Why AI Agents Double-Charge Customers: The Retry Logic Flaw Developers Miss

·1 views

A widely overlooked infrastructure bug in AI agent frameworks can cause duplicate financial actions, such as issuing a double refund, when a network timeout or dropped connection makes the outcome of an API call ambiguous. Unlike a simple success-or-failure model, distributed system calls have three possible outcomes: committed, failed, or ambiguous — and most retry logic incorrectly treats ambiguous as failed. The flaw is not caused by faulty AI reasoning; the model may have acted correctly, but the surrounding code blindly retries on any exception. The safe approach requires atomically reserving an action's unique key before making the external call, then marking it as ambiguous — not releasing it — if a timeout occurs. This principle, long established in databases via two-phase commit and in payments via idempotency keys, has yet to be widely adopted by agent frameworks that are already taking consequential real-world actions.

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)

.NET 10 Changes Shift Behavior for Small Integer Types in Generic Math

Microsoft's .NET 10 has altered how oversized bit-shift operations behave for small integer types like byte and short when dispatched through generic math interfaces such as IShiftOperators. Unlike concrete C# expressions that promote operands to int, generic methods constrained by IShiftOperators return the original type T, meaning shift counts equal to or exceeding the type's bit width now produce different results than in .NET 9. Developers maintaining libraries that target multiple runtimes should add cross-runtime tests covering shift counts at and beyond the type width to catch behavioral differences. Two mitigation strategies are available: rejecting out-of-range counts via ArgumentOutOfRangeException for inputs like parsers or authorization bitsets, or applying a modulo operation on the count where wrapping behavior is an intentional part of the contract. Microsoft recommends keeping cross-runtime assertions in place until all affected older runtime targets reach end of support.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

OpenAI Claims AI-Assisted Breakthrough on Navier-Stokes Millennium Prize Problem

OpenAI published a formal write-up on September 8, 2026, claiming an internal AI system produced an analytical proof that three-dimensional incompressible Navier-Stokes equations can develop a finite-time singularity — a long-unsolved Millennium Prize Problem. The internal model, described as significantly more capable than any publicly available system, ran approximately 10,000 concurrent agents for around 88 hours to generate the core proof. GPT-6 Astra then completed the Lean formalization — converting the proof into a machine-checkable format — in roughly 17 additional hours. OpenAI also acknowledged concurrent related work by Tristan Buckmaster of New York University and Levent Alpöge of Anthropic, indicating it intends to recognize their priority. The result represents a coordinated AI research workflow rather than a single public model independently solving the problem, and the underlying mathematical claims still await broader expert scrutiny.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

IonQ Launches Superion 256, a Rack-Compatible Quantum System Built for Mass Production

IonQ has unveiled the Superion 256, its sixth-generation quantum computing platform, designed to move quantum hardware from laboratory settings into commercial data centers. Built using trapped-ion technology and standard semiconductor fabrication in partnership with foundry SkyWater, the system fits within a standard server rack and consumes less power than a typical GPU rack. The shift to foundry-based production cut the design cycle from nine months to two months, enabling twelve times more wafer lots to be processed in a six-month period. The platform uses Electronic Qubit Control to manage qubits directly on-chip, replacing complex laser systems with standardized electronics. IonQ also has the Superion 10K in development, which aims to integrate CMOS technology onto the quantum chip and scale toward millions of qubits using the same underlying architecture.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Myna 0.15.0 adds self-update command with checksum verification and safe staging

The myna CLI tool has released version 0.15.0, introducing a built-in 'myna update' command that allows the binary to update itself in place. The update process performs numerical version comparison to avoid ordering errors, selects the correct architecture-specific build before downloading, and verifies the file against published SHA256 checksums before installation. To ensure safety, the new binary is staged in the destination directory and atomically renamed over the existing file, leaving the original untouched if a download fails. A '--check' flag exits with code 1 when an update is available, enabling scriptable cron-based checks, while '--version' allows installing any specific release in either direction. Users on version 0.14.0 or earlier must still use the curl installer for the initial upgrade to 0.15.0, after which 'myna update' becomes the standard upgrade path.

0 comments Read more at DEV Community