Code is clay now
A useful distinction between fast human-in-the-loop exploration and more automated production pipelines: agentic coding lowers the cost of changing code, but judgment still has to be applied either during each iteration or built into the factory.
Alberto Denia argues that agentic coding makes software cheaper to reshape: code behaves less like stone that is costly to alter and more like clay that can be modeled, tested in use, and refined. This makes it practical to try larger product or feature changes in the running system instead of relying only on plans and mockups.
The essay separates two ways to apply engineering judgment. The artisan stays in the loop, judging each iteration, which suits exploration and work where the product needs to feel right. The factory embeds judgment upstream in specs, generators, tests, evaluations, and guardrails, then escalates exceptions to people. Repeatable work in well-structured codebases is a better fit for this mode.
These are ends of a spectrum, not competing doctrines. A team can use hands-on iteration for novel problems and automated pipelines for routine changes. The cost of making changes may fall, but deciding which changes are right remains engineering work.