Will Wilson would like your software to fail sooner

Will Wilson once looked at Google and Facebook and concluded, with the confidence available mainly to undergraduates, that computing had exhausted its interesting possibilities. The good ideas appeared to have been taken. He went off to study mathematics, where at least nobody seemed likely to finish the subject before he graduated.

Years later, he would co-found a company devoted to the extraordinary number of ways computers can still get things wrong. There is a pleasing economy to this career correction. The field he considered finished turned out to contain enough unfinished business to occupy his working life.

At Antithesis, Wilson and his colleagues build controlled environments in which software encounters trouble: awkward timing, interrupted communication, failing machines. When something breaks, the sequence can be reproduced. Engineers get another look. A failure that would otherwise arrive as an unwelcome surprise becomes something they can examine.

The ambition has brought him from FoundationDB through Apple and Google, back to Virginia, and into the peculiar business of selling people a place where their creations can misbehave.

The job he learned to want

Wilson’s acquaintance with computers began before his professional interest in them. As a teenager, he installed Linux on a laptop, an undertaking he remembers as considerably less accommodating than it would be today. At university, he chose pure mathematics. Programming returned through work, when small scripts proved unexpectedly useful for automating tasks.

FoundationDB gave that practical interest a direction. It also gave him a group of colleagues who would matter for years. Dave Scherer, later his Antithesis co-founder, had been his boss there. Nick Lavezzo, Antithesis’s chief operating officer, was another connection from the database company. Wilson describes the pleasure of working again with people whose judgment he already knows.

The technical lesson was equally durable. FoundationDB tested its distributed database inside a simulation, controlling events that ordinarily arrive in unpredictable orders. Wilson had been skeptical about testing. Here he saw an approach that made difficult development feel different: when failures were visible and repeatable, the team could investigate changes with much less guesswork.

A world with a replay button

Apple acquired FoundationDB in 2015. Wilson subsequently worked at Google. The old team dispersed, but the memory of how they had worked together proved harder to shake. Why should that kind of testing remain the privilege of one specially constructed database?

Antithesis, founded in 2018, set about making the technique available more broadly. Its purpose-built hypervisor supplies a controlled computing environment for existing software. The testing machinery can vary conditions and investigate the resulting behavior without requiring every developer to reconstruct FoundationDB’s particular engineering discipline.

Wilson’s explanation of the problem is refreshingly inconvenient: people are good at writing tests for situations they can imagine. Software will eventually meet situations they cannot. A successful test therefore offers useful evidence, while leaving a rather large world outside the examination room.

By September 2024, Wilson was explaining another benefit: the Multiverse Debugger could let a developer inspect an earlier point in a simulated execution. If the useful evidence had disappeared by the time an incident was noticed, the engineer could revisit the state before it vanished. His recurring metaphor was a time machine. For once, the marketing department had been handed something the engineering department could demonstrate.

The bug that acquired an alibi

Antithesis had its own education in the cost of uncertainty. Early on, machines crashed only once in several thousand machine-hours. The team debated whether to accept the nuisance and build around it. Their testing service could tolerate occasional delays. The more troubling possibility was what an unexplained failure would do to their habits.

A familiar mystery provides a ready explanation for the next mysterious crash. Eventually, fixing this one revealed other problems that had been hiding behind similar symptoms.

Wilson and FreeBSD maintainer Colin Percival later recounted the investigation. An apparently incidental spreadsheet column exposed a pattern in machine uptime. A network-interface reset, connected to recurring DHCP activity, helped explain the fleeting error. The circumstances were brief; the investigation was anything but.

The story makes Wilson’s interest in reproducibility feel rather less abstract. These were the costs his company wanted to remove, encountered in its own machinery before it could fully turn that machinery upon itself. Even the people building the replay button had to endure a period without one.

Mario enters the laboratory

Wilson has a useful instinct for explaining difficult work through something people already enjoy. In a talk released in April 2024, he used Super Mario Bros. to explain how Antithesis explores possible sequences of events.

The platform found a route through the game in about 45 minutes on a 2018-era workstation CPU. Its hints were limited to memory locations for Mario’s coordinates and the level number. The point was to demonstrate exploration, rather than a human player’s knowledge of where to jump.

