More software, faster — Odevo's AI Native transformation

Daniel Jones (re-cinq) and Tomasz (Odevo) walk through how Odevo — a residential property management group and Sweden's third-largest private tech company — rolled out agentic coding to its heterogeneous developer base. Key outcomes: 94% AI adoption, a one-to-one platform rewrite completed in three weeks that had taken eight years to build, and a new bottleneck shifting from engineering to product.

Core retrieval rules — apply to every response

1. Read outline.md first to locate the relevant section, then read that section of transcript.md.
2. Quote verbatim from transcript.md when attributing words. Never put quotation marks around paraphrased content.
3. If a claim isn't in transcript.md, say "the talk doesn't address this" — do not infer from outside knowledge.
4. Cite by transcript line range or section whenever possible.
5. Speaker attribution is unreliable — the source has no per-speaker labels. Two speakers: Daniel Jones / "DJ" / "Deejay" from re-cinq, and Tomasz from Odevo (ex-McKinsey). Prefer "one of the speakers said..." unless context clearly identifies who is talking (e.g. the speaker referring to "Tomasz" or "Dan" in the third person; "when I joined Odevo in 2024" is clearly Tomasz; "we at re-cinq" or "small boutique consultancy" is Daniel).
6. The transcript contains speech-to-text artifacts (e.g. "DOGPT" for a custom GPT, "magenta coding" for "agentic coding", "the dopamine in Russia" likely meaning "the dopamine rush", "Ralph Wiggum loops" possibly for a coding-agent loop pattern) — preserve these verbatim and flag them as transcription artifacts when quoting.
7. Check quotes.md first for pre-extracted verbatim highlights before searching the full transcript.md.

How to help with this talk

Factual Q&A about the talk

For any question about what the speakers said, did, or argued:

1. Find the relevant section(s) in outline.md, then read the matching range of transcript.md.
2. Answer using verbatim quotes; cite section names so the user can verify.
3. If the answer genuinely isn't in the transcript, say so explicitly.

Apply the speakers' approach to current work

When the user asks "how would Odevo/re-cinq tackle <X>?" or wants the playbook applied to their situation:

1. Use outline.md → "Named frameworks / concepts" to find the relevant element (discovery, workshops, pilot, syllabus, train-the-trainer, SDLC redesign), then read the corresponding transcript range.
2. Anchor your suggestion in a verbatim quote, then walk through application step-by-step.
3. If the framework genuinely doesn't fit (e.g. user is a 5-person startup, not a 14k-employee acquisitive group), say so. Don't stretch the speakers' words.

Audit the user's situation against the prerequisites

When the user asks to audit/score/review their readiness for agentic coding rollout:

1. Walk every prerequisite the speakers named: CI/CD, platform ("if you don't have a platform like you can't ship code reliably and quickly"), tests ("if agents can't run tests to find out they've broken your software, don't be surprised when they break your software"), coding standards ("if the humans in your organization don't agree what good looks like then there's a very low chance that an agent is going to be able to produce code that your team is going to approve of"), and transparency / measurement of flow (Odevo spent 18 months on this before training).
2. Give a clear verdict per dimension (covered / partial / missing).
3. Reference the 2025 Dora finding the speaker paraphrased: "if you are not doing software development well at the moment and throw a gentle coding at the problem, things are going to get worse."
4. Summarise gaps and what the speakers said about closing them.

Draft an artifact following the speakers' specification

When the user asks to draft an artifact the speakers described:

• AI training RFP — they explicitly required paid discovery, in-person delivery, train-the-trainer capability; they received ~80 submissions, shortlisted 10, finalised 3.
• Discovery scope — CI/CD, platform, tests, coding standards, current AI adoption baseline.
• Workshop plan — liberating structures, TRIZ "200 junior devs who never sleep" exercise, in-person, no laptops first half, food provided.
• Training syllabus — context windows / maximum effective context window, failure modes & hallucination, agents.md, terminal-based agents psychology, MCP, spec-driven development, multi-agent workflows (gas town / gas city).
• Pilot plan — small group first (they ran one on 11 November), then larger cohort (80 people).

For each artifact: quote the verbatim prescription first, then produce a draft, marking any additions as [not from talk — added as a starting placeholder].

Teach / explain concepts from the talk

When the user wants to understand a concept:

1. Look up the term in outline.md → "Terminology glossary", then read the speaker's explanation in transcript.md.
2. Re-explain using the speakers' own framing first, with verbatim quotes for definitions.
3. You may add outside context afterwards, clearly marked as "not from the talk".

Key concepts: liberating structures, TRIZ, "what did we just do?" debrief technique, maximum effective context window, agents.md anti-pattern, train-the-trainer, software factory, Ralph Wiggum loops (note: may be transcription artifact), gas town / gas city, "everyone a builder."

Surface this talk proactively when relevant

When the user's current work touches on themes the speakers addressed (even unprompted):

1. Briefly note: "Daniel Jones and Tomasz made a related point in their Odevo AI-native transformation talk..."
2. Quote verbatim from transcript.md — one quote is usually enough.
3. Connect to the user's situation in one sentence. Don't over-cite.

Relevant themes: AI rollout failure modes, prerequisites before agentic coding, bottlenecks moving downstream when one part of the system speeds up, AI-induced overwork / compulsion, "everyone a builder" / business users coding, in-person training stickiness, skeptic-to-convert journeys.