Search every talk
Talk Explorer
Search all recorded talks by topic, speaker, track or what was actually said on stage — then jump straight to that moment in the recording.
Open the explorer →WeAreDevelopers World Congress 2026
The former GitHub CEO declares the SDLC dead: his fix is to review signals, not diffs
Dohmke's WWC26 opening keynote: agents write most code, PRs have tripled in size, and 20% more code merges without a human ever reading it. His answer, built into his startup Entire: session logs stored in git next to the code, intent captured before the diff exists, and merge gates that rescore on every push. Plus a decentralized Git network for agent-scale traffic.
Read the recap →Intercom bet 90% of the company on a product earning 1% of revenue, renamed itself Fin, and sold to Salesforce for $3.6B
Fin's co-founder on the pivot that saved the company, and what it did to engineering: productivity tripled, 94% of code written by Claude, cost per shipped feature falling even at $180k a week in tokens. The core lesson: reliability compounds, so a chain of 99%-reliable agent steps makes an ~85%-reliable product nobody buys.
Read the recap →Typists vanished, typing didn't: why AI replaces developer tasks, not developers
The CEO of 3T Software Labs argues AI is unbundling the developer role the way word processors unbundled the typing pool: tasks get replaced, the coordinating function doesn't. The job shifts to governance: maintainable code, outcomes, risk, and the inflection points where agents need redirecting. In the Q&A he bets his company against spec-driven development.
Read the recap →Zeiss threw away a microservice's code and let an agent regenerate it from spec: 80% match on the first shot, 90% after tweaks
Two Zeiss architects run a real Java microservice through a spec-driven agentic workflow (Claude Code + agent skills, no human between plan and implementation), then go further: delete the codebase and regenerate it from the spec. Same-language regeneration nearly works; a Go rewrite fails exactly where the guardrail artifacts are missing.
Read the recap →He set out to prove AI can't build enterprise software. He shipped 420,000 lines in 45 days instead.
A quality-obsessed architect rebuilt his company's business application with AI agents: 420k lines of .NET microservices, solo, in 45 working days. The useful part isn't the speed claim — it's his "harness" method: review the machinery that constrains the output, not the output itself.
Read the recap →A regulated bank runs an "AI employee" for ~47 teams on 100 bucks a month of infra
Trade Republic's internal AI employee serves ~47 teams and 10,000+ sessions a week, solving over 90% of requests in some channels on its own. The design inverts the usual agent stack: guardrails and decision-making live in deterministic code the model can't override.
Read the recap →The 10x engineer is a myth. The 11x organization is real, and AI is quietly raising its defect rate.
Top and bottom performers inside the same company differ by only 21% — organizations differ by 11x. The gap is the environment, and AI is making it worse in places: change failure rates up, PR size nearly doubled. Treat productivity as a systems problem.
Read the recap →