The Agent Economy
How Consumers Actually Use AI Agents
Executive summary
In 2023, Gartner predicted that by 2026, 20% of inbound customer service contacts would come from “machine customers”. This fall, personal agents started calling, emailing and messaging companies for the people who use them. Companies now serve two kinds of customer: people, and the agents people send. We asked nearly 2,000 US consumers whether their AI is already doing this, and whether they want it to.
66%
Interest is running well ahead of use
Two in three consumers are interested in an AI agent that deals with companies for them, yet at most 1 in 16 say an AI tool has contacted a company itself on their behalf. About half of everyone we asked is interested but hasn't started.
~3x
Early users want agents to take on more
The 14% who say an AI tool has contacted a company for them are about three times as likely to want an agent to handle the whole job (24% vs 8%), and 85% of them are interested in using one this way.
56%
Consumers are most ready to delegate negotiation
55.5% are comfortable sending an agent to get a better price, against 38.5% for making a booking. The first agents are likely to turn up in conversations about money.
76%
Once people try it, most keep doing it
Three in four early users have done it more than once, and more than 1 in 4 do it weekly or daily. The jobs include returns, refunds, cancellations, claims and negotiating a lower rate.
1 in 9
The line between the customer's agent and the company's is blurring
About 1 in 9 people who said their AI dealt with a company were describing the company's own chatbot. The business still runs much of what customers experience as “using an agent”.
Methodology
Finding 1: Interest is running well ahead of use
66% are interested, but at most 6% say an AI tool has contacted a company for them
We asked 1,966 consumers whether, in the last 12 months, an AI tool had messaged, emailed or called a company on their behalf (advice and drafts excluded), and how interested they would be in an agent that did.
are interested in an AI agent that deals with companies for them
at most, say an AI tool has contacted a company itself for them
Out of every 100 consumers
Prompt“If an AI agent could deal with companies for you, how interested would you be in using one?” 66.4% = 18.7% very + 47.7% somewhat interested. n = 1,966.
13.9% said an AI tool had contacted a company for them. When we asked what the tool did, 6.3% of all respondents picked “contacted the company itself”. We treat that as the upper bound: roughly 1 in 16. Applied to the roughly 262 million US adults, that would be up to about 16 million people. This is our estimate, not a count, and the true figure is likely lower.
Interest is much higher. Most of it sits with people who haven't started: among those who say an AI tool hasn't contacted a company for them, 63% are interested and 16% are very interested. That group is about half of everyone we surveyed. Most interest is still tentative, though: 47.7% are only somewhat interested, and only 10% would rather an agent handle it for them today (53.8% prefer to do it themselves, 32.9% want AI advice).
Finding 2: Early users want agents to take on more
Early users are about three times as likely to want an agent to handle the whole job
We compared the 274 early users with the 1,563 people who say an AI tool hasn't contacted a company for them.
The gap between early users and everyone else
Interested
Very interested
Prefer an agent handles it
Prefer to handle it myself
85% of early users are interested in using an agent this way, against 63% of people who say an AI tool hasn't contacted a company for them. They are more than twice as likely to be very interested (38% vs 16%) and about three times as likely to want the agent to handle it outright (24% vs 8%). 29% would still rather do it themselves, half the rate of those who haven't (58%).
as likely to want an agent to handle the whole job, among early users
Early users chose to try agents, so the cause could run either way: keen people try agents, or trying one makes people keen.
Finding 3: Consumers are most ready to delegate negotiation
55.5% are comfortable sending an agent to get a better price, against 38.5% for a booking
We asked how comfortable people would be letting an AI agent contact a company on their behalf for three jobs.
Comfort letting an AI agent contact a company for you
Getting a better price
Making a complaint
Making a booking
Prompt“How comfortable would you be letting an AI agent contact a company on your behalf for…” n = 1,966.
Getting a better price came out well ahead: 55.5% comfortable and 28.8% uncomfortable. Complaints split almost evenly (41.8% vs 40.8%). Making a booking was the only task where discomfort won (38.5% comfortable, 45.3% uncomfortable).
