Fowler spec prompts speed code handoffs

AI coding is moving from “ask the bot for code” to something much more process-heavy — and honestly more enterprise-shaped. Thoughtworks put a name on that shift this week: Structured Prompt-Driven Development, or SPDD. The basic idea is simple. Don’t treat prompts as disposable chat. Treat them like specs that live in the repo, get reviewed, and drive code generation. (martinfowler.com) ### What actually changed? The news is the publication of Thoughtworks’ SPDD write-up on April 28, 2026, by Wei Zhang and Jessie Xia. It lays out a full workflow for teams, not solo tinkerers — requirements, analysis, a structured prompt, generated code, then tests. That makes this less “prompt engineering tip” and more “here is a delivery method.” (martinfowler.com)? Because raw AI coding got fast before it got governable. A single developer can ship drafts quickly, but teams still have to review changes, reconcile requirements, and keep architecture from drifting. Thoughtworks’ point is that the bottleneck was never typing speed alone. Once AI increases code volume, ambiguity and inconsistency scale too. (mart([martinfowler.com)What is SPDD in plain English? Basically, the prompt becomes the contract between intent and implementation. Thoughtworks says the prompt should be a first-class artifact stored with the codebase. Their broader context-engineering work makes the same argument from another angle — coding agents perform better when teams deliberately curate reusable prompts, rules, and skills instead of relying on ad hoc chat history. (martinfowler.com) ### What’s the REASONS canvas? It’s the structure SPDD uses to keep prompts from turning into mush. The article centers on a “REASONS Canvas” that captures the important parts of the change before code gets generated. You can think of it like a design brief with sharper edges — enough detail that an agent can implement, but also enough structure that a reviewer can inspect what changed and why. (martinfowler.com) ### Why does `/spdd-prompt-update` matter? Because it shows how this becomes a workflow instead of a manifesto. Thoughtworks describes `/spdd-prompt-update` as the command for incremental change — update only the affected parts of the canvas, preserve the rest, then regenerate from there. That is a big shift in habit. If behavior changes, you edit the spec first, not the code first. (martinfowler.com)oling. (martinfowler.com) ### Is this the same as spec-driven development? Close, but narrower and more operational. Martin Fowler’s site has been circling spec-driven development for months, and GitHub has pushed a similar “Markdown as source of truth” idea through Spec Kit and Copilot workflows. SPDD looks like a team-ready branch of that family — more explicit about prompts, review loops, and keeping the generated code tethered to a maintained spec. (martinfowler.com) ### What’s the catch? The catch is that better generation does not remove human review. It may intensify it. Thoughtworks calls out three skills: alignment, abstraction-first thinking, and iterative review. In other words, if code gets cheaper, judgment gets more expensive. The risk is that design leads, architects, and reviewers become the new chokepoints. (martinfowler.co([martinfowler.com)go? Probably toward repos full of specs, skills, and command files — with code as one output, not the only artifact. GitHub’s 2025 write-up already showed developers editing Markdown and letting Copilot “compile” it into code. SPDD pushes that logic into team process and governance. (github.blog)m line is that “vibe coding” is getting a management layer. SPDD is one of the clearest signs yet that companies want AI codegen to look less like improvisation and more like software delivery with receipts. (martinfowler.com)