The ‘Pain in the Ass’ Math That Could Save Vibe Coding From Itself
AI companies are trying to make formal software verification, which uses mathematics to check whether code meets precise requirements, faster and easier to use. The approach could improve software reliability and security, but it remains uncertain whether AI can make it practical beyond high-risk systems.
&Truth has raised $70 million and used AI to verify part of the code that encrypts Signal conversations; rival Harmonic has raised $295 million and says its Aristotle agent can verify C software and Verilog hardware designs. Anthropic and Theorem also used AI to find bugs missed by extensive testing. Experts caution that verification can be costly and cannot catch every problem, including those caused by user error or hardware failure.
In short
- Start-ups are using AI to make mathematical code checks more practical.
- &Truth raised $70 million; Harmonic has raised $295 million.
- Experts say verification offers assurance but cannot catch every bug.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
AI-assisted formal verification could make software safer by checking whether code meets precise requirements, rather than relying only on tests that cover selected cases. If AI lowers the cost and expertise needed, the method could reach beyond safety-critical systems and help catch security flaws before software is widely used.
The case against
Formal verification can be costly and difficult, and proving that code meets stated requirements does not show that those requirements are complete or correct. It also cannot prevent every failure, including problems caused by user error or hardware, so AI may make the process more accessible without making it practical or sufficient for ordinary software.
Explained simply
Software verification uses mathematics to check whether a program follows a precise set of rules. It differs from ordinary testing, which checks how software behaves in selected situations; verification aims to show that specified properties hold across all situations covered by the proof.
The work is attracting attention as AI tools make it easier for people to create software, sometimes without deep programming experience. Companies including &Truth and Harmonic are applying AI to verification, while Anthropic and Theorem have used AI to help find bugs. Signal, the messaging service, is among the organisations whose encryption code has been partly checked this way.
Verification can strengthen confidence in important software, but it takes time and money, and a proof only covers the rules and code examined. It cannot prevent every failure, such as problems caused by user mistakes or faulty hardware, so its wider use in everyday software remains uncertain.
Art Business Celebrity Companies Culture Entertainment Software Technology