What AI changes about building software — and what it doesn't

The honest version, from people who use these tools all day: AI-assisted engineering is real. Work that took a team a quarter can take a small senior team weeks. Boilerplate, migrations, test coverage, first drafts of almost anything — dramatically compressed.

What hasn't compressed: judgement. Knowing which product to build, which corner is safe to cut, which architecture will still be standing in three years, whether the generated code is subtly wrong. AI raises the ceiling of a good engineer and the floor of a bad one — and the market is about to be flooded with the latter's output.

What this means for buyers

• Production hours are worth less. Don't pay 2019 timelines for work AI accelerates.
• Senior judgement is worth more. The scarce skill is deciding, reviewing and architecting — not typing.
• Small teams now beat big ones for most SME-scale work: less coordination, same output.
• 'We use AI' is not a differentiator. Ask to see the quality controls around it.

Where AI belongs in your business

Same test as any technology: frequency × time × error cost. Reading inbound documents, drafting routine replies, extracting data from unstructured mess, triaging support — strong candidates. A chatbot on your homepage because 2025 happened — usually not. The wins are boring, specific and measurable, which is exactly why they work.