Descriptions:
Dan Shipper, CEO of Every, shares a firsthand account of running a company where 99% of code is written by AI agents — specifically Claude Code, Codex, and similar coding agents. Every operates six business units and four production software products, including Kora (an AI email management app) and Monologue (a speech-to-text app), with just 15 people and approximately $1 million in total funding raised. The company has grown monthly recurring revenue by double digits every month for six consecutive months and has over 7,000 paying subscribers and 100,000 free users. Each product is built and maintained by a single developer.
Shipper attributes this leverage to a workflow he calls compounding engineering — a four-step loop of Plan, Delegate, Assess, and Codify. The differentiating step is Codify: rather than letting tacit knowledge stay in engineers’ heads, teams capture everything learned during planning, delegation, and debugging as explicit prompts stored in CLAUDE.md files and agent slash commands. Where traditional engineering makes each new feature incrementally harder to add due to growing complexity, compounding engineering is designed so each feature makes the next one easier.
The talk also examines an organizational threshold Shipper has observed: the difference between teams where 90% versus 100% of engineers use AI is qualitative, not incremental. Even a small holdout forces the whole team to maintain backward-compatible, human-paced workflows. Grounded in real production numbers and named tooling, the talk is among the more credible firsthand accounts of AI-native software development available.
📺 Source: AI Engineer · Published December 18, 2025
🏷️ Format: Workflow Case Study
