Speaker

Will Wilson

Will Wilson

Co-founder and CEO of Antithesis, a company building autonomous testing infrastructure. His background is in pure mathematics (representation theory, large cardinal theory) before a self-taught transition into software engineering that led him to FoundationDB, the distributed database acquired by Apple in 2015.

At FoundationDB, Wilson helped build and operate the system’s deterministic simulation testing framework — a proprietary environment that could run an entire network of database processes inside a single Linux process with mocked networking, fake concurrency, and fully reproducible execution. That framework was the enabling condition for FoundationDB to build features — including a from-scratch Paxos consensus implementation and a complete rewrite of its concurrency control algorithm — that would have been reckless without it.

Antithesis takes that idea and generalises it: rather than requiring engineers to write their software inside a special framework, it implements a hypervisor that makes arbitrary software deterministic at the virtual-machine level. Software runs unmodified; the hypervisor controls all hardware-level sources of non-determinism and enables massively parallel state-space exploration through copy-on-write memory deduplication across VM forks.

Core positions

Wilson holds that non-determinism is the root structural problem in software testing: it prevents bugs from being reproducible and degrades randomised search (fuzzing, property-based testing) into random guessing. He is a practitioner of property-based testing — the technique of specifying what a program should do and generating vast numbers of random inputs to find violations — but argues against the mathematical tradition’s assumption that properties must be exhaustively specified before testing can begin. Coarse properties (does not crash, meets SLA) catch a large fraction of bugs in practice because software’s own chaotic behaviour amplifies subtle faults into visible symptoms.

On AI coding, he argues that verification is the new bottleneck: generating code at scale is now easy; deciding that it is correct is not. He also warns that putting AI agents in a closed loop with a rigorous validator reintroduces the ‘evil genie’ failure mode of classical program synthesis — the agent optimises to satisfy the test rather than to solve the problem.

He deliberately chose to found Antithesis in Washington DC, and to work in a neglected problem area, on the theory that combining genuine economic importance with low status is an arbitrage opportunity for companies willing to do the work early.

In the wiki