How we use AI every day to write software
Not 'AI' as a buzzword. A concrete operational change that reduces shipping time and bugs. Here's how.
At Spiralway we don't have an 'AI team' separate from the development team. AI is part of the flow, like the editor or version control. This is the short version of what that looks like in practice.
Accelerated discovery
Before, a one-week discovery consumed a lot of time transcribing conversations, synthesizing requirements, and assembling a first document. Today a one-hour conversation is auto-transcribed, summarized, and turned into an editable scope skeleton in minutes. That gives time back to what matters: thinking about the problem.
Assisted code, not automatic code
We use LLMs to write code all day, but no PR hits production without a human reviewing, refactoring, and understanding every line. The difference from three years ago is speed: the first draft of a new function, a test, a migration, or a script, is ready in seconds. The engineer's time is freed up for design, edge cases, and architecture decisions.
Tests, docs, and ops
AI particularly accelerates three things classic teams postpone: tests, documentation, and ops scripts. What used to be 'I'll do it later' now happens in the same PR. Result: less technical debt, fewer production surprises.
What AI doesn't do
Decide what to build. Understand the client's real context. Make an architecture decision that affects the next three years. Those remain human. That's why you talk to the people building, not an intermediate PM: important decisions aren't delegated to a machine.