Everyone's talking about LLMs making engineers code faster. Juhani Lehtimäki and Jasper Morgan argue that's the boring half of the story, and that companies handing out Cursor licences inside a Scrum process are still about to ship the wrong product but maybe faster than ever before.

In this episode we unpack the Seed Framework: a way of building software that uses LLMs where they're actually strongest, not as a junior-engineer replacement, but as a validation engine. The idea is to build a real, working, deliberately throwaway piece of software in days, put it in front of stakeholders against real data, and let that artefact surface the spec that JIRA tickets, Figma click-throughs and Design Sprints never could.

Then we walk through two real case studies:

• VertMatch.run: a trail-running analytics tool where the first Seed was thrown away, because real data made it obvious the "obviously useful" dashboard wasn't actionable.

• IndoorBike.app: an open-source indoor cycling app where the very first Seed, ridden on a real trainer against real workouts, was right on the first try. Three days, end to end.

The bigger claim: waterfall isn't the villain it was made out to be. Waterfall on a guess is the villain. When the spec is validated and locked, waterfall is the most efficient engineering model that exists and LLMs are what finally let us validate the spec before we commit a team to building it.

━━━━━━━━━━━━━━━━━━━━

━━━━━━━━━━━━━━━━━━━━

#SeedFramework #LLM #SoftwareEngineering #AICoding #Scrum #ProductDevelopment #ClaudeCode #VibeCoding