your feedback loop should have 100x’d over the classic human-driven Lean Startup feedback loop you were doing 1 year ago.

Seriously: If recursive self-improvement is not a core part of your product, you are not going to make it.