Insights
Notes from the work.
Practical writing on building enterprise AI that reaches production.
BigQuery's AI.AGG summarizes millions of rows in one SQL query. Count the rows it quietly leaves out
AI.AGG, in preview, runs a map-reduce of Gemini calls over text and images inside BigQuery. Useful for logs and feedback, but some rows are skipped without an error, so check completeness.
Read
AlloyDB Express creates Postgres databases in under two seconds for coding agents. Review the upgrade button
Google's AlloyDB Express, in preview, gives developers and AI coding agents throwaway Postgres databases that scale to zero, then upgrade in one click. The risk moves from creation to promotion.
Read
Coding agents are leaving the laptop. AWS and Databricks show the new path: gateway, microVM, pull request
An AWS sample runs Claude Code tasks in isolated microVMs that return pull requests; Databricks routes Claude Code and Codex through its model gateway. Each fixes one control, not all.
Read
AWS now sells web research as one tool call. Your agent gets an answer and someone else's reading of sources
Web Answer on Bedrock AgentCore is GA: it searches an Amazon-operated index, reads pages and returns a cited answer for US$10 per 1,000 calls. When raw results are safer.
Read
The best visual AI assistant finished 53% of guided tasks in Salesforce's test. Speed was part of the failure
Salesforce's Interactive Visual Gym tests assistants that watch a user's camera and talk them through a task, with a simulated user who reacts. Leading models struggled, often by speaking too late.
Read
Most of what an AI agent does between thoughts runs on CPUs. NVIDIA's Vera pitch is a capacity warning
NVIDIA argues that sandboxes, databases, search and code execution make the CPU a bottleneck for agents, and cites partner benchmarks for Vera. The planning lesson holds whichever chip you buy.
Read
Five decisions an AI agent's model should never make. Give it everything else
Taxbit, Okta, AWS and a Microsoft reference architecture converge: the model proposes, deterministic systems decide identity, thresholds, amounts, order and the record.
Read
An agent reading yesterday's warehouse is wrong about today. Confluent and Databricks race to serve the now
Confluent's Streamhouse and Lightning Tables, and a Kafka-free CDC path from Debezium into Databricks, both aim to give agents current business state. Freshness belongs in the agent's specification.
Read
Google's new TPUs are built for confidentiality and memory, not just speed. Long-running agents set the brief
TPU 8tx adds hardware confidential compute for regulated customers, TPU 8i Max doubles memory bandwidth, and TPU Sync treats model state as infrastructure. What matters for enterprises.
Read
Microsoft closed the shared default environment to its own makers. Every low-code agent now has an owner
With hundreds of thousands of low-code agents and apps in its tenant, Microsoft moved makers into 170,000+ personal environments, with tiers, connector rules and Sentinel enforcement.
Read
Four in ten companies add flaws faster than they fix them. Google's CodeMender bets on fixes with a human gate
Google's security update adds sandboxed autonomous remediation in CodeMender, OIDC tokens for agents and topic rules in Model Armor. The bottleneck it targets is the fix queue, not detection.
Read
Microsoft Research proved an AI-written login correct for every input. What it could not prove is the lesson
PDD Kit turns a developer's request into a formal specification, has agents write code and proofs, and labels what remains tested but unproved. A research prototype with a useful habit.
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