Agentics

OpenClaw just shipped a CVE in reverse: instead of fixing a breach, it's locking down the defaults that made breaches possible.

Autonomous AI agents fail at 70% of basic workplace tasks, and 84% of enterprises cite governance risk as the barrier. Asana answer: agents that live inside work management infrastructure, not standalone chat. The catch: the beta numbers come from the already-converted.

410 million ChatGPT data violations in one year. 18,033 terabytes uploaded. 100% of AI systems with critical flaws. The enterprise AI security gap is not a future problem.

A SQL injection let a red team poison McKinsey Lilli with one HTTP call. The security tools meant to stop this class of attack do not exist yet — and 40% of enterprise apps will have agents by 2026.

Beijing banned it from government devices. Alibaba, Tencent, and Baidu deployed it anyway. The OpenClaw story is about who builds foundational infrastructure — and who is too afraid to.

Google shipped a platform where AI agents can extract their own service account credentials by default. Palo Alto found the proof-of-concept. The fix is a documentation footnote. Sound familiar? It should — this is the AWS IAM pattern, fifteen years later.

Microsoft runs 1,300 Azure SRE Agents that have mitigated tens of thousands of incidents across its own infrastructure. AWS shipped the same thing three weeks later. Both use MCP to reach each other's infrastructure without bespoke connectors. That is how a protocol becomes plumbing.

JetBrains spent years building AI pair programming tooling. Now it is retiring that product and replacing it with something structurally different: not a better AI, but a governance layer for autonomous agents. The architect of the pair programming era is betting the bottleneck has moved.

When agents explore in parallel, they find more fixes faster — but keep repeating the same ones. When they hand off to specialists first, they find fewer fixes but deeper ones. The authors call for routing between these modes at runtime.

The holy grail of design tools was 'what you see is what you get.' Figma just shipped 'what you see is what runs.'

Multi-agent debate systems are often credited to model capability. A new study isolates the protocol effect: RA-CR converges faster for consensus tasks, but argument diversity stays constant whether agents interact or not. Always-on debate is not automatically the right answer.