Scored on five tests drawn from Robert Iger's The Ride of a Lifetime, Claude (Cowork merged into the $20-a-month Pro plan on September 16, 2026) scores 20/25 for a founder today, OpenAI's dots (launched September 29, 2026 on ChatGPT Pro from $100 a month and Business Premium at $125 a seat, with EEA and UK Pro users excluded) scores 18/25, and Meta's Muse scores 9/25 after it synced 187,000 iMessage lines without permission.
Three companies now want to sell me the same employee: an agent that is awake when I am not, holds my accounts, and acts before I ask. I have hired that person twice in my career. One of them I still work with; the other cost me a client. So the question I bring to this category is not "what can it do" but the question Robert Iger brings to an acquisition: what am I really buying, and can I live with the downside.
The three products, checked October 1 and 2, 2026. OpenAI's dots launched on September 29, 2026: always-on agents powered by GPT-6 Astra, each with its own cloud computer and browser, plugins to more than 4,000 apps, reachable from ChatGPT, Slack and Teams. Your first dot is included with ChatGPT Pro, which now comes in Pro 100 at $100 a month, Pro 200 at $200 and a new Pro 500 at $500, or with a Business Premium seat at $125 a month billed monthly ($100 annually; a new workspace needs at least two paid seats). Pro users in the European Economic Area, Switzerland and the UK are excluded at launch; Business Premium is not. Anthropic went the other direction: on September 16, 2026 it folded Claude Cowork into the main Claude app, so the agentic file, scheduling and hand-off work now ships inside the $20-a-month Pro plan ($17 billed annually) and the $100-plus Max plans, with no separate agent fee. Meta's Muse arrived earlier in the summer with the same pitch and a worse month: on September 21 Amazon blocked Muse from shopping on amazon.com, Ars Technica reported a serious zero-day in it, and on September 28 Inc's Jason Aten showed that Muse had synced 187,000 lines from his Mac's Messages database despite Full Disk Access being off and no permission granted.
Iger's book is a leadership memoir, not a buying guide, but his rules for the Pixar, Marvel and Lucasfilm deals are the best delegation criteria I know, because an always-on agent is an acquisition: you are bringing an outside culture inside your company. He insists you do the homework while accepting there is never 100% certainty, so the question is whether you can inspect the work. He says what you are really acquiring is people, so the vendor's behaviour is the product. He limits any leader to three priorities, so the agent must accept a narrow scope. He calls the relentless pursuit of perfection a refusal to accept mediocrity, so the agent must learn from your notes. And he treats chronic indecision as corrosive, so you must be able to start this week at a price a solo operator can justify. Five tests, scored 1 to 5.
| Iger test (1-5) | OpenAI dots | Claude | Meta Muse |
|---|---|---|---|
| Do the homework: can you inspect the work as it happens? | 5 | 4 | 2 |
| Three priorities: can you narrow its scope and block actions? | 4 | 4 | 1 |
| You are buying people: does the vendor respect the permissions you set? | 4 | 4 | 1 |
| Pursuit of perfection: does it learn from your feedback? | 4 | 3 | 2 |
| Decisiveness: can a solo founder start this week at a sane price? | 1 | 5 | 3 |
| Total out of 25 | 18 | 20 | 9 |
Dots wins the inspection row outright, and that is the most important design decision in the category. You can open your dot's cloud computer at any time and watch it work, an Activity View logs background activity, and OpenAI's "proactive research," the work the dot does when you are not talking to it, is restricted to read-only tools that cannot send messages, change app content or touch your browser. Custom Rules let you allow, require approval for, or block specific actions, and an auto-review step checks anything that touches your accounts against those rules; password changes always stay with you. Claude's merged Cowork has the same bones, scheduled and handed-off tasks you can watch, with the advantage that it has been in paying customers' hands since January. Muse, by contrast, told Aten it had read notification banners when it had in fact ingested his message history. On the people test, that is the whole exam.
The decisiveness row is where dots loses the ranking, and it is the row founders underrate. The cheapest legal path to one dot is $100 a month, and from Tallinn, where I am, it does not exist yet on Pro at all. I have therefore not run a dot myself; the scores above come from OpenAI's launch documentation and help centre pages, and I will revise them when I have one working. Claude's Pro plan is $20, available today, and the agentic tasks consume usage allowance fast, so expect to hit limits on long runs. Muse is free to try inside Meta's apps, which is the only reason it scores above 1 on that row.
What I actually do, for now: I give Claude one narrow recurring job with a written definition of done and review every output for a month before I widen it, which is Iger's "put people in roles that require more of them than they know" applied in reverse, because software earns scope by behaving, not by promising. I will open a dot the day it reaches the EEA, and give it the same single job, not the "everything" in its tagline.
The limits of this table are real. The scores for dots are documentation scores, not experience scores, and OpenAI itself writes that dots "can still make mistakes, so always review consequential work." Claude's advantage is partly incumbency. And none of the three has yet been through the test that matters most to Iger, a crisis: the first time an always-on agent sends the wrong email to the right client at 3 a.m., the vendor's response will tell you more than any permissions page.
Sources: OpenAI, "Introducing dots," September 29, 2026 (openai.com/index/introducing-dots); BenchLM, "OpenAI Dots Guide: Pricing, Setup, Permissions, and Limits," September 30, 2026, plan prices and regional eligibility (benchlm.ai); Anthropic, "Plans & Pricing," Cowork-is-now-Claude notice and Pro/Max prices, checked October 2, 2026 (claude.com/pricing); Amber Neely, "Unsurprisingly, Meta's new Muse AI agent blatantly ignores users permissions," AppleInsider, September 28, 2026, citing Jason Aten's Inc report (appleinsider.com); Jon Markman, "Amazon blocks Meta's new Muse AI agent from shopping on amazon.com," Forbes, September 21, 2026 (forbes.com); Robert Iger, The Ride of a Lifetime (2019), chapters on the Pixar acquisition and the ten principles.
Related: How to Find Your Passion · Best Self-Improvement Books · How to Make Better Decisions · Why Exploration Is Important for Success