Insights
Notes from the work.
Practical writing on building enterprise AI that reaches production.
Databricks' agents now reach the terminal, your SaaS tools and outside AIs. Each carries the user's identity
Genie Code CLI, Genie Agents' MCP connectors and the Genie One MCP server all run with the requesting person's permissions. When analytics agents act, data permissions become action permissions.
Read
Cornerstone cut database diagnosis from 45 to 10 minutes. Its agents never touch production alone
Cornerstone OnDemand's database agents gather evidence across SQL Server, Jira and dashboards, while destructive actions wait for a person and default to deny. An AWS pattern formalizes the split.
Read
AI-written code is a supply chain problem first. GitLab's governed software factory starts there
GitLab announced controls for dependencies, secrets, artifacts and AI spend around its agent platform. Each ships at a different maturity, and the order matters more than the headline.
Read
An agent's next permission can depend on its last action. AWS Strands Box makes authorization temporal
AWS released Strands Box in developer preview: an open source agent sandbox whose Dogwood policies allow or deny each action based on what the agent has already done.
Read
Claude Haiku 5.5 costs 90% less below 100,000 tokens. That line is now an architecture decision
Anthropic's Haiku 5.5 makes high-volume agent steps far cheaper, but only while a request stays under 100,000 tokens. Routing, compaction and caching now decide most of the bill.
Read
The next AI pilot has to prove operations and jurisdiction. NTT DATA opens a production test floor in Munich
NTT DATA is opening an AI Factory in Munich where organizations can validate AI architectures on production-grade infrastructure before committing. It shows what a serious pilot now has to prove.
Read
The cheapest context is the one your agent never loads. Google turns its developer docs into an agent API
Google's Developer Knowledge API serves official documentation to agents as searchable chunks, grounded answers and fresh Markdown pages. It is a pattern enterprises can copy for their own knowledge.
Read
Enterprise AI is becoming a sales channel, not just a feature. Radisson's ChatGPT plugin shows what that takes
Radisson and Accenture built a ChatGPT plugin for hotel discovery that hands bookings to Radisson's own site. The reusable part is the MCP and API layer behind it, not the chat.
Read
Not every agent decision needs a generative model. OpenAI's Decisions API prices the decision layer apart
OpenAI opened its Decisions API in public beta: GPT-6 Luna returns probabilities, choices or scores instead of prose, billed on input only. The decision layer is becoming a product category.
Read
For long-running agents, the useful limit is the environment, not the number of steps. Jump Trading shows how
Jump Trading lets GPT-6 Astra agents run long quantitative research loops on their own, inside a monitored environment with human-defined criteria and review before any signal is used.
Read
When an agent joins the team chat, context is both its strength and its risk. Cisco brings agents to Webex
Cisco will let teams @mention Claude Managed Agents and other agents into Webex spaces, meetings and calls. Shared context makes them useful, and makes access the first design question.
Read
The most valuable enterprise agents may never leave the transaction system. SAP's Joule rollout makes that bet
SAP is rolling out Joule Work this month and counts more than 200 Joule Agents that run inside its deterministic business workflows. The architecture matters more than the agent count.
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