I Let My AI Agent Plan My Last 3 Business Trips — It Was Better Than My Travel Agent

I used to pay a travel agent $150 per trip to book my business travel. She was good. She knew I hated layovers under 90 minutes, preferred aisle seats, and would rather stay at a mid-range hotel close to the venue than a fancy one across town. For three years, that was money well spent.

Then I started running more of my business through an AI agent — first for email and calendar, then invoicing, then basically everything my virtual assistant used to do. At some point I thought: can this thing plan a trip?

The answer, after three real business trips, is yes. With caveats. But the caveats are smaller than I expected, and the savings — in both money and time — are bigger than I expected.

Here’s the full story.

The Setup: What I Actually Asked My Agent to Do

Before I get into the trips, let me explain the workflow. I didn’t just type “book me a flight to Austin” and hope for the best. That would be a disaster. I built a travel planning workflow with specific instructions, and it took me about two hours to get right.

The workflow covers five phases:

Search and compare — flights, hotels, rental cars. Pull options from multiple sources, rank by my preferences (direct flights first, then shortest total travel time, then price).
Build itinerary — combine all bookings into a single document with confirmation numbers, addresses, check-in times, and a day-by-day schedule.
Monitor for disruptions — watch for flight changes, gate changes, cancellations. Alert me immediately if anything shifts.
Collect receipts — grab confirmation emails, pull receipt PDFs, organize by trip for expense reporting.
Post-trip summary — total cost breakdown, time savings estimate, anything that went wrong.

If you’ve read my post on setting up agent workflows, you know I’m obsessive about defining the “done” state before I start. Travel planning is no different. The agent needs to know what “good” looks like or it’ll optimize for the wrong thing.

I run this through Agent-S, which handles the orchestration across all the different tools and data sources. The key is that it’s not just a chatbot — it actually executes multi-step workflows, monitors for changes, and takes action without me babysitting it.

Trip 1: Domestic Conference in Austin (March 2026)

The first test was low-stakes on purpose. A three-day conference in Austin, Texas. Direct flights from my home airport. Straightforward hotel situation. If the agent screwed this up, the consequences were mild.

What I gave it: Conference dates (March 11-13), my home airport, budget range ($1,200-$1,800 total), preference for hotels within walking distance of the convention center, and my usual flight preferences.

What it did:

The agent pulled flight options from multiple aggregators and came back with a comparison table in about four minutes. Not a list of links — an actual structured comparison with departure times, total travel time, price, seat availability, and layover details. It flagged that a direct flight on the morning of March 11 was $87 cheaper than the afternoon option and would get me there in time for the evening networking event.

For hotels, it found 14 options within a mile of the convention center, filtered down to 6 that matched my price range, and ranked them by a combination of distance, review score, and whether they had late checkout (which I always want on the last day of a conference). It noted that one hotel was running a conference attendee discount that brought it $45/night below the next cheapest option.

Total time I spent: about 12 minutes reviewing the options and saying “book option A for flights, option 3 for hotel.”

The result: $1,340 total trip cost. My travel agent’s typical Austin trip ran about $1,500-$1,600. The $45/night hotel discount alone saved $135 over three nights — something my travel agent had missed on a previous Austin trip because she wasn’t cross-referencing the conference website for partner hotel deals.

What impressed me: The itinerary document. It wasn’t just a list of bookings. It included the conference schedule (pulled from the public website), suggested restaurants near the hotel with ratings, Uber cost estimates from the airport, and a note that the hotel had a free airport shuttle I could book 24 hours in advance. My travel agent never did any of that.

What didn’t work: Nothing went wrong on this trip, which meant I didn’t get to test the disruption handling. Almost too smooth.

Trip 2: International Client Meeting in London (April 2026)

This was the real test. International travel has more moving parts — passport considerations, time zones, currency, ground transportation in a foreign city, and the general complexity of a transatlantic trip.

What I gave it: Meeting dates (April 22-23 at the client’s office in Canary Wharf), preference for arriving a day early to adjust, return flight on April 24, hotel near Canary Wharf, and a budget of $3,500-$4,500 for everything.

