If your self-service form has eighty-nine questions on it, you haven't designed a decision surface, you've handed off a configuration job. The five-not-eighty-nine test: can the business owner read the surface in five minutes and know what they're choosing? Here's how to apply it.
Three quarters of incident debugging time used to come from not knowing what the AI did. Logging every tool call into a queryable store changed that, small infrastructure investment, outsized return.
Microsoft Ignite reframes the stack around agents. Google ships Gemini 3 and Antigravity. The $45B Microsoft-Nvidia-Anthropic circular deal. Nvidia prints another record. Brussels proposes an AI Act delay. UPS cuts 48k jobs. A loud week.
After fifteen accounts and three cycles of OAuth auth fatigue this year, I rebuilt the office stack mostly with open-source self-hosted alternatives. Worth being concrete about what works, what doesn't, and what the trade-off curve actually looks like.
Destructive operations get a second look. Soft-delete by default. Backup before mutation. The bias is toward keeping the thing, and that bias gets stronger, not weaker, when an agent is the one running the command.
Most of the AI conversation centers on the largest models. The interesting work in 2025 happens at the small end of the spectrum, where a handful of 1-8B-class models are doing more useful daily work than the marketing acknowledges.
OpenAI surprise-launches GPT-5.1 with a tone-picker. Anthropic reportedly commits $200B to Google Cloud. Moonshot's Kimi K2 Thinking, a trillion-parameter open-weight model that beats GPT-5. Microsoft Ignite warming up. My take on the week.
The legal framework for treating accumulated organizational knowledge as a real asset is older than the AI conversation and incomplete in ways the AI conversation makes more visible. Worth pulling on the thread.
When something is broken you get three attempts to fix it before you must stop and re-plan. Why 'just one more try' is the failure mode, and why the rule matters most when an agent is the one trying.
The NVIDIA device plugin, node taints and tolerations, the multi-tenancy reality of one GPU per pod vs MIG vs vGPU. The scheduling primitives that work at homelab scale on a single 1080 Ti and the ones that change shape entirely at fleet scale.
DSPy lets you optimize prompts the way you'd optimize a model, programmatically, against an objective, with structured evaluation. The default DSPy workflow sends a lot of prompts to a hosted optimizer. The local-first version is doable, faster than expected, and worth the setup.
OpenAI-AWS $38B makes multi-cloud official. Moonshot's Kimi K2 Thinking is the first open-source model to beat GPT-5 on agentic benchmarks. Apple reportedly pays Google $1B/yr for Siri. Ironwood TPUs ship. The concentration story cracked.
