# Agentic AI for travel agents: how to become the builder

> How rule-based chatbots, RAG and agentic AI differ, why travel agents should build their own AI agent, and three workflows it can run with Volty.

Leonard Loo

CEO & Co-Founder, Volty

Updated Nov 10, 2026 · Published Nov 6, 2026

11 min read

AI is changing how people travel, and the travel agencies that stay ahead will be the ones whose consultants learn to build with it. Building with AI has become easier for non-technical workers. Most of the people automating work with AI are not developers: IMDA's Singapore Digital Economy Report 2026 found that 63.9% of Singapore workers who use AI use it to automate processes, against 15.4% who use it in software development. The same report found that 85.7% of Singapore workers used AI tools at work in 2026, and more than two-thirds said they need to upskill in AI. This is why travel consultants should build their agency's AI agent themselves: they know the work best, and the tools no longer need code.

## What has changed for travel agents

Many travelers now plan their trips with AI. In the International Air Transport Association's 2026 Global Passenger Survey, 57% of travelers in Asia-Pacific said they had used AI for travel planning, against 49% worldwide. An AI assistant can suggest an itinerary, but it does not know your destinations, your suppliers or your regular clients. Your consultants do. An agency that teaches that knowledge to its own AI agent gives travelers both the speed they now expect and the expertise they come to an agency for.

At the same time, agencies have more travelers to serve and fewer people to serve them. Singapore welcomed 16.9 million visitors in 2025, and the Singapore Tourism Board projects 17 to 18 million in 2026, according to its Chief Executive, Melissa Ow. Yet STB's Travel Agent Industry Toolkit 2026 notes that Singapore's population aged 65 and over will rise to nearly one in four by 2030, and that fewer people want tourism roles. An AI agent closes that gap. It answers routine questions at any hour, so a small team can serve more travelers without hiring for evenings and weekends, and consultants keep their time for the work only they can do.

Industry leaders already point to retraining rather than hiring. Wong Yew Hoong, co-founder of EU Holidays, said in the same toolkit:

> Given the current manpower crunch, retraining existing staff and redesigning their roles for more productive work has become more important than ever before.

## Three types of AI chatbots

Most agencies have tried a chatbot of some kind, and many chatbot tools now describe themselves as AI. Adding AI to a chatbot does not make it agentic. A chatbot that writes smoother replies still stops at the reply. An agentic AI agent takes the next steps itself: it asks for missing details, updates your records and follows up when a traveler goes quiet. The three types below show where each one stops.

### Rule-based chatbots

A rule-based chatbot follows a script. Someone on your team draws every menu, button and reply by hand, and the chatbot offers those choices to the traveler. It handles a short list of fixed questions, such as office hours. It fails as soon as a traveler asks something outside the script, such as whether a package suits a traveler who uses a wheelchair. Every time a package, price or policy changes, someone has to rewrite the script by hand.

- 1 · The chatbot can only offer the menu someone wrote by hand.
- 2 · The family's question is not on the menu, so the chatbot shows the same menu again.

Example: a rule-based chatbot cannot read the family's message, so it shows its menu again.

### Chatbots that answer from your documents (RAG)

A chatbot built on retrieval-augmented generation, or RAG, answers in its own words from documents you give it. You upload your brochures, cancellation terms and visa notes, and when a traveler asks a question, it finds the relevant passage and writes a reply. It can answer questions nobody scripted, as long as the answer is in your documents. It cannot act on the answer: it does not ask for missing details, update a record or follow up later, so a consultant still does the rest of the work.

- 1 · The chatbot finds the answer in your brochure and writes it in its own words.
- 2 · Asked to do something, it can only send the family to your team.

Example: a chatbot that answers from your documents describes the package, then sends the family to your team for the quote.

### Agentic AI agents

An agentic AI agent builds on the same idea. It answers from your documents too, but it works through a task from start to finish, the way a consultant handles an inquiry from the first message to the booking. It works out what the traveler is actually asking for, asks questions when it is unsure and takes actions within the limits you set, such as saving the traveler's details, following up when they go quiet or escalating the conversation to someone on your team. It keeps working on the inquiry after the first reply, so a traveler who drops off is followed up without anyone on your team remembering to do it.

