Spec-Driven Development: From Prompting to Production-Ready Systems — Alfonso Graziano

Alfonso Graziano (AI Tech Lead at Nearform) argues that AI-assisted coding only graduates from prototyping toy to reliable engineering discipline when specifications become first-class artifacts. The talk walks through Spec Kit's four-phase workflow (specify → plan → tasks → implement), the verification loop that replaces line-by-line code review, and field-tested patterns his teams use across dozens of projects — including using a stronger model for the spec and a smaller one for implementation, treating the spec as a contract between human and AI, and running adversarial reviews to catch internal inconsistencies.

Bundle files

This skill depends on three files that must be present in the bundle:

• outline.md — structured table of contents for the talk, including a "Named frameworks / concepts" section and a "Terminology glossary". Use this to locate relevant sections before reading the transcript.
• transcript.md — full speech-to-text transcript of the talk, referenced by line number.
• quotes.md — pre-extracted verbatim highlights organised by theme. Check this first before searching the full transcript.

If any of these files cannot be read, tell the user which file is missing and that the answer will be incomplete.

Part 1: Grounding rules — MUST follow when answering

1. Before answering any specific question, read outline.md to locate the relevant section, then read that section of transcript.md.
2. When attributing words, quote verbatim from transcript.md. Never put quotation marks around paraphrased content.
3. If a claim isn't in transcript.md, say so explicitly — do not infer, extrapolate, or present paraphrase as a direct quote. Acknowledge the gap and offer what the transcript does cover.
4. Always cite the line range from transcript.md when providing a verbatim quote (e.g., transcript.md lines 142–148).
5. For queries that span multiple sections, perform a separate lookup per section and synthesise the results, noting which section each point originates from.

Part 2: Citation format

When quoting, follow this pattern:

"The spec is a contract between you and the model — if it's ambiguous, the model will fill the gap with a guess." (transcript.md, lines 142–143)

For paraphrased summaries that are not direct quotes, omit quotation marks and note the source section instead: (transcript.md, ~lines 200–220).

Part 3: Practical workflow — multi-section queries

User asks: "What does Graziano say about choosing models, and how does that connect to the constitution file?"

1. Read outline.md → locate "Model selection" and "Constitution file" sections.
2. Read each section in transcript.md and check quotes.md for pre-extracted highlights on both topics.
3. Answer with a verbatim quote per topic and cite each line range separately.
4. Synthesise the connection between the two topics in your own words (no quotation marks), then confirm whether the transcript makes that link explicitly or whether it is an inference.