Thinking about the new economics of software.

Why AI changes the economics of software teams.

The opportunity is not simply to replace developers. It is to redesign the way experienced people spend their time.

Traditional delivery models were shaped around a practical constraint: software implementation required a large amount of human production work. As projects grew, organizations added developers, analysts, testers, coordinators and layers of management to keep the work moving.

AI changes the amount of implementation a capable engineer can direct. Code generation, test creation, codebase exploration, refactoring, documentation and investigation can increasingly happen in parallel. That does not remove the need for engineering judgment. It makes judgment more valuable.

The strongest operating model is therefore not “one developer replaces ten.” It is a smaller team in which senior people stay close to the business problem, use AI to compress execution time, and add specialist depth only where the product requires it.

• Keep product decisions close to the people building.
• Use domain experts to validate the real workflow early.
• Give AI agents bounded implementation work rather than product authority.
• Measure the team by working outcomes, not visible headcount.

When an AI prototype is not yet a product.

AI has made the first 70% of a product dramatically easier to see. The remaining work is what allows real users to trust it.

A prototype can now look remarkably complete: polished screens, working flows, a database, even integrations. That is valuable because it lets founders and operators discover the product by using it rather than describing it.

But production introduces a different set of questions. Who can access what? What happens when an integration fails? Can data be recovered? Are permissions enforced on the server? Are secrets protected? Is the codebase maintainable? Can someone understand a failure at 2 a.m.?

Productionization is not a ritual of rewriting everything. The right approach is to assess what is sound, keep it, and deliberately engineer the missing system around it: architecture, identity, data integrity, APIs, tests, deployment, monitoring and operational ownership.

Make existing software usable by AI agents.

Most enterprise systems were built for humans clicking screens. Agents need a safer, more explicit way to use the same business capabilities.

The next modernisation layer is not necessarily another front end. It is a machine-usable capability layer: APIs, documented business functions, permissions, events and controlled tool interfaces that allow an AI agent to inspect information and take approved actions.

This can often be added progressively. Start with one useful capability—retrieve a case, create a request, update a status, generate a document—then expose it through a secure API or MCP interface with clear identity and authorization.

The result is a platform that can continue serving human users while also becoming accessible to agents, automations and new AI-enabled products.