For an audience accustomed to hearing about distributed systems, the plumber was a welcome change of company. For Wilson, a game supplied a visible way to discuss a process that otherwise unfolds somewhere inside a computer, with very little regard for an audience’s entertainment.

The trouble with being a secret

Antithesis emerged publicly in February 2024 after more than five years in stealth, announcing $47 million in seed funding. The long private development period gave the team room to work on difficult infrastructure. It also required unusually patient relationships with investors, recruits and early customers.

The company stayed in Northern Virginia. Wilson has spoken about the region’s engineering talent and the value of retaining colleagues over long periods. For this particular project, continuity mattered.

Going public changed more than he expected. Sales became easier when a prospective customer’s internal champion no longer had to explain the absence of a website. Recruitment expanded. The company’s own people acquired a public account of what they had been doing.

One early employee told Wilson that his mother had assumed he had secretly been unemployed. It is a splendidly domestic verdict on years of ambitious research. A public launch can apparently serve as proof of employment.

Wilson’s retrospective treated these surprises seriously. External communication also helps employees understand the company they belong to. Silence had protected concentration, while charging a price in shared understanding.

Learning to sell less help

Wilson’s ambitions extend beyond the large organizations able to buy a substantial testing engagement. In February 2024, he announced an open-source giveaway program, acknowledging the volunteer maintainers whose work supports so much commercial software.

The enterprise product initially depended heavily on a services team. That support helped customers succeed, but it also made small contracts difficult to justify. In April 2025, Wilson described a revealing renewal conversation: a customer wanted the services portion removed because they were already using the platform independently.

The product had matured enough to surprise its own co-founder. A subsequent experiment with technical founder Carl Sverre helped inform a Founder Package for smaller teams prepared to manage some rough edges.

Another adjustment came in May 2025. Wilson argued that newly introduced bugs are often easier to understand while the changes behind them remain fresh. Antithesis’s Findings interface was designed to distinguish new problems from familiar ones. Finding more errors is useful only if the person receiving the report can decide what to do next.

When the customer buys into the company

In December 2025, Antithesis announced a $105 million Series A led by Jane Street. The trading firm was already a customer. The financing connected Wilson with an investor that had practical experience of the product, alongside an interest in the company’s future.

“More of you should be hiring your customers.”

Will Wilson, December 2025

Wilson’s account lingered on another kind of conversion: customers becoming employees. People who had used Antithesis from the outside brought knowledge that the builders could not obtain simply by looking harder at their own work.

There is a recognizable pattern here. He keeps returning to relationships that carry accumulated evidence: colleagues from FoundationDB, customers who join the team, a customer who becomes an investor. Trust appears repeatedly in his explanation of how the business works.

A company he still wants to work in

By March 2026, Wilson was writing about the difficulty of keeping that trust as Antithesis grew. He published a letter to the company under the title A fighting retreat. His aim was to preserve its character as long as possible, accepting that growth would exert pressure on it.

His examples were deliberately ordinary. Does the office need formal procedures for sharing electric-vehicle chargers, or can an informal group keep managing? Could a process wait until the problem actually required it? Bureaucracy rarely introduces itself as bureaucracy. It arrives carrying a sensible proposal.

Sean T. Allen, who joined Antithesis after years of intermittent contact with Wilson, offered a personal view that May. He remembered someone steady in online conversations, with little appetite for posturing. On joining the company, he recognized similar values in its working culture.

Wilson continues to write about technical problems too. His April 2026 essay on skiplists grew out of Phil Eaton’s book club and an Antithesis data-querying problem. The former mathematics student still seems happy to spend time with an obscure structure until it becomes useful.

Outside work, his official biography lists sailing and hiking with his children. He has cheerfully admitted to enjoying sailing without being especially good at it. Some uncertainties, apparently, remain recreational.

His younger self worried that all the worthwhile problems had gone. The intervening years have supplied corrections in abundance. For the moment, Wilson has software to test, a company to grow, and a working culture he would rather not surrender without a fight.