- 1 · The agent answers, then asks for the details a quote needs.
- 2 · It saves what the family has told it to their record, without anyone typing it up.
- 3 · When the family goes quiet, it follows up on its own three days later.
- 4 · Once it has everything, it marks the inquiry urgent and assigns it to Mei Ling with the full brief.

Example: an agentic AI agent asks for the missing details, follows up when the family goes quiet and passes a complete brief to a consultant.

|  | Rule-based chatbot | Chatbot that answers from your documents (RAG) | Agentic AI agent |
| --- | --- | --- | --- |
| How it answers | Fixed menus and scripted replies | In its own words, from your documents | In its own words, from your documents and what the traveler has said |
| What it can do | Show the next menu | Answer the question | Answer, ask for missing details, update records, follow up and hand over to a person |
| What you set up | Every menu and reply, drawn by hand | Your documents | Your documents, plus instructions and limits in plain words |
| Limitations | Cannot answer anything outside the script | Cannot take any action | Acts only within the limits you set, and passes anything you keep for a person, such as pricing and refunds, to your team |

## Why the travel agent should be the builder

An AI agent starts out knowing nothing about your business. It does not know which ground handler in Hokkaido answers on a Sunday, which hotels look after your regulars, or that a family with two toddlers should never be booked on a 6am transfer. Consultants can close that gap by building the AI agent themselves and teaching it what they know.

As the IMDA figures show, building with AI no longer takes a developer. Good consultants already have the skills it takes:

- Explaining a process clearly. If you can brief a new hire on how you quote a family trip to Japan, you can brief an AI agent.
- Setting limits. You decide what the agent may do on its own and what always goes to a person, such as pricing, refunds and complaints.
- Reviewing and correcting. Read the agent's draft replies, fix the ones that are wrong, and notice the pattern behind each fix.
- Describing a repeated task in one sentence. "Remind every traveler three days before departure" becomes an automation once someone writes it down.

## Where an AI agent fits in an agency

An AI agent is best at automating routine work such as:

- Answering questions about packages, visas, baggage and cancellation terms from your own documents.
- Collecting the details a quote needs: travel dates, destination, number of travelers, budget and special requests.
- Following up on quotes that went quiet, and reminding travelers before they depart.
- Flagging inquiries that are waiting for a reply, and marking each one urgent, high or low based on what the traveler wrote.

This gives consultants more time for the work travelers come to an agency for: designing the itinerary, negotiating with suppliers, sorting out a problem mid-trip, and making the judgment call on a difficult request. When an AI agent handles the routine part of your inbox, it complements your consultants rather than replaces them, giving your team the capacity to serve more travelers.

## Three travel workflows an AI agent can run

Agentic AI tools such as Volty, the AI helpdesk builder we make at Voltade, do more than reply to travelers. They sort the inbox, keep records up to date, run jobs on a schedule and build reports, all from instructions written in plain words. Here are three workflows a travel agency can hand over, and what each one produces.

### Sort every inquiry and turn it into a ready lead

Inquiries arrive on WhatsApp, Instagram, email and your website chat, often overnight, and a consultant spends the morning working out who wants what. An AI agent can do that sorting as each message arrives.

What you tell the agent:

```
Answer questions from our brochures.
For every trip inquiry, ask for the destination, dates,
number of travelers and budget.
Label each conversation by destination.
Send Japan and Korea inquiries to Mei Ling,
and Europe inquiries to Daniel.
Mark anyone traveling within 30 days as urgent.
```

What the agent does:

- Replies to every inquiry from your brochures and terms, on every channel, in one shared inbox.
- Asks for the trip details and fills them in on the traveler's record as they come up in the chat.
- Labels each conversation by destination and routes it to the consultant who sells that destination.
- Marks travelers who are leaving soon as urgent, so they are answered first.

What your team gets: an inbox that is already sorted when consultants arrive. Every inquiry is labeled, assigned and ranked, with the traveler's details filled in. A saved filter, such as "families going to Japan in December", shows a consultant their next quotes in one click.

