Why 'Ship 2x Faster' Is the Wrong Goal | Steve Pereira on AI & Flow

Typing was never the bottleneck. Coordination was. Every AI coding tool promises faster developers, but Steve Pereira argues that generating 10x more code without touching coordination just buys you 10x the review burden. In this conversation, Feifan (co-founder of Tanagram, building Lore) sits down with Steve, "The Value Stream Guy" and co-author of Flow Engineering, to dig into what actually drives impact when teams adopt AI: not lines of code, but work that reaches customers and moves the business. They get into why AI often just moves the bottleneck and hides it, measuring at the finish line instead of mid-race, and how coding agents are starting to make knowledge work visible for the first time. Steve breaks down value stream mapping for software teams, the "software factory" mindset borrowed from manufacturing, why the biggest wins are usually in the boring broken parts of your pipeline, and the social-capital model behind curiosity and execution. If you lead an engineering team, or you're trying to figure out where AI actually pays off, this one's for you. What we cover: Why "2x faster" is the wrong metric, and what to measure instead How coding agents make invisible knowledge work visible Pointing AI at existing pain (CI/CD toil, mystery processes) instead of shiny new projects The one move a tech lead or VP of Eng can make when "we have AI now" but nothing changes Value stream mapping, explained for software delivery The "software factory": what real factories teach us about building software systems The curiosity vs. execution pendulum across an engineering career Chapters: 0:00 Why typing was never the bottleneck 0:30 Meet Feifan (Lore) and Steve Pereira 2:01 AI moved the bottleneck, it didn't remove it 3:19 The metric nobody measures: ship to the finish line 6:23 Shared context and the speed of decisions 9:46 Point AI at existing pain, not shiny new projects 12:49 Tiger teams vs. business as usual 15:30 The software factory: encoding patterns over tribal knowledge 19:06 Keep the factory simple: guardrails vs. teachable moments 23:58 Capturing and distilling knowledge with AI 27:38 When work becomes visible 30:35 Asking for help, and becoming a connector 37:23 Engineering is problem-solving, not typing 39:25 The curiosity vs. execution pendulum 41:42 Coordination and the social capital model 45:44 Define the outcome, let the system catch failures 49:20 One move for leaders: value stream mapping 52:40 Wrap-up About Steve Pereira: Steve Pereira is a workflow and value stream expert with two decades spent improving how work flows through engineering organizations, from tech support and IT to build/release engineering and founding CTO roles. He's the co-author of Flow Engineering: From Value Stream Mapping to Effective Action. → Steve's site: https://stevepereira.ca → Flow Engineering (book): https://itrevolution.com/product/flow... About Feifan Zhou: Feifan is the founder of Lore, Lore puts all your AI work in one place for teams that live in Claude Code, Codex, or Cowork. Get a shareable URL for any session showing what worked (or didn't work), along with skill usage and token efficiency. Search decisions, related work, and context across your team without waiting for someone to reply. Connect: → https://lore.link/ → Feifan's LinkedIn -   / feifanz   → Steve's LinkedIn -   / devopsto   #SoftwareEngineering #AI #FlowEngineering #EngineeringLeadership #ValueStreamMapping #DeveloperProductivity #AICodingAgents