Category
DeepSeek proved reasoning is teachable to small models with plain fine-tuning on worked traces. For agent builders, that collapses the reasoning-model premium and resets what counts as capability.
Lila Sciences runs a warehouse of AI-guided robotics doing experiments 24/7. It's the clearest sign yet that agents don't have to stay trapped in files and API calls, and the economics of that shift are the story worth studying.
Three years after the term 'AI engineer' was coined, the discipline has a tested playbook. Here is what the shift from prompting to agent harnesses tells you about where autonomous agent work is heading.
A maintainer's GitHub activity jumped when Opus 4.8 and GPT-5.6 shipped. That commit chart is a better read on where agent labor is moving than any benchmark.
A developer used a consumer agent to review and ship a major open-source release for about $149. That number is the story: the marginal cost of software maintenance just repriced.
Dylan Field's Figma is embedding AI as an agent-assisted layer rather than a replacement engine. The choice reveals the real strategic question facing enterprise software: which parts of the workflow does the human keep, and which does the tool absorb.
Meta's new work treats data creation as an agentic process rather than an upstream chore. If it holds, agent capability growth becomes self-reinforcing, and the competitive map of AI training shifts.
Self-driving labs and Qwen's jump from screen to robot arm both cross the same line: from describing the world to changing it. Here is how to find where your own agents sit on that line, and whether you put them there on purpose.
Claude Fable 5 spots problems and fixes them without being asked. That shift from reactive assistant to self-directed problem-solver moves the work of oversight from giving instructions to setting boundaries.
When Simon Willison built a new agentic editing plugin, he didn't reinvent the wheel. He copied Claude's. Here's what that tells you about where the real value in AI agents lives.
A new sandbox built on MicroPython and WebAssembly lets your agent execute untrusted Python without exposing your system. Here's why it matters for autonomous agents, and where it still leaks.
On-demand capability loading in Pydantic-AI v1.105.0 is being sold as a performance feature. It's actually an admission that the monolithic-agent pattern doesn't survive contact with real users.
OpenClaw's move to modular plugins exposes a critical tradeoff: flexibility versus dependency hell, with implications for security and scalability.
The critical SQL injection vulnerability in Strapi's content-type builder is not just a code flaw but a symptom of systemic weaknesses in AI agent security architectures.
The TanStack malware incident exposes fundamental cracks in the trust model of package ecosystems, forcing a reevaluation of how we secure software supply chains.
The discovery of OpenClaude's sandbox bypass vulnerability signals that traditional sandboxing approaches may no longer be sufficient for securing AI agents in production environments.
Command injection flaws are increasingly exposing AI agents to systemic risks, forcing a fundamental rethink of how agent runtimes handle untrusted inputs.
Anthropic’s Claude Code team advocates for HTML as the preferred output format over Markdown, signaling a broader shift in how AI agents structure and render content.
Hermes Agent’s v0.13.0 release, dubbed 'The Tenacity Release,' signals a critical shift in agent design priorities from ephemeral task execution to durable, fault-tolerant workflows, reshaping the competitive landscape for multi-agent systems.
Anthropic's partnership with SpaceX for Colossus GPU access signals a strategic pivot: AI's next frontier isn't better models, but compute dominance at scale.
