Startups & Ideas
The Pricing Page Is the Product: A Field Guide to Charging for AI
Seats, credits, tasks or outcomes? Every AI pricing model makes a promise about who carries the cost of the machine. Here's how to choose yours.
by Theo Brooks
#003
Who Pays for the Machine?
The Ideabro Brief #3: who pays for the machine?
Every AI product makes a quiet promise on its pricing page: who carries the cost of the machine. Charge per seat and you carry it. Charge per credit and your customer does. Charge per outcome and you'd better agree on what an outcome is.
Theo's field guide to AI pricing is this week's lead because the question has escaped startup circles. If you buy software at work, you're now reading pricing pages that look like phone contracts from 2004. His one-sentence test will help you read them too.
We also reviewed agent builders and note-takers the way we think reviews should work now: by kind, not by brand. Brands change monthly. Kinds last.
And Lena explains the four main ways governments regulate AI, without picking a side.
— Maya
In this issue
- 1
- 2
Agents & AutomationFive Kinds of Agent Builders, Reviewed: Canvases, Chat Agents, Frameworks and More
You don't need a brand name; you need the right kind of tool. We built the same three agents with each of the five main kinds of agent builder and scored what actually mattered.
by Dev Raman
- 3
Policy & EthicsFour Ways to Regulate AI, Explained Without the Jargon
Risk tiers, sector rules, liability and licensing: a plain-language, side-free guide to the main approaches governments use, plus five questions to ask about any new proposal.
by Lena Hartmann
- 4
Tools & WorkflowsAI Note-Takers, Reviewed by Type: Bots, Built-Ins, On-Device and DIY
Product reviews go stale in a month. The kind of meeting note-taker you choose decides most of what you'll love and hate about it. We compared all four.
by Maya Okafor
Also worth your time
- In Show & Tell: a food bank's volunteer scheduling, built with three prompts and a shared sheet. Daniel's "check it in a fresh chat" trick is one of the best ideas we've seen this month.
- A teacher's class built its own benchmark. Hannah's Year 11s wrote 30 'explain it like a lab partner' questions and scored three assistants blind. Their favourite wasn't the most accurate one. The thread is in the forum.
- From the Tool Shed: Relay Canvas, where approval gates are a building block rather than an afterthought. It's where we'd send a non-technical team building its first agent.
Prompt of the week
Before you hand anything to an agent, turn it from a vague task into an errand, something software can tell it has finished.
Turn this vague task into a clear errand for an AI agent: {{task}}
Write:
- Goal (one sentence)
- Inputs it can use
- Tools and permissions it needs, and nothing more
- Definition of done, specific enough to check
- Things it must ask me about before doing
- What it should hand me at the end
One more thing
Several of you asked whether we take advertising. We do, carefully, and always labelled. The rules we hold ourselves to are on the advertise page.
Every weekday
Get the next one in your inbox
The five AI stories that matter, explained in five minutes.