Field report E-15 / R1 · sheet /strategy
Corporate StrategyFiled from the territory
Only 2% of 2,053 Organizations Can Authorize Customer AI Agents
Customer-led AI assistants now complete purchases and shape model knowledge. A scan of 2,053 organizations finds average actionability of 2.5/100 and only 2% allowing access withdrawal.
- Grid
- E-15 / R1
- Surveyed
- November 19, 2026
- Reading
- 4 min read
- Correspondent
- Amara Osei

Field notes
- More than 10 million people in the US and Canada set up Meta Muse agents within six weeks; 25% use them daily.
- The average Muse purchase exceeds $200.
- Only 2% of 2,053 scanned organizations let customers authorize an assistant with access they can limit and withdraw.
- Average actionability was 2.5/100; only 2% published a product catalog that machines could read.
- Top-performing organizations were mostly software businesses, scoring 57.7 on average against a field average of 38.8.
By Amara Osei
More than 10 million people in the United States and Canada set up Meta Muse agents within six weeks. Only 2% of 2,053 organizations scanned by Ready Index could let customers authorize an assistant with access they can limit and withdraw.
Meta told investors on its latest earnings call that 25% use Muse daily and the average purchase exceeds $200. Tens of millions of Americans trust personal assistants, including Meta’s Muse and Instinct, to book, buy and arrange parts of their lives. They are delegating commercial tasks rather than browsing, creating a distinct audience for marketers.
For more than two decades, websites filtered bots; many new arrivals now function as customers in a new form. Websites report record visitor totals as assistants move at machine speed, prompting marketers to intensify bot filtering. That reflex risks excluding the channel that completes delegated tasks.
What happens when marketers block AI agents?
Blocking an assistant creates two losses. A blocked agent neither complains nor abandons its task; it moves to a competitor, and the customer never learns the option existed.
Over time, agent feedback shapes the next generation of models’ knowledge of the web. A business that an assistant cannot use may enter that context as somewhere not worth trying again.
Ready Index frames the minimum requirement this way: “Letting agents in is the floor rather than the goal.” Companies must decide what assistants may do, where people approve a step and when human intervention takes over. The study found no organization at top readiness, where agents operate as a supported channel across web, interface and voice.
What can 2,053 organizations actually support?
The actionability gap is measurable: scanned organizations averaged 2.5 out of 100 for building something an assistant can act upon. Only 2% published a product catalog in a format machines can read; only 7% published any position on whether assistants were welcome. On a separate measure, only 7% could connect to an assistant at all.
The strongest performers were overwhelmingly software businesses, which averaged 57.7 against a field average of 38.8. Their advantage was structural rather than creative: software products already let machines read machines.
Content neglect adds a slower risk that conventional site management can conceal. When a page became outdated, a promotion ended or a referral program expired, marketers could archive it or remove it from navigation. An assistant does not navigate menus; a hidden page remains as findable as the homepage.
Only 23% had published a file stating what was current and what an assistant could use. For the remainder, an agent could return an offer that expired two years earlier as though it remained active.
How should marketers prepare for agent-led traffic?
An assistant arrives with intent: a direct, specific statement of what the customer wants, delivered before the customer reaches the site. That level of pre-arrival intent has no precedent in the history of marketing. A working conversation also reveals the person’s underlying request more precisely than a referrer or bounce rate.
Ready Index argues that the connected 7% will pull ahead before rivals recognise the channel.
Search, social, display and email channels built over the past 30 years assumed a person at the other end. That assumption now fails as agents become a channel in their own right.
Businesses will compete through agent language (the instructions and interfaces software agents use), for the attention and ease of use of software rather than people.
None of the recommended fixes requires exotic technology. Ready Index’s five-part operating checklist is:
- Publish a product catalog in a format machines can read.
- Publish a file stating where content lives and what assistants may do.
- Give customers revocable authorization for a narrowly defined task.
- Set a written policy on whether assistants are welcome.
- Track why agents arrive, what they do and how many become customers.
The urgency rests on allocation rather than novelty: agent-facing positions are being assigned now according to which businesses can complete a task. The mobile comparison is direct: marketers debated responsiveness while competitors built sites working only on phones.
Ready Index’s conclusion is blunt: “The companies winning it are not winning with better creative. They are winning because an assistant could read them.”
via fastcompany.com/rss (Source)
More from Amara Osei
Show full bio
Market editor covering people, launches and day-to-day practice in corporate strategy and chief-executive leadership at CEO Frontier.
449 articles
Adjacent sheets
- Brands' Next Customer Is an AI Agent—and It Ignores Brand Size
- AI Agents Are Eroding the Moats Built on Consumer Laziness
- ByteDance to Launch AI Agent Rivaling Meta's Muse Within Weeks
- Amazon's Jassy Keeps Door Ajar for Outside AI Shopping Agents
- AI Now Decides Which Brands Shoppers Consider — and Nike Is Losing