The 6 AI tools I actually pay for and use every week in my Chick-fil-A
Most AI tool comparisons are written by people who do not run an operation. Here are the six I actually pay for, what each one does in my restaurant, and what the whole stack is actually for.
Every week somebody asks me what AI tool to buy first. The honest answer is that it depends on your problem. The more useful answer is to just show you what an operator with a working stack actually pays for.
Fair warning: half of this list is one company. I am heavily Anthropic-leaning right now because they shipped the operator features first: computer access, browser control, agents that actually do work. If that changes, my stack changes. I am writing this in July 2026 and this stuff moves fast.
1. Claude
My primary thinking partner. Long-form analysis, comparing SOPs, drafting reviews, talking through operational problems out loud until they make sense. It handles long documents without losing the plot, and the output lands close to my voice without much coaxing.
The lesson I paid for: use Projects from day one. I spent months retyping my operation’s context into every new chat before I figured out I could save it once and let it carry forward.
2. Claude Cowork
The unlock. This is where AI stopped being a chat window and started being an operational partner that actually does things: edits my website, drafts documents straight into my folders, runs the ten systems I wrote about in the ten systems post. The website you are reading right now was largely built and edited in Cowork sessions, including this sentence.
Still a research preview, still occasional friction, still worth it.
3. Claude in Chrome
Drives the browser for the repetitive web work I do not enjoy: vendor portals, dashboards, review platforms. A few sites block automation and still need my hands. Most do not. The boring clicking I used to do after close mostly happens without me now.
4. ChatGPT
My second opinion. When an output is about to drive a real decision, I run it past a second model and look for disagreement. Different models have different blind spots, and the disagreements are where the danger lives. Also still my pick for generating images for training materials and team stuff.
5. Gemini
My whole operation runs on Google Workspace, so Gemini lives where my email and sheets already live. For heavy reasoning I go to Claude. For work that starts inside a Gmail thread or a spreadsheet, the tool that is already standing there wins.
6. Manus
Autonomous research runs: go compare the catering competitors near me, gather the pricing, bring me a summary. It works unsupervised for an hour and comes back with something that used to cost me half a day. Scope the task tight or it wanders. Everything I said about accountability for AI applies double to the tools that run unsupervised.
What the stack costs and what it returns
A few hundred dollars a month all-in. The financial return is real and I will spare you the inflated consultant math: the hours alone justify it many times over.
But hours are a raw material, not a result. Here is the accounting nobody does. If AI gives you back ten hours a week and you fill them with more busywork, your stack returned nothing. It just sped up the hamster wheel.
The best return my stack ever produced was not a margin point. It was a question I ran through it: who on my team deserves recognition that has not received it? The answer was a name I did not expect, a quiet team member covering for a sick coworker, and it turned into a thank-you note in her first language. Subscription math cannot price that.
What to do this week
Audit what you already pay for and cancel anything without a specific weekly use case. Pick one tool from this list and run it on real work for a week before adding anything else. And when the hours start coming back, decide on purpose where they go. The tools automate the ordinary. What you amplify with the extraordinary is the only line on the invoice that matters.
The questions I point my stack at are free at jonharmeling.com/prompts.