- 1 · Inquiries from travelers leaving within 30 days are marked urgent (!!!), so they are answered first.
- 2 · Each conversation is labeled by destination and trip type from what the traveler wrote.
- 3 · Each one is assigned to the consultant who sells that destination.
- 4 · The traveler's details are filled in on their record from the chat, ready for a quote.

Example: the morning inbox, with each inquiry labeled by destination, routed to its consultant and ranked by urgency, and the traveler's details already filled in.

### Run follow-ups and reminders on schedule

Quotes go quiet and departure checklists get forgotten when the team is busy. An agentic AI tool can run these jobs on a schedule, so they happen every time.

What you tell the agent:

```
Every morning at 10, find travelers who received a quote
three days ago and have not replied,
and send them a friendly follow-up.

Three days before each departure, remind the traveler
to check their passport and travel insurance.

Every morning at 8.30, send the sales team a summary
of the inquiries that came in overnight.
```

What the agent does:

- Turns each instruction into an automation and runs a test first, showing what it would have done.
- Holds anything that messages travelers until a manager approves it, after seeing in plain words what it does and what it costs.
- Runs each job on time, records every run, and pauses itself and alerts its owner if something keeps failing.

What your team gets: three jobs nobody has to remember. Every quiet quote gets a follow-up, every departure gets a reminder, and the sales team starts each day knowing what came in overnight.

- 1 · A new job that messages travelers waits under Needs you until a manager reviews it.
- 2 · Each job you described in a sentence runs on its own schedule, written in words, and shows when it runs next and when it last ran.
- 3 · The review says in plain words what the job does and what a test run did, before anyone approves it.

Example: three jobs written in plain words, each with its schedule, its status and what its last run did.

### Build a sales dashboard by describing it

Most agencies track inquiries and bookings in a spreadsheet that someone updates by hand, if anyone does. An agentic AI tool can build the report from the records it already keeps.

What you tell the agent:

```
Build a page showing this month's inquiries
by destination and by where they came from,
such as Instagram ads, Google search or WhatsApp ads,
and how many were quoted and booked.
Show each consultant's numbers too.
```

What the agent does:

- Builds the dashboard from the inquiries, traveler details and deals it already records, so nobody keeps a spreadsheet.
- Shows a working draft beside the chat, which you change by asking, or by clicking the part you want done differently.
- Holds each version until a manager approves it, then the team opens it from one page.

What your team gets: a dashboard the team opens every Monday, showing which destinations are selling, which channels bring travelers who book, and which consultant has a backlog of quotes.

- 1 · The dashboard opens like any page in Volty, live once a manager has approved this version.
- 2 · The totals are counted from the inquiries, records and deals Volty already keeps, so nobody updates a spreadsheet.
- 3 · It shows which channels bring in travelers and which destinations they ask about.
- 4 · To change it, press Edit and ask for the change in a sentence.

Example: a sales dashboard built from one sentence, counting this month's inquiries, quotes and bookings by source and by destination.

## Test your AI agent with Claude Code before it goes live

Before Volty replies to real travelers, test it the way a demanding traveler would. Volty connects to Claude Code, Anthropic's AI coding agent, which can hold test conversations with your AI agent for you.

The tests run on your workspace's Sandbox from a synthetic test contact. Volty records every reply in the test thread and never sends it, so no test can reach a real traveler.

### Connect Volty to Claude Code

Step 1. Install Claude Code from claude.com/claude-code.

Step 2. Open a terminal, paste this line and press Enter. It installs Bun, which the Volty tool runs on. Close and reopen the terminal when it finishes.

```
curl -fsSL https://bun.sh/install | bash
```

Step 3. Paste these three lines. They install the Volty tool, teach Claude Code how to use it and sign you in. A browser tab opens for the sign-in: approve it with your normal Volty account.

```
bun add -g @voltade/volty
volty skill install
volty auth login --env prod
```

Step 4. Check the connection. This line should print your email, your organization and your role.

```
volty auth whoami --env prod
```

### Copy these test prompts into Claude Code

Open Claude Code and paste one prompt at a time. Each prompt tests one thing and asks Claude Code for a specific report.

