Insights
Notes from the work.
Practical writing on building enterprise AI that reaches production.
Snyk's new security agents follow one rule: the agent that writes the fix never grades it
Snyk launched Evo Agentic AppSec, a set of agents that scan, review, block packages and fix code, checked by its deterministic engines. Most of it is in preview; the design principle is ready now.
Read
Microsoft wants Copilot to compile your team's knowledge for agents. Someone has to own the compilation
SharePoint knowledge bases, in Frontier preview, turn team emails, meetings and documents into Markdown that agents read. Faster answers, and a new layer that needs provenance rules.
Read
Tamnoon's cloud remediation agents never hold a credential. That is the design decision worth copying
AWS published how Tamnoon built agents that investigate cloud findings, score their safety and plan fixes on Bedrock, while execution runs elsewhere with short-lived, customer-controlled access.
Read
Block's security agents keep the raw evidence behind every investigation. That turned a rare task into an hour
Block uses OpenAI's Daybreak and its own harness for detection, investigation and patching. Its most transferable practice is storing query evidence so changes can be replayed against past cases.
Read
Anthropic went from 2,500 to 30,000 merged changes a day. Git's assumptions broke before Git did
Anthropic replaced Git internally with Scion after coding agents multiplied its merge rate. Most companies will never need Scion; the assumptions it dropped are worth checking anyway.
Read
Two model retirement notices in two days. November 16 and January 14 are now dates on your change calendar
OpenAI will remove GPT-4o and GPT-4.1 API snapshots on January 14, 2027. GitHub retires three Copilot models on November 16. What a model dependency inventory needs before either date.
Read
Meta's Muse Spark Open needs about a terabyte of memory. Downloading the weights is the easy part
Meta released Muse Spark Open, an 816B-parameter open-weight model for coding agents, on October 15. Four questions decide whether an enterprise should run it, rent it or leave it.
Read
GitHub's AI can now find, fix and retest vulnerabilities. It still cannot block a merge or approve one
AI Scan and Agentic Autofix reached general availability on October 15. One advises, the other opens draft pull requests. Read the limits as design: the pull request stays the control boundary.
Read
ServiceNow gives every AI agent a signed passport. What counts is who checks it at the door
ServiceNow's AI Passport, due in Q4 2026, turns an agent's authorized scope into a signed credential checked at runtime. The value depends on enforcement points the agent cannot bypass.
Read
AI security scanners find more problems. GitHub now lets enterprises see whether developers fix them
GitHub added enterprise-wide AI Scan metrics for pull requests: findings, fixes, remediation rates and autofix use. Useful for rollout decisions, if remediation is not mistaken for lower risk.
Read
Microsoft's own chips already run Copilot. Maia 300 cuts Microsoft's costs first; yours depend on the contract
Microsoft plans multi-gigawatt Maia deployments from 2027 and projects 40% more tokens per megawatt than GPU systems. Customers can't select the chips; they get capacity, price and terms.
Read
If an agent solved the same problem fifty times, why is it still reasoning? When agent work should become code
The best autonomous agents should make parts of themselves unnecessary. Five questions decide when a repeated agent task should become a tested playbook, and when the reasoning is worth paying for.
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