Practical ways to evaluate, adopt, and operationalize AI tools.
Read the latest ThinkFast notes in ai.
Shadow AI is already in your company. So is the bill.
Pax8 says 35% of AI-engaged SMBs found shadow AI and 49% got a surprise AI bill. A CFO's one page inventory for agents, access and who owns the spend.
Don't give agents the checkbook without a ledger
Fragment raised a $12M Series A so AI agents cannot move money the books cannot explain. A CFO's view on agent ready accounting and where approval stays.
Claude Sonnet 5.5: price the workhorse, not the flagship
Claude Sonnet 5.5 keeps $2/$10 per-million-token pricing, half of Opus 5.5. Budget AI by cost per finished job, not by which model scores highest.
OpenClaw Enterprise makes the agent control plane free
OpenClaw Enterprise is a free, MIT-licensed control plane for persistent AI agents. The license cost of governance is now zero. The discipline is on you.
Agent commerce gets a rulebook: sell to bots too
The Personal Agent Protocol from Meta, Sierra, Walmart, Stripe and others sets rules for AI agents that shop for people. Here is what sellers should do now.
TypeSafe’s Jev: decision AI for invoice ops, not chat
TypeSafe AI launched Jev, a System One model for typed decisions. For operators, the AI accounting angle is AP automation with confidence gates.
Coding agents need a Herdr spend/PII control surface
Before a company-wide Claude Code and Codex rollout, put the fleet under Herdr so you can see state, attribute spend, and attach guardrails.
Enterprise AI harnesses beat model wars
Mid-market buyers should diligence the AI harness (context, governance, model swap, control plane), not the demo model. Salesforce set the frame on Sep 10.
Local Mac coding agents: when Claude Code still wins
Wire Hugging Face models through LM Studio into OpenCode or DeepSeek Harness. What M5 Pro vs Max changes, and when Claude Code still wins.
Too many AI design tools: how to audit your stack
We audited 41 AI design tools on one machine and found 14 of 20 jobs contested. Here is the repeatable process for cutting a stack down to what works.
AI-Driven Scaling: Workflows for Core Operations
AI for small business operations works best when applied to repeatable drafting, summarizing, and triage inside a clear operating rhythm.
Give your AI a second brain (and five operating rules)
One file and an Obsidian vault give my AI assistants a persistent second brain and five operating rules that survive every session reset.
Generate images in Claude with a free model council
Claude has no built in image generator. Here is how to generate images in Claude by driving free OpenAI and Gemini CLIs in parallel.
How to Evaluate AI Workflows Before You Buy
A practical checklist for separating durable leverage from expensive novelty, with a six-factor scorecard you can run in under an hour.
The practical AI stack: tools worth standardizing first
Where small teams should put AI before chasing bespoke automation, using a three-layer stack that compounds instead of sprawling.