What it did:

The flight search was more interesting this time. The agent found that flying into London City Airport instead of Heathrow would save me 45 minutes of ground transportation to Canary Wharf, even though the flights were slightly more expensive. It calculated the total cost including ground transport both ways and showed that London City was actually $60 cheaper door-to-door despite a $120 higher airfare. That’s the kind of analysis I wouldn’t have thought to ask for, and my travel agent definitely never ran those numbers.

It also found a business class upgrade available for $380 on the outbound overnight flight, and flagged it with the note: “Overnight transatlantic, arriving for a client meeting — upgrade may be worth it for sleep quality.” That’s… actually good judgment. I took the upgrade.

For the hotel, it prioritized proximity to the client’s specific office address (not just “Canary Wharf” generally), found that one hotel was a 4-minute walk versus the others at 12-15 minutes, and that the closer one was only $30/night more. Easy call.

The savings moment: Here’s where it got interesting. The agent found that booking the hotel through the airline’s travel portal earned double loyalty points AND was $22/night cheaper than booking direct. Over three nights, that’s $66 in direct savings plus roughly 4,500 bonus points. My travel agent had never once cross-referenced airline loyalty portals for hotel bookings. Across the full trip, the agent saved roughly $400 compared to what I would have paid through my travel agent’s typical booking approach.

The itinerary: This is where the agent really earned its keep. The London itinerary included:

Tube directions from London City Airport to the hotel (with a note that the DLR was faster and cheaper than a taxi at that time of day)
The client’s office address with a walking route from the hotel
Currency conversion rates and a note about contactless payment being universal in London
Weather forecast for the meeting dates pulled two days before departure
A restaurant recommendation near the client’s office for the post-meeting dinner, filtered by “good for business dinners” and “has a private dining option”
Time zone adjustment reminders in my calendar

What went wrong: The agent booked a hotel that required a credit card authorization hold in British pounds, and my card flagged it as fraud. The agent couldn’t fix that — I had to call my bank. Not the agent’s fault, but it’s worth noting that there are parts of the travel chain that still require a human with a phone.

Trip 3: Multi-City Tour — Chicago, Denver, Phoenix (May 2026)

Three cities in five days. This is the trip type that used to take my travel agent two or three rounds of back-and-forth to get right. It’s also the trip where the disruption handling finally got tested.

What I gave it: Three meeting dates across three cities (May 5 in Chicago, May 7 in Denver, May 8 in Phoenix), with the constraint that I wanted to minimize total travel time and ideally not take any flights before 7 AM.

What it did:

The multi-city optimization was genuinely impressive. The agent mapped out four different routing options with different tradeoffs:

Option A: Home → Chicago → Denver → Phoenix → Home (logical geographic order)
Option B: Home → Chicago → Phoenix → Denver → Home (cheaper flights but more total air time)
Option C: Home → Denver → Chicago → Phoenix → Home (best individual flight times but tight connection in Chicago)
Option D: Home → Chicago → Denver → Phoenix → Home with a red-eye return (saves a hotel night)

Each option had total cost, total travel time, number of connections, and earliest required departure. Option A won on every metric except price (Option D was $180 cheaper but meant a red-eye, and I’m too old for that).

The hotel situation was also well-handled. Three different cities, three different hotels, all booked with the same loyalty chain where possible to maximize points. When it couldn’t find the same chain in Phoenix at a reasonable price, it explained why and offered the best alternative.

The disruption test: On May 6, my Chicago-to-Denver flight got cancelled. Weather delay cascading into crew timing issues — classic airline chaos. Here’s what happened:

The agent detected the cancellation from the airline’s notification email within about 8 minutes. It immediately:

Found three alternative flights to Denver on the same day (two on other airlines, one later flight on the same airline)
Checked whether my Denver hotel had a flexible check-in time in case I arrived later
Flagged that the earliest alternative would get me there by 4 PM, which was tight for my 6 PM dinner meeting
Recommended the 11:30 AM alternative on a different airline as the best balance of arrival time and cost

I approved the rebooking in about 30 seconds from my phone. The agent handled the rest — new confirmation, updated itinerary, adjusted ground transportation estimate, and a note to file for a refund on the cancelled flight.

