Real uses, not pipe dreams.
Plain-language writing about AI that actually works — grounded in real projects, honest about limits, and free of hype. One concrete example and one clear takeaway per post.
-
The automation checklist I use before writing any code
The questions that separate a good automation candidate from a bad one: repetition, stable inputs, clear rules, ownership, exceptions, and checks.
-
Reading invoices and POs with AI
Document extraction is the useful, unglamorous AI job: draft the fields, check the risky ones, and stop hand-entering the boring 80%.
-
What is safe to paste into an AI tool
A plain-English three-bucket rule for employees: public, internal, and restricted, with examples of what to rewrite before pasting.
-
We tried ChatGPT and it did not stick
The problem is usually not the model. It is the rollout: no owner, no specific task, no workflow, and no verification step.
-
Keeping company data out of AI training sets
A practical owner-side guide to safe AI use: one sanctioned tool, the right data settings, and a short list of what never gets pasted.
-
This week in AI: agents got real, and so did the guardrails
A practical read on a loud week in AI: agents are moving from answers to action, safety testing is showing why permissions matter, and transparency rules are no longer theoretical.
-
What AI actually costs a small business
Two halves: seats and usage. Real per-million-token rates from the vendors' own pricing pages, worked examples in dollars, and the two choices that account for most of the difference. Updated monthly.
-
AI sales and outreach tools worth your time: Clay, Perplexity, and friends
Sales tooling is the loudest corner of the AI market. An honest map of what to buy, what to build, and what to skip — Clay for lists, Perplexity for research, and the homegrown option nobody sells you.
-
MCP connectors, explained for business owners
MCP is the standard socket that lets AI plug into your email, spreadsheets, and CRM — turning advice into errands. What it unlocks, and the security questions to ask before plugging in.
-
Mostly script, a little AI: the hybrid workflows that actually ship
The workflows that survive real business data aren't all-script or all-AI. They're deterministic pipelines with one narrow AI call at the messy joint — cheap, fast, and trustworthy.
-
Automating the browser: scripts, extensions, and how to do it ethically
AI can write the JavaScript and browser extensions that gather public data and automate web chores. What that unlocks — and the ethics rules that separate research from being a bad neighbor.
-
Automating business workflows with Python (and where AI actually fits)
When your data is consistent, a Python script written with AI's help beats calling AI every run — faster, cheaper on tokens, identical every time. How to spot the workflows that are ready.
-
Which AI model for which job: a plain-language map
Model names change every quarter; the map underneath doesn't. Words, code, making images, reading documents, analyzing data — and the right-sized tool for each.
-
Tokens: what they are, what they cost, and how to stop wasting them
Every AI bill is denominated in tokens, and most people can't define one. What you're actually paying for, why long chats cost more than you think, and the script–small-model–big-model ladder.
-
Agentic AI, explained: what it is, what it isn't
An agent is a model in a loop with tools — powerful for multi-step work that needs judgment, wasteful for work that doesn't. The plain-language version, minus the hype.
-
Turning a gigabyte of messy government data into something useful
NHTSA publishes over a gigabyte of raw vehicle-complaint text. No human can read it — but the patterns inside it are real. Here's how Python text extraction surfaced part-failure insights the dataset never states outright.
-
No token tax: when a plain script beats calling AI every time
Calling an AI model costs a little money every single time — and gives you a slightly different answer every time. For truly repetitive work, a deterministic script is cheaper, faster, and more reliable. Here's how to tell which one you need.
-
Excel first: the fastest way for a non-technical person to feel AI is real
Forget chatbots and demos. The moment AI becomes real for most people is the moment it fixes the spreadsheet they've been fighting for years. That's why every training I run starts in Excel.
-
Asking precisely: the one skill that separates useful AI from disappointing AI
The difference between "AI is useless" and "AI just saved me four hours" is almost never the AI. It's the question. Precision in asking is a learnable skill — and it's the centerpiece of everything I teach.
-
Shadow IT vs. sanctioned tools: build things without making IT your enemy
The worst way to bring AI into your workplace is behind IT's back. The best way turns them into your strongest ally. The difference is a conversation most people skip.
-
What AI can't do (yet): an honest list
Anyone selling you AI should be able to tell you where it fails. Here's my working list — the real limits I run into building with these tools every day, and what they mean for your business.
Reading is free. So is the first call.
If any of this sounds like your situation, let's talk about it — plainly, and with no obligation.