THE COLLECTION
EVERY SIGNAL, SORTED.
Search the whole shelf by topic, tool, term, or plain old curiosity.
France, Mistral AI, Managed Agents, Agents SDK, Durable Execution, Stateful Sessions, Human Approval, Sandboxes, Idempotency, Tracing, Privacy, Reliability and Enterprise AI
Mistral made agents durable across crashes. Make resumability visible
Mistral says its new agent runtime can keep a task alive for hours or days and resume from the last finished step after a crash. That convenience creates a new question: what happened between the checkpoint and the restart?
PyTorch 2.15, Distributed Training, NCCL, GPU Clusters, Reliability Engineering, Rollback Testing, Flight Recorders, Distributed Checkpoints, CUDA, ROCm and Production Infrastructure
PyTorch 2.15 rewrote the GPU handshake. Test the rollback before the cluster hangs
PyTorch made its rewritten NCCL backend the default for distributed GPU jobs. The release includes hang diagnostics and temporary rollback switches, but the safest migration begins with a canary that proves both paths.
Global Open Source, Research Agents, Reliability, Workflow Recovery, Subagents, Observability, Token Costs, Scientific Research and Software Maintenance
Feynman kept finished research alive after a crash. Agent reliability is getting less glamorous
The open-source research agent now preserves results from subagents that finished before the parent workflow died, trims repeated prompt overhead and gives a damaged authentication file a useful error. None of that looks like magic. That is precisely why it matters.
AI Agents, Reliability, Business Workflows, Databases, Evaluation and Human Oversight
The agent said done. The database kept a different story
ThinkingBox reruns 507 business tasks twenty times and grades the records an agent leaves behind. The cheerful final answer is no longer the evidence.
Agent Memory and Reliability
Changing the model can quietly scramble what your AI remembers
A controlled study found that an agent can keep the same stored memory and still retrieve or interpret it differently after its underlying model changes.
AI Evaluation and Reliability
The benchmark judge changed its mind overnight
A preregistered audit sent identical work to language-model judges and received materially different rankings. The ruler was moving while everyone argued about the measurement.