For comparison: the one time I had a cancellation while using my travel agent, I called her, went to voicemail, waited 40 minutes for a callback, and by then the best alternative flights were gone. She ended up rebooking me on a flight that arrived three hours later than what the agent found. I’m not saying she was bad — she was juggling other clients. But an agent that’s monitoring my trip doesn’t have other clients.

The receipt collection: After the multi-city trip, the agent compiled every receipt — flights, hotels, meals charged to hotel rooms, Uber rides — into a single expense report document. It categorized everything, calculated totals by city, and flagged one duplicate charge from the Denver hotel. That duplicate charge was $47, and I probably never would have caught it manually.

This feeds directly into my invoicing and bookkeeping workflow. No manual data entry. The receipts flow straight into my expense tracking system.

The Numbers: Agent vs. Human Travel Agent

After three trips, here’s the honest comparison:

Metric	Human Travel Agent	AI Agent
Cost per trip	$150 flat fee	~$0 (part of my existing agent subscription)
Average booking time (my time)	20-30 min of back-and-forth	10-15 min reviewing options
Flight/hotel savings found	Baseline	$180 average per trip vs. agent
Itinerary quality	Good (basic bookings + confirmations)	Excellent (full day-by-day with extras)
Disruption response time	30-60 minutes (depends on availability)	Under 10 minutes with alternatives ready
Receipt collection	Not included	Automatic
Loyalty point optimization	Sometimes	Systematic

Over three trips, the direct savings break down like this:

Travel agent fees saved: $450 (3 × $150)
Booking savings found by agent: ~$540 ($0 on Austin since it was comparable, ~$400 on London, ~$140 on the multi-city routing)
Caught duplicate charge: $47
Total quantifiable savings: ~$1,037

That’s not life-changing money. But it’s real money, and it’s on top of time savings. I tracked the time using the same method I described in my 30-day ROI tracking post — about 3.5 hours saved across the three trips when you factor in the elimination of back-and-forth, receipt chasing, and itinerary building.

When I look at the real cost of running my AI agent setup, travel planning adds almost nothing to the marginal cost since it’s part of the same platform handling everything else.

What the Agent Still Can’t Do

I want to be honest about the gaps because I’ve written about how I handle agent mistakes and I think transparency matters.

Reading the room on hotel neighborhoods. When I’m visiting a client, there’s a soft calculus about where to stay. Is this a conservative Fortune 500 client who’ll notice if I’m at a budget hotel? Is this a startup founder who’d think I was wasting money if I stayed at the Ritz? My travel agent knew my clients. She’d say, “For the Goldman meeting, stay at the Langham. For the startup in Shoreditch, stay at the CitizenM.” The agent optimizes on distance, price, and reviews. It doesn’t understand social signaling.

I’ve partially solved this by adding client-tier tags to my CRM that map to hotel budget ranges. “Enterprise client = $250-400/night, startup client = $120-200/night.” It’s a crude proxy, but it works for 80% of situations.

Complex visa and documentation requirements. For my London trip this wasn’t an issue (US passport, no visa needed). But for more complex international travel — countries with visa requirements, vaccination documentation, or specific entry rules — I wouldn’t fully trust the agent yet. It can look up requirements, but the stakes of getting it wrong are too high for me to skip double-checking.

Negotiating corporate rates. My travel agent had relationships with certain hotel chains and could sometimes get corporate rates that aren’t publicly available. The agent works with publicly available pricing, loyalty programs, and aggregator deals. It’s good at finding the best public price, but it can’t call a hotel sales manager and negotiate.

Last-minute human judgment calls. On the multi-city trip, I spontaneously decided to extend my Phoenix stay by a day because the client meeting went well and they invited me to a team offsite. Changing flights, extending the hotel, and adjusting the itinerary — the agent could do all of that. But knowing that I should stay because the client relationship was at an inflection point? That was a human read of the room that no agent would have suggested.

