Insights
Notes from the work.
Practical writing on building enterprise AI that reaches production.
Gemini 4 Argon is built for agents that run for hours. Test what happens in hour three
Google's frontier model targets long autonomous work, at an introductory price that doubles on December 1. A million-token window does not show how it behaves deep into a task.
Read
GitHub lets enterprises set Copilot's default model in JetBrains IDEs. A default is not a policy
Copilot for JetBrains now takes an enterprise-managed default model and a setting to stop MCP servers starting automatically. Useful, but neither restricts what a developer may use.
Read
Microsoft's leaders want an emergency brake for AI. Stopping an agent is the easy part of an incident
Nadella asked for a way to pause models mid-task, Brad Smith for mandatory incident reports, the White House for disclosure. The brake is the simple part. Enterprises lack what comes after.
Read
Collibra now reads agent quality signals from four AI platforms. A shared dashboard is not a shared metric
Collibra 2026.10, in pre-production since October 11, reads pass rates and token use from agents on Databricks, Bedrock, Gemini Enterprise and Azure AI Foundry. Comparing them is harder.
Read
Tulip's agent builds factory apps after an engineer approves its plan. In GxP plants, that gate is the product
Tulip's AI Authoring plans manufacturing apps from SOPs, drafts changes after an engineer approves, and adds review agents and code-based apps for GxP. None of it is generally available.
Read
When coding agents keep misusing your API, test your docs. Microsoft's AX playbook shows how
Microsoft published its Agent Experience Practitioner Playbook, a method for evaluating what coding agents do with an SDK, its docs and its skills. Internal platform teams need it most.
Read
Databricks' AI gateway policies can now read why a request was made. The caller writes that answer
Unity Gateway policies can now check caller-supplied request tags before a model or MCP tool call, in beta. Useful for attribution and routing. Databricks says a tag alone proves nothing.
Read
Stop comparing token prices with GPU hours. Price inference per completed task, over the hours you pay for
A GPU that lost money over a week, an agent made 29x cheaper on the same model, a decision model that bills only input. Our framework for choosing how to buy inference.
Read
Mashreq gives coding agents its integration standards, not its payment rails. Inside a bank's IBM Bob pilot
Mashreq reports up to 60 to 70% less time on integration design and troubleshooting in an IBM Bob pilot, with agents reaching systems through an MCP gateway. What does it enforce?
Read
A $2-an-hour H100 earned four times its cost in a busy morning. Over the week, it lost money
A Microsoft engineer served Qwen3.8-27B on an Azure Spot H100. Busy hours paid four to one; the full week did not cover the compute. Utilization, not the GPU price, decided it.
Read
Microsoft's Decision-1 prices a decision at $0.042 per million input tokens. Calibration is what you buy
Microsoft-Decision-1, in public preview in Foundry, returns a probability for each predefined option instead of text. The question is whether its confidence holds on your data.
Read
Google stopped selling Gemini Code Assist on October 9. Your renewal date is the real decision date
New Gemini Code Assist Standard and Enterprise subscriptions ended October 9. Existing ones keep working, but seat additions and annual renewals stop after January 31, 2027.
Read
Have an AI use case stuck between prototype and production?
Tell us what you’re trying to ship. We’ll reply with honest next steps.
Discuss a use case