For businesses, that points to pricing and other conversations about money as the first place agents are likely to appear. These are the conversations where a business keeps or loses revenue: a rate review, a refund, a cancellation, a renewal.
Finding 4: Once people try it, most keep doing it
76% of early users have done it more than once
We asked early users how often they use AI to deal with companies.
How often early users use AI to deal with companies (n = 274)
More than 1 in 4 do it weekly or dailyOnly 23.7% of early users described it as a one-off. What the AI did varied: 62.0% said it answered a question or looked something up, 50.0% said it drafted a message they sent, and 44.9% said it contacted the company itself. The jobs people described included returns, refunds, cancellations, claims and negotiating a lower rate.
Finding 5: The line between the customer's agent and the company's is blurring
About 1 in 9 were describing the company's own chatbot
About 1 in 9 people who said their AI dealt with a company (30 of 271) described a company's own chatbot or in-app assistant, at a retailer, a bank or a delivery app. To them, the company's AI was their AI. The business still runs much of what customers experience as “using an agent”.
were describing the company's own chatbot
“I had a dispute with amazon and alexa contacted them for me and actually resolved it as well.”
Survey respondent. From hand-coded open-text answers, so treat this finding as directional.
What this looks like in the wild
Public accounts since September are consistent with the survey.
Insurance
Switching insurance in minutes
A Muse user posted that the agent found equivalent car insurance about $3,500 a year cheaper, bought the new policy and cancelled the old one.
X, September 10
Travel
A delay claim, settled by an agent
Another user posted that their agent secured a $250 airline credit in five minutes after a seven-hour delay.
X, September 21
Hospitality
Blocked, then reinstated
A restaurant booking platform banned a consumer whose agent was checking for tables every 0.4 seconds, then reinstated the account two days later.
Lorikeet blog
Commerce
Businesses are picking sides
Amazon blocked Meta's Muse from September 20. PayPal opened checkout to it the next day. eBay has banned “buy-for-me” agents since February.
Adweek; PayPal on X; Power Systems Design
“Personal agents like Instinct, Muse, Grok bot and Town are the fastest growing consumer product since ChatGPT. On behalf of your customers they now research, recommend and buy, and they engage businesses persistently, repeatedly and ruthlessly along the way.”
Steve Hind · Co-founder and CEO, LorikeetKey implications
What agent-ready means
An agent-ready business can serve a customer who arrives as an agent. The agent can find accurate answers about prices, policies and orders from the business itself. It can reach the business by chat, email or phone without being blocked. It can complete the job it was sent to do, under the same identity, consent and compliance checks the customer would face. And the business can tell which contacts come from agents, and count them.
01
Plan for customers who arrive as agents.
Two in three consumers are interested, about half of everyone we surveyed is interested but hasn't started, and most early users keep coming back. Expect a rising share of contacts to come from agents.
02
Start with conversations about money.
Getting a better price is the job consumers are most ready to delegate, and it is where revenue is at stake. Make sure an agent asking about a rate, a refund or a cancellation gets a correct, fast answer.
03
Give agents a front door.
Blocking an agent strands the human customer behind it, as one restaurant platform found when it banned an agent user and reinstated the account two days later. Make it easy for an agent to complete the job the customer sent it to do.
04
Give an agent no more authority than the customer has.
Identity checks, consent and compliance steps stay fixed. An agent should be able to do what the customer could do, and nothing more.
05
Count agent contacts.
Agent-written messages are often clean and polite enough to pass as human, so they are easy to miss. Tag them so you know how many arrive and what they ask for.
Numbers that matter
66%
are interested in an AI agent that deals with companies for them
1 in 16
at most, say an AI tool has contacted a company itself for them (6.3%)
About half
are interested but say an AI tool hasn't contacted a company for them yet
3 in 4
early users have done it more than once
~3x
as likely to want an agent to handle the whole job, among early users (24% vs 8%)
56% vs 39%
are comfortable sending an agent to get a better price, versus letting one make a booking
1 in 9
people who said their AI dealt with a company were describing the company's own chatbot
Give your customers' agents a front door.
Lorikeet B2A gives agents a concierge that answers clearly, resolves the job inside your rules, and gives an agent no more authority than the customer has.