When Software Gets Cheap, Focus Gets Expensive
Agents and Engineers | An AI Podcast
0:00 / 0:00
When Software Gets Cheap, Focus Gets Expensive
15 просмотров · 5 дней назад
Agents and Engineers | An AI Podcast
21 подписчик
15 просмотров · 5 дней назад
Greg Ceccarelli is the co-founder and Chief Product Officer of SpecStory, which builds tools for developers using AI coding agents. He previously led product at Pluralsight and worked on the first GitHub Copilot technical preview as GitHub's Director of Data Science.
Dan and Greg discuss what becomes scarce when software is easier to produce. Greg warns that a quick MVP can create false confidence before a founder has established customer demand or willingness to pay. He describes mining roughly 1,300 SpecStory sessions and nearly 5,000 commits to identify patterns in his own agentic engineering practice.
Those patterns put verification and durable documentation at the center of the work. Greg uses architecture documents to preserve context between agent sessions and prefers steering local agents while they run so he can maintain his mental model. The conversation examines the limits of unattended work and why product judgment remains difficult even as implementation gets faster.
Chapters:
00:00:00 Introduction and guest background
00:00:55 What agentic teams are running into
00:06:56 Startup leverage, MVP traps, and maintaining SpecStory
00:09:26 When software gets cheaper, distribution matters more
00:12:31 Hand-written code, craft, and code as liability
00:16:21 Mining 1,300 sessions into 25 patterns
00:19:01 Verification and as-built architecture docs
00:23:55 Co-writing docs with LLMs
00:25:15 Keeping docs fresh through skills, Git, and verbose commits
00:27:50 Trunk-based development for agentic teams
00:30:26 Local steering versus cloud-agent pull requests
00:32:14 Goal and rider plans, long runs, and Gas Town
00:35:52 Replacing issue trackers with weekly docs
00:38:19 Larger teams and issues as agent-ready specs
00:42:45 Parallel projects and concentration limits
00:44:47 Local agents, permissions, and risk judgment
00:46:57 The cognitive pull of managing agents
00:51:58 Scarcity, token costs, and model choice
00:58:57 Copy, design, naming, and taste
01:05:04 Why creative output resists verification
01:07:12 Closing