AI agents take over the search and comparison part of an intermediary's work first. Judgment on complex cases, accountability, and execution after the sale stay with people for much longer. So the intermediaries most exposed are the ones whose main value is an information gap, a comparison, or customer inertia. The ones that add judgment and accountability keep their place, and the best of them automate their own routine work.
What an Intermediary Sells
Every intermediary sells some mix of four things:
| What it sells | Examples | Can an agent do it? |
|---|---|---|
| Search | Finding carriers, products, listings, candidates | Yes, now |
| Comparison | Quotes side by side, matching a need to an option | Yes, now, if suppliers publish data machines can read |
| Accountability | A license, a signature, liability, a fiduciary duty | No. A person or firm must carry it |
| Execution | Claims, problems, renewals, logistics after the sale | Partly. Routine steps yes, exceptions no |
The useful question for any business is how much of its revenue comes from the first two rows. The essay The Commanding Heights by Will Manidis and Joe Roberts asks investors the same thing about every toll position they own: "Does intelligence open a new route around your holdings' tolls?" The answer depends on what the next-best route costs the buyer. When an agent can search and compare in seconds, that cost falls close to zero for simple purchases.
What Happened to Travel Agencies
The travel industry went through this change when booking moved online. The Bureau of Labor Statistics counts employment at US travel agencies.
Employment peaked at 170,600 in 1999. By 2011, it was 82,100, a fall of 52%, as simple flight and hotel bookings moved to websites. Then it rose again, to 92,300 in 2018. That recovery is consistent with the remaining agencies focusing on complex trips, groups, and high-value travel, where a person adds judgment and fixes problems. The pandemic cut the series again in 2020. The BLS now expects little or no change in travel agent jobs from 2025 to 2035.
The lesson is not that every intermediary disappears. The simple, comparison-only part of the work leaves first. The complex part stays, and it can grow.
Agents Already Search and Compare
The same shift is now happening with AI, at much larger scale.
- Google AI Mode passed 1 billion monthly users in Q2 2026, according to Alphabet's earnings release.
- Amazon's Rufus assistant reached more than 300 million customers in 2025. Amazon says it delivered nearly $12 billion in incremental annualized sales (Q4 2025 results).
- Payment networks built rails for agents to buy. Visa Intelligent Commerce (April 2025) and Google's Agent Payments Protocol (September 2025) let agents complete purchases with the user's authorization.
Buying through an agent is still early. Walmart told Wired that purchases completed inside ChatGPT converted at about one-third the rate of its own website, as Search Engine Land reported. Search and comparison moved first. The final purchase moves more slowly.
Suppliers Meet the Agent Directly
When the comparison moves into an AI assistant, suppliers can go there directly. In May 2026, Liberty Mutual launched an app inside ChatGPT that quotes auto insurance from its own rating engine. The press release contrasts its quotes with aggregator apps that show only a range of prices. The customer buys the policy on Liberty Mutual's own website.
Carriers have sold direct for decades. What changes is the cost of comparing them. An agent can collect several carrier quotes in a minute. An intermediary whose only service was collecting those quotes now competes with that minute.
The Strongest Intermediaries Automate Themselves
Freight brokerage is a matching business, exactly the kind of work an agent could take over. The largest broker is automating its own work instead.
C.H. Robinson reports that it runs hundreds of its own AI agents. In October 2026, it said they had sent more than 80,000 ocean and air quotes to more than 3,600 customers. Quotes now take 2 to 3 minutes instead of 1 to 3 days. Customs processing time fell 23%. Before these agents, after about 10 years of earlier automation, only about 5% of ocean and air booking requests were automated. Uber Freight also runs more than 30 AI agents across the shipment process.
A broker that answers in 2 minutes is a broker that a customer's agent can work with. Automation turns the intermediary into a fast, reliable endpoint instead of a slow step to route around.
Test Your Business: The Agent Exposure Score
Score each question from 0 to 2. A higher score means more exposure.
| # | Question | 0 | 1 | 2 |
|---|---|---|---|---|
| 1 | Is your main value finding or comparing options? | No | Partly | Yes |
| 2 | Can a buyer get the same prices or products directly from the supplier? | No | Some | Yes |
| 3 | Do suppliers publish their prices, products, or quotes in a form machines can read? | No | Some | Yes |
| 4 | Do customers stay mainly because switching is effort, not because of your service? | No | Partly | Yes |
| 5 | Is a typical deal simple and repeatable? | No | Some | Most |
| 6 | Does the customer rely on your judgment for complex or unusual cases? | Often | Sometimes | Rarely |
| 7 | Do you carry accountability: a license, a signature, liability, or claims handling? | Yes | Partly | No |
| 8 | Do you execute after the sale (service, claims, problems, renewals)? | Yes | Partly | No |
| 9 | Can a customer's AI agent get a quote or answer from you in minutes? | Yes | Hours | Days |
| 10 | Do you hold data about the customer that a new provider cannot get? | Yes | Some | No |
0 to 6: resilient. Your value is judgment and execution. Use AI to serve more clients with the same team. 7 to 13: at risk. Parts of your work are search and comparison. Automate those parts, and move your people to the complex cases. 14 to 20: exposed. An agent can do most of what you sell. Change the offer now.
Questions 1 to 5 measure what an agent can take. Questions 6 to 10 measure what an agent cannot easily take. Question 9 is the fastest one to change.
What to Do This Quarter
- Make yourself readable by agents. Publish your products, coverage, prices, or availability in structured form, and offer a quote API or an app in the AI assistants your customers use. See how AI search finds your site.
- Automate your own back office first. Quote, intake, document, and follow-up work should take minutes. Your speed becomes the reason an agent routes to you.
- Move your people to judgment. Put staff time on complex cases, exceptions, claims, and advice. Those are the parts customers still pay a person for.
- Use your relationship data. History, preferences, and past claims help you serve customers better than a stranger's agent can. Keep that data clean and use it with the right permissions.
- Give your clients your own agent. A client who asks your agent does not need to ask a general one.
Put Agents to Work for Your Business With OpenNash
OpenNash builds AI agents that run inside your business, measured in hours saved and cost per case. Our forward-deployed engineers work with your team to:
- Automate quoting, intake, and service so you answer in minutes, with human approval on exceptions.
- Make your products readable by agents through APIs and apps in the AI assistants your customers use.
- Connect your systems with the right permissions, so agents use your relationship data safely and every action is logged.
- Run the agents in production for a flat monthly fee, one workflow at a time.
Book a call with an OpenNash forward-deployed engineer or email us at hello@opennash.com. In the first session, we score your business on the exposure test and review your top AI use cases and data. Then we show you which ones deliver the highest business value and the fastest ROI.