What we write. What we read. What we give away.
Short notes from the field, practical guides you can run this week, and the outside sources we actually trust.
Latest notes
From the blog
Cutting Edge·Apr 23, 2026Computer-use agents need production wrappers
Browser and shell agents are ready for bounded work when scopes, screenshots, approval points, budgets, and receipts are built into the wrapper.
Operations·Apr 22, 2026MCP needs an operating model
MCP servers need inventory, scoped credentials, consent rules, logs, and receipts because tool access becomes a business surface fast.
Fundamentals·Apr 20, 2026Agents only pay when workflows change
AI adoption pays when a team changes the workflow: pick one clean input, define the output, measure accepted work, and keep review in the loop.
Guides
Playbooks you can run this week
Startup90 minutesLehigh Valley Business Startup Checklist
A local-first checklist for registering a Pennsylvania business, getting tax accounts in order, and checking city-level requirements in Allentown, Bethlehem, and Easton.
Preview guide
AI Operations45 minutesAI Readiness Checklist
A practical way to choose the first AI workflow worth testing before buying tools, hiring help, or automating the wrong work.
Preview guide
AI Operations50 minutesAgent Workflow Starter Map
A mapping guide for turning one recurring business process into a scoped agent workflow.
Preview guide
Recommended reading
What Alex actually reads
A short list we revisit. Primary sources over commentary, practitioners over pundits.
Anthropic Engineering
How the people building Claude think about prompt caching, tool use, and agent evaluation.
OpenAI Research
Primary source on model capabilities, safety research, and benchmark releases.
Google DeepMind Blog
Frontier research writeups that rarely make it into vendor marketing summaries.
Simon Willison's Weblog
Daily, skeptical, hands-on coverage of LLM releases with runnable examples.
Interconnects by Nathan Lambert
Research-grade analysis of the open-source model ecosystem and what the numbers actually mean.
Model Context Protocol docs
The spec we build against when we stand up agent tool surfaces for clients.
Rather skip the reading?
Thirty-minute working call. We walk through what fits your situation and leave you with a plan.