How This Fits Into My Bigger Agent Setup

Travel planning isn’t a standalone thing for me.It’s one workflow inside a larger system where my agent handles most of the operational work in my business. The travel workflow connects to:

Calendar management — the agent blocks travel time, adds flight times to my calendar, and adjusts meeting buffers around travel days
Expense tracking — receipts automatically flow into my bookkeeping system
Client communication — it can draft “looking forward to meeting Tuesday” emails timed to send after I’ve confirmed bookings
Follow-up — post-trip, it queues follow-up tasks based on who I met

This is what I mean when I talk about agent workflows that actually move the needle. It’s not about any single capability being mind-blowing. It’s about the connections between workflows creating compound value. The travel planning alone saves me maybe $350/trip and a couple hours. But when it feeds into expense tracking, calendar management, and client follow-up automatically? The total value is much higher than the sum of the parts.

If you’re looking at building something similar, Agent-S is what I use to tie all of this together. The multi-step workflow orchestration is what makes travel planning work — it’s not just “search for flights” but “search, compare, book, monitor, collect receipts, and handle problems” as one continuous workflow.

Would I Go Back to a Human Travel Agent?

No. And I say that with genuine respect for my former travel agent, who was great at her job. The math just doesn’t work anymore.

The agent is faster at searching and comparing. It’s better at optimizing across multiple variables simultaneously. It never goes to voicemail when my flight gets cancelled at 11 PM. And it costs me nothing incremental per trip.

The one scenario where I’d consider a human agent: if I were planning a complex personal trip — a two-week vacation with multiple countries, family members with different preferences, and the kind of “make it special” requests that require taste and creativity. Business travel is optimizable. Vacation travel is an art. I’m not ready to hand the art over to an algorithm.

But for business travel? Three trips in, and I’m not looking back.

If you’re still doing business travel planning manually — or paying someone $150+ per trip to do what an AI agent can handle — you’re leaving money and time on the table. It was one of the easier workflows I’ve set up, and the ROI was obvious within the first trip.

Frequently Asked Questions

Can an AI agent actually book flights and hotels, or does it just search?

It depends on your setup. My agent handles the search, comparison, and recommendation phase autonomously, and I approve the final booking with one click. For most business travelers, you want that approval step — you don’t want to wake up to a $4,000 flight booked without your okay. The key is that the agent does 95% of the work (searching, filtering, comparing, building the itinerary), and your only job is reviewing the final recommendation and saying yes. The entire approval process takes me about 10-15 minutes per trip compared to the hour-plus I used to spend going back and forth with my travel agent.

How does the AI agent handle flight cancellations and rebookings?

The agent monitors airline notification emails and booking status in real time. When it detects a cancellation or significant schedule change, it immediately searches for alternatives, evaluates them against my preferences and constraints (arrival time for meetings, budget, seat preferences), and presents me with ranked options. On my multi-city trip, the entire process from cancellation detection to having a new confirmed booking took under 15 minutes. The critical advantage over a human agent is availability — my agent doesn’t have other clients, doesn’t take lunch breaks, and responds at 11 PM on a Tuesday just as fast as at 2 PM on a Wednesday.

Is AI travel planning worth it for infrequent travelers?

If you take fewer than 4-5 business trips per year, the setup time for a dedicated travel workflow might not pay off on its own. But if you’re already running an AI agent for other business tasks — email, calendar, invoicing — adding travel planning is incremental. The workflow took me about two hours to configure, and it paid for itself on the first trip. For someone starting from zero, I’d recommend getting the core agent workflows running first and adding travel once you’re comfortable with how the agent handles multi-step tasks.

What about travel policies and corporate compliance?

If you work within a corporate travel policy (specific airlines, hotel chains, booking platforms, spend limits), you can encode those as constraints in the agent’s workflow. Mine is simpler since I run my own business, but the same principle applies — define the rules upfront, and the agent will optimize within them. It’s actually more consistent than a human agent for policy compliance because it never “forgets” a rule or makes an exception without asking.

How do you handle loyalty programs and points across multiple airlines and hotels?

I gave my agent my loyalty program memberships and tier statuses. It factors these into every comparison — not just the sticker price, but the points earned, tier qualification nights, and whether a booking through a specific portal earns bonus points. On the London trip, this cross-referencing found $66 in direct savings plus bonus points I never would have gotten booking the obvious way. Over a year of business travel, loyalty optimization alone probably covers the cost of running the agent.