Code is exhaust. The bottleneck moved upstream: context, intent, judgment, and ownership.
The 2nd edition of The Bottleneck Is Never the Stack is out.
https://t.co/Cj3FR47rrB
#AgenticAI#EngineeringLeadership
A prompt can describe what an agent should do. It cannot establish what the agent is permitted to do.
That part belongs in the platform: scoped identity, short-lived credentials, network policy, and a record of what the agent tried and got denied.
https://t.co/3Dz4LkYWV9
I built AgentMachinist for solo work: one bounded task in, one reviewed local change out. You approve the exact spec before any code is written, and an independent review runs before you integrate.
Works with Claude Code, Codex, OpenCode, Pi, and Goose.
https://t.co/FzpWwsDtrf
Apple approved two of my apps two minutes apart on Tuesday. GSD Task Manager 3.6 adds Saved, a place to park a task with a note until you're ready for it. PourCraft Coffee 2.6.1 can start from the water and sets temperature by roast. Both are live now. https://t.co/kXBZzSj1NM
Black October packed three incidents into one week: compromised eMudhra certs, PHP2Shell, and a poisoned LiteLLM release.
Different causes, same weak spot. Can you turn a finding into a safe change fast, and prove it reduced exposure?
https://t.co/sAENdczjyq
System Headroom 1.5 (@systemheadroom) is out on the Mac App Store. My menu bar monitor can now tell you when an app hogs the CPU for too long, shows memory pressure and swap in the header, and gives every listening port a menu to open or copy its address. https://t.co/lXpMddeHKH
An agent that saves you money threatens somebody else's revenue. That's why Amazon blocks Meta's Muse agent while Walmart lets it in.
A company's agent policy tells you where its margin lives.
https://t.co/6b21kXcKuV
In September I asked Muse, Meta’s personal agent, to find subscriptions nobody in the house was using. It flagged two and recommended canceling both.
What stayed with me was the direction of the recommendation. An agent working for me had just found revenue somebody else was about to lose.
Twelve days after Muse launched, Amazon blocked it. Last week, five of the largest U.S. banks blocked third-party agents from customer accounts. Walmart, the same week, built Muse a connector. Visa and Mastercard built an ID check.
Same technology, four different answers. The essay works through why, and ends with the three questions I’d ask before calling any platform ready for independent agents.
https://t.co/fot8bjNw76
Audio edition on Never the Stack, 24 minutes.
When a company blocks your agent, what would the agent have skipped?
#EngineeringLeadership #PlatformEngineering #AgenticCommerce #AIStrategy
Black October showed AI finds flaws faster than ever. The bottleneck is remediation: ownership, judgment, delivery. My essay covers what broke and 5 changes that help.
#Cybersecurity https://t.co/sAENdczRnY
StillView 5.1 is out. Browse a folder and all its subfolders in one collection, then narrow it with the folder tree or a path search. Text recognition now works even with Apple Intelligence off. Free for Apple silicon Macs on macOS 27. https://t.co/3fhzMl6BqP
I'm attending All Day AI on October 22. Join me!
Free, all-day community event. 122 speakers from leading AI companies. 30-minute sessions across 4 stages, 6am–9pm ET. No vendor pitches—drop in for the sessions that interest you. https://t.co/5oUETgR8Ko
Prompts tell agents what to do. They can't enforce what's permitted. Enforce permissions in the workspace, identity, network, and tool layer — then test those controls directly.
#AgenticSDLC#AISecurity https://t.co/3Dz4LkYWV9
TypeSafe says Jev "can't hallucinate." That zero is a schema guarantee, not a measurement. A well-typed answer can still be wrong.
Jev returns decisions, not text — $0.042/M tokens.
https://t.co/CgmzWZtTkn
#AIStrategy#GenAI