In 2025, I rarely used AI-assisted development. My early experiments had been underwhelming; the output was often confidently wrong, and fixing it felt slower than writing the code myself.

Within a few months, that changed. Today, I use AI every day, but I treat everything it produces as a proposal to understand and validate, never as an authoritative answer.

The shift began during a performance investigation where AI proved genuinely useful. From there, I experimented with almost pure vibe coding before introducing more discipline through specifications, agent roles, focused handoffs, separate review and tighter validation.

AI now helps me plan changes, explore unfamiliar code, summarise support cases, generate tests, implement features and review pull requests. It saves time and lets me make progress on more things, but it also brings new costs: cognitive load, growing review queues, token usage, context switching and the risk of losing touch with the craft of writing code by hand. Am I more productive, but less fulfilled?

I now believe that using these tools effectively is becoming a core engineering skill. In this session, I’ll share what has worked, what has failed and the practical habits I use to stay in control.

This is not a story about replacing developers. It is about benefiting from powerful tools without surrendering judgement, understanding or responsibility.