The one from an accredited lab, off the internet, in thirty seconds. Not a terminal, not a framework.
- The product layer is where the difference is now — not the model
- There is an incentive for insiders to make this sound harder than it is
Cat Getsy took a landline phone company past a million dollars in under a year on work that would otherwise have taken six to ten hires. Not because the tools built it — because she onboarded them like staff. Three tools below, one per lever.
The difference between the red one and the rest is not the model.
At your fingertips, allegedly
Each tool below is one lever from the episode, with input fields attached. Start anywhere — they do not depend on each other.
The most common wrong assumption in this field is that being good at AI means knowing every tool. It does not. Pick one, click every button in it, and work down this list.
Graduating from a chatbot to something that can actually do work takes about as long as making a coffee.
The one from an accredited lab, off the internet, in thirty seconds. Not a terminal, not a framework.
One toggle. It holds more context, runs longer, and takes on tasks that take real time rather than answering and stopping.
Slack, Drive, Gmail, Calendar. The places the work already lives.
Not errands. The kind of thing you would hand an analyst or a manager who is genuinely helping you get knowledge work done.
The point where it stops being a conversation you have and starts being a system you keep.
Make an ordinary folder — finance, or whatever the domain is — and point the tool at it. Every conversation you aim there accumulates in one place.
It saves markdown into that folder without being asked — plain text notes on what the project is and what you prefer. That is its memory, and it is editable.
Describe it in plain English in the automations panel. It asks a couple of questions, confirms the tools, and gives you a toggle.
None of this is about knowing more tools. Pick one, click every button in it, and get good at the moves.
Assess the goal before rewriting. Make it interview you. Ask for the bull case and the bear case. Push once when it claims it cannot.
Dictate. Every desk in a San Francisco startup has a gooseneck mic. When you are listening rather than reading, append: give it back as a transcript that could be read aloud.
Final-say financial accounting stays supervised. So does anything you would want a second pair of eyes on.
0 of 12 cleared
Cat Getsy built Physical Phones into a million-dollar business in under a year, and runs Cat Labs, where people screen-share the actual prompts behind what they built. Every rule in these three tools is hers, from this conversation.
Open on YouTubeBetween the agents and the automations, Cherie and Jean gave away a fair amount about themselves. Answer first — the fact arrives afterwards.
Question 1 of 5
Where did Cherie, Jean and Cat stay during Cannes?