Anthropic's Claude formally proves Fermat's Last Theorem in 11 days using AI agents

·1 views

Anthropic's AI model Claude completed a machine-verified formal proof of Fermat's Last Theorem between August 7–17, 2026, a task that mathematician Kevin Buzzard of Imperial College London had secured a five-year research grant to accomplish. The proof was produced through the Prove2Me platform, where dozens of AI agents autonomously divided work across a shared theorem dependency graph, generating 13 million lines of Lean code and verifying over 29,500 sub-theorems. Buzzard reviewed and independently compiled the entire output, confirming it passed verification, though he noted the proof follows the 1990s Darmon-Diamond-Taylor approach rather than modern methods, and covers only prime exponents p ≥ 17. Human researcher involvement was minimal, limited to occasional high-level guidance, while the agents largely self-corrected errors and cross-checked results autonomously. The achievement marks the first fully machine-verified proof of the theorem, which was originally conjectured by Pierre de Fermat in 1637 and first proved by Andrew Wiles in 1995 across a 129-page paper.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Developer Ditches OpenSpec for AIDLC Workflow to Fix AI-Generated Code Quality

A software team lead spent over a month overhauling his team's AI coding workflow after noticing that while development speed had improved over six months, code quality remained poor and bug-fixing often took far longer than initial development. The team had been using Spec Driven Development (SDD) frameworks like OpenSpec, but found these tools difficult to standardize across a multi-team, long-running project. Complex commands and inconsistent adoption led some developers to abandon the workflow entirely and revert to unstructured 'vibe coding.' The lead replaced OpenSpec with a new approach called AIDLC, which incorporates a more detailed software development process and stronger team collaboration practices. According to the author, the switch resulted in a significant improvement in code quality across the team.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How Dependency-Driven Structure Can Improve Game Wiki Troubleshooting Guides

A developer essay on DEV Community proposes a structured documentation model for game wiki guides, using the business simulation game Big Ambitions as a reference case. The model suggests organizing guides around explicit dependency relationships, where each prerequisite is described by its condition, its connection to other workflow elements, what readers can observe, and where to go next if something is missing. The author recommends dividing troubleshooting pages into sections with distinct responsibilities, such as summary, prerequisites, procedure, verification, and diagnostics, so readers can navigate directly to what they need. A dedicated verification section is emphasized to define what success looks like, rather than leaving outcomes ambiguous after a procedure ends. The piece also advocates storing version context and source evidence alongside factual claims, giving editors a clear way to maintain accuracy as game content changes over time.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Agent hook scripts fail open by default, posing silent security risks

Agent-CLI hooks, small programs that run before or after tool calls in AI coding harnesses like Claude Code, default to a fail-open state when they crash or return malformed output. This means a hook that throws an exception or receives unparsable input may silently allow a tool call to proceed rather than blocking it. The author of Handrail, an open-source hook pack, argues the fix must be structural: a top-level error trap that routes any uncaught failure to an explicit deny decision. The recommended pattern uses a narrow allow-list so that only known-safe cases produce an allow response, with everything else — including unknown or unparseable inputs — defaulting to deny. The Handrail toolkit, available under the MIT license on GitHub, packages this fail-closed design along with a fixture test suite and installer for Claude Code settings.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Why Engineering Leads Should Never Write Code, According to One Developer

Software engineer Anton, who works primarily with PHP/Symfony and Go, argues that a technical lead's role should be strictly limited to preparing task facts, handing over work, and accepting results — never writing code directly. His workflow caps concurrent executors at three: two on main work and one on fixes, deliberately avoiding fan-out to prevent file conflicts and limit bottlenecks. According to Anton, the moment a lead starts coding, the critical fact-preparation step breaks down, making subsequent task batches more costly than any time saved. Each task must be handed to an executor as a complete, self-contained message requiring no further reading or clarification. After a set of tasks closes, a structured audit maps every original requirement to what was done and what proof exists, flagging any logic added outside the original statement.

0 comments Read more at DEV Community