The biggest thing AI changes about software development isn't how fast you write code. It's how much of the work you can hand off.
Most developers meet AI through tools: better autocomplete, a chat window, a coding agent. The deeper shift is harder to see. As you learn to delegate larger parts of the development process, your leverage grows, your role changes, and the most valuable skills move from writing code to defining, verifying, and directing the work that produces it.
Agentic Programming is a practical roadmap for that shift. At its center is the AI Fluency Ladder: five levels of delegation, from writing code by hand to systems that turn an objective into deployable software on their own. The book shows what changes at each step: the bottleneck that moves up, the trust that has to be earned, and the architecture required to earn it.
The first half builds the foundation: how large language models actually behave, the eight ways they fail, and the mental models agentic programming depends on. The second half turns to construction. It shows how to build an agentic harness (workflows, coordinators, verification, retries, and recovery, the architecture that makes delegation reliable enough to trust) and follows that architecture up the ladder to autonomous execution. A hands-on appendix builds a working harness from an empty repository, and its complete source is available as open-source companion code. A closing part covers the economics of the ladder and what the shift means for engineers, teams, and careers.
Throughout, the book reports what worked on real projects, puts honest numbers on the gains, and is candid about the limits, including the failures that proved most instructive.
It's written for software developers, architects, engineering leaders, and technical founders. No machine-learning background required.
If you build software and want to understand where the field is heading, and how to move with it one level at a time, start here.