Dominik Tornow argues that reusable specifications can let coding agents build implementationsCode generation uses AI or another automated system to create source code from instructions, examples, schemas, or higher-level specifications. tailored to existing infrastructure. He describes Resonate’s attempt to derive multiple durable execution servers from a shared protocolSpecification-driven development defines intended behavior and acceptance criteria before implementation begins..
A direct move from abstract specification to production code failed under concurrency and network or process failures. Adding a concrete specification helped, but humans initially had to drive the target-specific design.
Tornow describes using deterministic simulationDeterministic simulation reproduces a modeled system's execution under controlled conditions so that the same setup and choices yield the same sequence of events. Repeatability helps investigate difficult failures. to expose stale reads and failed updates, recording diagnostic facts hidden from production code. He says the agent used repeatable traces and fuzz testingFuzzing tests software by generating many varied inputs and observing failures such as crashes, hangs or violated properties. to refine an algorithm, derive a concrete specification and implement it.
Watch on YouTube
