This week: Claude Opus 5 launches, Kimi K3 open weights arrive as the largest release ever, Nvidia backs open models, and the EU AI Act deadline nears.
Two things happened in dev tooling this week worth poking at: AI agents got their own runtime process outside your editor, and Ruby's JIT quietly closed a performance gap everyone assumed was permanent. Two interactive labs.
CrewAI wires role based AI agents into a crew that finishes a real job. Setup, 7 DevOps and security workflows, cost tips, and the gotchas nobody mentions.
A Ruby core committer ran Claude Code 600 times across 13 languages to build a mini Git clone. Dynamic languages came out cheapest and fastest — and it lands right as GitHub ships usage-based agent billing.
How AI agents automatically collect and organize SOC 2 and ISO 27001 audit evidence year-round, so compliance teams stop scrambling before every audit.
the shed // metrics & monitoring Build a real Prometheus /metrics endpoint in pure Ruby — no prometheus-client gem, no […]
the shed // config management A ~180-line, dependency-free configuration engine that only touches the filesystem when reality doesn’t match what […]
the shed // windows & wmi Audit real BitLocker drive-encryption state via WMI — not the policy setting, the actual […]
This week's AI roundup: Nvidia's new agent security alliance, Microsoft's cyber AI model, an Apple-Nvidia market cap flip, and a sovereign AI mega-deal.
Agent-to-agent handoffs between CI/CD stages carry full context forward, so failed builds stop starting root-cause hunts from zero every single time.
This week's AI roundup: OpenAI's rogue agent hack, a Teams phishing wave, record vulnerability counts, MCP goes stateless, and a chip stock selloff.
Cursor is an AI-native code editor built for agentic development. Setup, 7 DevOps and security workflows, pricing tips, and gotchas nobody mentions.