Stop Maintaining, Start Evolving: Applying AI-Native Practices to Brownfield Codebases — Katie Roberts

Katie Roberts shares lessons from a hands-on experiment: using AI to reverse-engineer and rebuild an S3-compatible object storage system (~350k lines of Rust, written in a few months) after MinIO stopped maintaining theirs. The talk argues that AI-assisted development of large complex systems is feasible if you stay in the loop, use an external test oracle to ground behaviour, kill flaky tests immediately, and treat refactoring (and throwing code away) as a routine part of the feedback loop — not as failure.

Supporting files

Two files accompany this skill and must be consulted before answering:

• outline.md — contains: section headers with line-range pointers into the transcript, a terminology glossary (test oracle, human-in-the-loop, refactor weeks, type-driven authorisation, AI-built tracing, etc.), and a "Named frameworks / concepts" index.
• transcript.md — the full talk transcript. Speaker labels are absent; the bulk is Katie Roberts presenting, with unlabelled audience questions at the end. A participants list appears in the transcript header.

Grounding rules and general workflow — MUST follow when answering

Apply these steps for every query:

1. Read outline.md to find the relevant section(s) and any applicable glossary or framework entries.
2. Read the corresponding range(s) of transcript.md.
3. Answer using verbatim quotes from transcript.md. Never put quotation marks around paraphrased content. Cite line numbers so the user can verify.
4. If a claim isn't in transcript.md, say "the talk doesn't address this" — do not infer positions from outside knowledge.
5. Speaker attribution is unreliable — the source has no per-speaker labels. Prefer phrasing like "an audience member asked..." for questions, and only attribute to Katie when the content is clearly her presenting. Cross-reference any named addressee with the participants list in the transcript header / outline.md before attributing.

How to help with this talk

Apply the speaker's approach to current work

When the user asks "how would Katie tackle X?" or wants the talk's approach applied to their own situation:

1. Use outline.md → "Named frameworks / concepts" to find the relevant idea (test oracle, human-in-the-loop, refactor weeks, type-system enforcement, etc.).
2. Follow the workflow above to retrieve and quote Katie's exact wording.
3. Walk through applying the approach step-by-step to the user's case, anchoring each step in a verbatim quote.
4. If her approach genuinely doesn't fit (e.g. she had an external test oracle in S3; the user has no equivalent), say so. The Q&A explicitly covers this risk — see the "tautology" exchange at the end of transcript.md.

Teach / explain concepts from the talk

When the user wants to understand a concept Katie covered (test oracle, flaky tests with AI, type-driven authorization, AI-built tracing):

1. Look up the term in outline.md → "Terminology glossary" first.
2. Follow the workflow above to retrieve and quote Katie's explanation.
3. Re-explain using her own framing and examples, with verbatim quotes for key claims and definitions.
4. You may add modern context or comparisons afterwards — but mark them clearly as "not from the talk".

Factual Q&A about the talk

For any question about what Katie said, did, or argued:

1. Follow the workflow above exactly.
2. Note: the abstract promises Strangler Fig content the delivered talk does not substantively cover — be honest about that gap if it comes up.

Surface this talk proactively when relevant

When the user's current work touches on themes Katie addressed (even if they haven't asked about the talk):

1. Briefly note: "Katie Roberts made a related point in her brownfield AI-native talk..."
2. Quote verbatim from transcript.md — one quote is usually enough.
3. Add one sentence connecting the quote to the user's situation.
4. Especially relevant when the user is: writing AI agent tests, debating 100% coverage, dealing with flaky tests, doing AI-assisted performance work, or discussing how much to stay in the loop with AI coding agents.
5. Do not over-cite. If the connection feels strained, stay quiet.

Key quotes

quotes.md contains pre-extracted verbatim highlights from this talk, organised by theme. When formulating answers, check quotes.md first for strong citable evidence before searching the full transcript.md.