AI
The 5-Minute Feature as a North Star
In 2005, CruiseControl was the original CI tool. The promise was simple: push a commit, grab a coffee, come back to a green or red build. That was your 5-minute feedback loop. Klemetti argues GenAI now applies the same logic to full features. Agile arrived 25 years ago, DevOps 15, and GenAI is the next shift that compresses the entire development life cycle to that same 5-minute window.
Tools like Lovable and GitHub’s coding agent are already doing it on new and existing codebases. Most organizations will not get there by accident.
Why Cutting Headcount is the Wrong Frame
Accenture announced 11,000 layoffs days before this talk, and private equity clients keep asking Klemetti whether they can produce the same code with fewer developers.
“And that’s fundamentally wrong question” — Marko Klemetti
R&D organizations spend 5-10% of their budget on actual coding. Optimize that by 50% and total R&D improves 2.5-5%. That is not an order-of-magnitude change. Fine-tuning the existing workflow just produces leaner silos, not a faster organization. The companies that win will rebuild how they produce digital solutions, not trim the headcount on the old process.
Team Topologies as the GenAI Org Model
Team Topologies defines four team types: stream-aligned teams that ship value directly to customers, platform teams that provide the shared tooling and pipelines, enabling teams that help stream-aligned teams get better, and complicated-subsystem teams handling hard specialist domains. Klemetti’s argument is that this structure removes the handovers between requirements, coding, testing, security, and operations that today take hours or days per step.
Conway’s Law means the org chart produces the architecture. Change the chart first, then pour GenAI on top. Otherwise you get faster silos, not faster software.
Agents as Platform Services
Platform engineering gave organizations a shared layer of templates, pipelines, and tooling. Klemetti predicts agents follow the same pattern. A dedicated agentic orchestration team ships agents the way Spotify’s Backstage ships service templates. A GDPR compliance agent, for example, gets subscribed into a team’s pipeline once and fires automatically on every change. When the agent logic updates, the platform team pushes the change, not each stream-aligned team individually.
“probably requiring way less managers” — Marko Klemetti
Self-organized pizza-sized teams, each with access to shared agents, make many coordination layers unnecessary.
Q&A
Who is capitalizing well on GenAI and who is doing it badly? Klemetti named Lovable as the model: launched January, valued at $2 billion, built from scratch using GenAI rather than bolting it onto an existing process, and competing in markets where expensive incumbents have no IPR protection on their applications. ▶ 41:47
Notable Quotes
And that’s fundamentally wrong question Marko Klemetti · ▶ 9:49
hours at best, usually days, which means Marko Klemetti · ▶ 31:54
probably requiring way less managers Marko Klemetti · ▶ 34:52
Key Takeaways
- A 5-minute feature end-to-end is the north star — most organizations will never reach it by optimizing silos.
- Cutting developers from 100 to 50 improves total R&D output by only 2.5-5%, not orders of magnitude.
- Agentic compliance and review agents deployed via a platform team eliminate handover delays without reorganizing constantly.