Insights
Notes from the work.
Practical writing on building enterprise AI that reaches production.
In Lithuania, agents produce 85% of ChatGPT Enterprise output; in Slovenia, 22%. The gap is depth
OpenAI's research on EU business customers finds agentic tools behind more than half of output tokens, with wide gaps between countries and firms. What the numbers can and cannot say.
Read
Your agents will not break the AI stack first. They will break the systems built for human pace
Anthropic's source control, Third Moment's backtests and Databricks' job scheduler hit the same limit: systems sized for people. A framework to audit yours before agent volume arrives.
Read
When agents run experiments, the slow part is around the model. Third Moment and Databricks show it
Third Moment runs 10,000 research agents for up to 48 hours; Databricks' researchers launched 8x more training jobs with agents. Both hit the same wall: waiting, launching and errors.
Read
AI agents enter chip factories through simulation, not the machines. NVIDIA's fab push shows the boundary
At SEMICON West, NVIDIA and chipmakers showed agents running physics simulations and digital twins before anything touches a tool. Each speedup is a different workload; none is a general benchmark.
Read
An SRE agent's diagnosis is a starting point, not a finding. AWS and Microsoft both built the handover
AWS connected DevOps Agent to Resilience Hub tests, and Microsoft put Azure SRE Agent investigations next to GitHub Copilot. Both treat the agent's analysis as evidence for an engineer to check.
Read
Does your RAG app need a separate vector database? Databricks measured search in Postgres at 4M passages
A Databricks engineer sized hybrid vector and keyword search in Lakebase Postgres with a real app and public corpora. The numbers give teams a way to decide where retrieval should live.
Read
Before AI security agents, Atlassian put 44 billion daily events into one schema. That is the order that works
Atlassian normalized its security telemetry to OCSF on Databricks and reports 80% lower ingestion cost. Its agent plans sit on top of that foundation, and most are still roadmap.
Read
Coding agents can read your repository. Figma Specs tries to give them the decisions behind the product
Figma launched Specs in open beta: Markdown requirements that live next to designs and code layers, readable by agents such as Codex through Figma's MCP server. Useful context, if someone owns it.
Read
GitHub's HydraFusion Max doesn't pick a model for your task. It plans who drafts, who judges and who repairs
GitHub updated HydraFusion and added a Max option, both in research preview. Models now get different roles and reasoning effort per stage. Measure it by cost per accepted change, not by benchmark.
Read
AI made producing a change cheap. Proving it is right is now the budget line nobody planned
This week's security releases share one pattern: generation scales, verification does not. La Madre's view on why verification capacity should be planned, funded and measured like compute.
Read
Reward an AI auditor for finding problems and it learns to invent them. Anthropic's research shows the fix
Anthropic trained a small model to audit AI systems with reinforcement learning. The naive reward produced invented findings; pairwise judging and benign examples fixed it.
Read
Chainguard's AI attacks your running app to prove a flaw is real. Evaluate the sandbox before the model
Chainguard Harness builds an app from source, boots it in an isolated microVM and lets Claude Mythos 5.1 try to exploit it. In AI security testing, the containment is the product.
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