Take back control with spec-driven development | DVT Insights.
The rise of AI-assisted coding is prompting renewed attention on how software is planned before it is built. This was a central theme in a recent DVT webinar on spec-driven development and developer responsibility.
Hosted by DVT’s Karl Fischer and presented by principal AI engineer Ewan Mc Phail, the session examined why spec-driven development is being revisited as AI makes code generation faster and easier.
Fischer said the pace of change in AI is pushing teams towards more structured ways of working, with spec-driven development emerging as one response.
Spec-driven development places requirements, architecture and constraints at the centre of the process before code is generated. In an AI-assisted environment, Mc Phail argued, that discipline is becoming harder to ignore. When code can be produced rapidly, the quality of the initial brief plays a much bigger role in shaping the outcome. “AI has turned everything on its head. One of the most practical ways I’ve found to tame the beast is spec-driven development,” he says.
The concern is not simply speed, but what happens when software is generated without enough upfront thought. Without clear specifications, Mc Phail warned, teams can accumulate technical debt far more quickly than before.
“If you don’t do your thinking upfront, AI can generate mountains of technical debt in minutes. You end up with an architecture and app that is totally useless, unmaintainable,” he says.
To illustrate the point, Mc Phail described building a chatbot-based occupational health and safety application using AI-assisted development. “This took me three days from start to finish to get working. About 80% of that was done by AI,” he says. However, Mc Phail said most of the effort still went into defining the specification, underscoring a shift from code-first to thinking-first development.
Fischer also pointed to the downstream impact on delivery and client outcomes, suggesting that faster prototyping gives teams more room to refine what clients actually value. Mc Phail said this is already changing how teams engage with stakeholders.
Despite the gains in speed and productivity, both speakers stressed that responsibility for the final output still sits with developers. “As a developer, it doesn’t matter whether AI spits out a million lines of code, that code is your responsibility,” Mc Phail says. “Whoever ships owns.”
Fischer added that successful adoption of AI-driven development still depends on experience and sound engineering judgment.
The discussion also highlighted how detailed specifications can help teams divide work more cleanly, especially when multiple components are being developed in parallel. A shared, evolving specification gives developers a common reference point and makes it easier to keep architecture, intent and implementation aligned across the team.
The broader message from the session was that while AI may compress development time, it does not remove the need for careful design, technical judgment or ownership of the end result.
A recording of the webinar is available below:
