Antithesis, an autonomous software verification company, has demonstrated how an AI coding agent can modify its own code. Until now, you couldn’t trust AI agents to check your work. Antithesis allows AI to self-correct as it writes code, eliminating a major bottleneck to AI adoption.
Since its launch, Antithesis has provided thorough verification of complex software. A set of tools announced today will enable AI agents to use Antithesis without human intervention. If the agent is unable to fix the error, Antithesis alerts a human developer to the problem and recommends a solution.
As engineering teams around the world are discovering, the use of AI has shifted the bottleneck in software development from code creation to verification. It takes just seconds for AI to generate code, but it takes days or weeks to review, test, validate, and build trust in the code. Given the speed at which AI operates, it is impossible for human developers to verify its output as soon as it is generated.
Also read: AiThority Interview with Glenn Jocher, Ultralytics Founder and CEO
AI coding agents require some form of thorough, independent, and unspoofable verification. Without fundamental changes in the way AI models are built, it is unlikely that AI will be able to create completely reliable software. They will continue to hallucinate, misinterpret their mistakes, and attempt to cheat by deleting tests, etc., making external validation essential. This goes beyond safety-critical systems such as flight controls, banking software, and subway signaling systems. This is equally true for programs that users habitually use, such as chat apps, design software, and even large-scale games.
Until now, code, whether written by humans or AI, had to be reviewed and tested using inadequate and outdated tools and methods. Even before the advent of AI coding, software testing was an inexact struggle for human developers. Depending on how you look at it, nearly half of software development time is spent testing and debugging, and yet unknown unknowns can slip through and lead to nasty and costly outages.
Even today, even as software becomes more complex, most organizations still rely on techniques that only capture surface-level problems and cannot reliably reveal the underlying emergency behavior that causes outages, data corruption, and cascading system failures.
Antithesis removes validation obstacles and improves testing. With Antithesis, developers know they can trust the code that AI generates, allowing them to use AI in areas that were previously too risky. This missing component delivers the productivity gains that AI has long promised.
“Today, we are taking a major step towards solving the validation gap that has hindered the potential of AI coding,” said Will Wilson, CEO of Antithesis. “Without rigorous validation, AI tools only create new bottlenecks that require humans to painstakingly test and review the results. Our universal trait-based testing and deterministic simulation technology can solve this problem in a practical way today.”
Late last year, Antithesis announced a $105 million Series A led by client Jane Street. Jane Street is a global technology-driven quantitative trading firm known for building the world’s most advanced software systems. This investment highlighted the emergence of Antithesis as a critical infrastructure for companies operating complex distributed systems. The funding will be used in part to accelerate Antithesis’ product innovation, and this advancement is a meaningful step in that direction.
Antithesis is a fundamentally new way to test and validate software before release. Run fully deterministic, massively parallel simulations to test years of real-world operations in hours. Antithesis intelligently explores the nooks and crannies of a customer’s codebase and strategically injects common failures to ensure that the system always behaves as intended. The platform fully reproduces bugs found in your own environment for rapid debugging. The company is based in Northern Virginia and was founded in 2018 and launched from stealth in 2024.
Also read: The infrastructure war behind the AI boom
[To share your insights with us, please write to psen@itechseries.com]
