Insights
Notes from the work.
Practical writing on building enterprise AI that reaches production.
Build or buy now applies to retrieval too. What Cohere's Compass Cloud and Embed 5 change
Cohere is turning its retrieval stack into a managed service with API, MCP and permission enforcement, plus a new embedding family. How to decide what to build and what to buy.
Read
Keep the deterministic parts deterministic: what Snowflake's AI accounts payable system gets right
Snowflake built InvoiceIQ, an internal accounts payable system that mixes AI with rules, audits every action and sends only real exceptions to people. A pattern for production AI in finance.
Read
Most enterprise RAG failures start in retrieval. Measure it as its own system
Cohere's RCP-nDCG@10 uses rubric-based, calibrated AI judgments to score retrieval when labeled test sets fall short. Why enterprise teams should evaluate retrieval separately from the final answer.
Read
Google's most useful AI agents don't chat. They sit inside code review and block vulnerabilities
Google describes scanning, triage and fix agents embedded in its development lifecycle, with deterministic checks and human review. What enterprises can copy from agents built into existing controls.
Read
AI spend now needs guardrails, just like permissions. Platforms are shipping them, and CFOs want a say
Snowflake made per-user AI quotas generally available, Google added spend caps for agents, and IBM reports CFOs taking a larger AI role. How to design cost controls before agents scale.
Read
Who holds what? Custody is becoming the first question in enterprise AI architecture
A month of announcements points the same way: agents create new copies of sensitive data in sessions, snapshots, traces and indexes. Why a custody map should come before the security review.
Read
Databricks is putting R$1.5 billion into Brazil. What it says about enterprise AI demand outside the US
Databricks plans to invest about US$300 million in Brazil over three years and train 150,000 people. Why Brazil is a core enterprise AI market, and what still depends on enterprises.
Read
An AI agent can return a 200 and still be wrong. Snowflake's Agent Observability measures what uptime can't
Snowflake is bringing agent tracing, cost signals and LLM-as-judge evaluation to Observe. Why production AI monitoring must cover behavior, quality and cost, and why the telemetry is sensitive too.
Read
Google's Agent Substrate is built for fleets of agents, not one. That changes what platform teams own
GKE Agent Substrate suspends idle agents to storage and resumes them in isolated sandboxes in under a second. Why agent execution is becoming a platform problem, and what it leaves open.
Read
Same Claude, different boundary: on Azure and AWS, the delivery path is the governance decision
Claude now reaches enterprises through Microsoft Foundry, Amazon Bedrock and several AWS-billed paths. The model is the same; data processing, identity and tool access are not.
Read
IBM Bob goes self-hosted. Where an AI agent runs is now part of the product decision
IBM made its agentic development platform available on premises, in private and sovereign clouds and air-gapped. What deployment topology changes for regulated engineering teams.
Read
When AI agents start maintaining AI systems, evaluation becomes the control surface
Palantir's AIP Evolve, now generally available, has agents propose model migrations, cost cuts and eval improvements. Why this only works with strong evals and human review.
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