Software Inversion

More people can now turn an idea into something they can try. An operator can work through a question with an AI assistant, a designer can explore a working interaction, and an engineer can test an approach without first building a model from scratch.

That changes the opening conversation. “Can we build it?” becomes less useful on its own. We can often build several versions before we understand which one deserves to exist.

Knowing the work changes what gets built

The important knowledge may sit with the person who handles the awkward requests, spots the unreliable source, or knows why a seemingly redundant step is there. Getting that person into the work changes the brief. Their contribution becomes more valuable when they can show a better version instead of waiting for someone to translate a request.

This is why we work across product, service, architecture, and leadership. A problem does not become purely technical because an AI model is involved. Sometimes the useful intervention is a working tool. Sometimes it is a clearer decision, a different conversation, or permission to stop doing something.

Easier to start does not mean finished

Model access removes some work; it does not settle whether an answer is dependable, whether the source is fit for the task, or whether anyone can maintain the result. Evaluation and engineering still matter. Their shape follows what people will rely on and what happens when it fails.

If a first version is cheaper to make, use that freedom to learn sooner. The advantage is being able to examine more possibilities with the people who care about the result.

Original illustration generated with AI for this belief.