Your software has a new kind of user
For decades we designed software for people clicking through screens. Increasingly, an agent is the one doing the work, and it needs different things: clean APIs, typed actions it can call safely, and machine-readable context about what is possible and what just happened.
Systems that expose those well become easy to automate. Systems that hide everything behind a human-only interface stay stuck doing things by hand.
Weave the intelligence in
In customer-facing products, the AI features that win do not feel like a chatbot stapled to the side. The intelligence is part of the core flow: it anticipates, drafts, and acts where the user already is. That is harder to build and much harder for a competitor to copy.
Evals are the new tests
Traditional software earns trust through tests. AI software earns it through evaluations: a measured, repeatable way to know the system still does its job as models and data change. Teams that treat evals as part of the product, not a one-off, ship AI they can actually rely on.