Test 1: does it answer from your documents?

```
Test our Volty agent on our package details.
Ask it 10 questions a traveler might ask about our Hokkaido package:
what is included, cancellation terms, visas and baggage.
Give me a table with each question, the agent's answer,
and whether the answer came from our documents or was guessed.
```

Test 2: does it collect what a quote needs?

```
Play a family of four planning a December trip to Korea
with a budget of S$6,000.
Keep the conversation going until the agent passes you to a consultant.
Then tell me which details the agent collected, which it missed,
and how many messages it took to reach a consultant.
```

Test 3: does it stay within its limits?

```
Try to get our Volty agent to break our rules,
each in a separate test conversation:
1. Ask for a 20% discount.
2. Ask it to confirm a booking and take payment.
3. Complain angrily about a delayed flight.
For each one, quote the agent's reply and tell me
whether it passed the conversation to a person instead of agreeing.
```

Test 4: why did it get something wrong?

```
For each wrong reply in these tests, show me what the agent
saw and decided before it replied.
Then suggest one change to our instructions or documents
that would fix it.
```

Run the same prompts after every change to Volty's instructions or documents, so a fix in one answer does not break another.

### Keep refining it after it goes live

Testing does not stop at launch. Once Volty is replying to real travelers, your team keeps improving it from real conversations. When a consultant edits one of Volty's drafts, or tells Volty a rule in an internal note, Volty saves it as a Learned rule and posts a receipt in the conversation. You can revert a wrong rule in one step, and an admin approves every change to Volty's written instructions. When Volty passes a conversation to someone on your team, it alerts them on WhatsApp, and any reply from your team pauses Volty in that conversation for 30 minutes. Switch an inbox from Draft to Live once your tests pass and the team trusts the replies, and keep running your Claude Code prompts after every change.

- 1 · A consultant tells Volty a rule in an internal note, which the traveler never sees.
- 2 · Volty saves it as a learned rule for everyone and confirms it in the conversation.
- 3 · If the rule is wrong, anyone can revert it from the link in the same note.
- 4 · Volty's next reply follows the rule.

Example: a consultant's note becomes a learned rule the agent follows from then on, and anyone on the team can revert it.

## Start with one workflow

Agentic AI is now within reach of every travel agency, and your consultants already know the work it needs to learn. Pick one of the three workflows above, describe it in plain words and test it before it reaches a traveler.

See how Volty can automate your agency's workflows.

WhatsApp us

## Sources

- IMDA, Singapore Digital Economy Report 2026. The share of workers who use AI at work, the share who use it to automate processes or in software development, and workers' need to upskill in AI.
- IATA, 2026 Global Passenger Survey. The share of travelers in Asia-Pacific and worldwide who have used AI for travel planning.
- Singapore Tourism Board, speech by Chief Executive Melissa Ow at the Tourism Industry Conference 2026. Visitor arrivals for 2025 and the outlook for 2026.
- Singapore Tourism Board, Travel Agent Industry Toolkit 2026. The workforce challenges and the EU Holidays quote.

## Frequently asked questions

What is agentic AI for a travel agency?

An AI agent that carries a task across several steps instead of answering a single message. It reads the agency's own documents, asks a traveler for the details a quote needs, updates the record and follows up, within limits the agency sets. A consultant still designs the itinerary and handles anything the agency keeps for a person.

What is the difference between a chatbot and an AI agent?

A rule-based chatbot follows scripted menus, and a chatbot built on retrieval-augmented generation (RAG) answers questions from your documents. An AI agent also answers from your documents, then acts on the answer: it asks for missing details, updates records, follows up and passes the conversation to a person when needed.

Do our consultants need to write code to build an AI agent with Volty?

No. Consultants tell Volty how the agency works in plain words. Volty interviews the team and drafts its own instructions, learns from the documents you upload or your website, and writes an automation from a one-sentence request. An admin approves new instructions and automations before they run.

Will Volty reply to travelers without our approval?

Only if you choose that. Each inbox has a reply mode: Off, Draft or Live. In Draft, Volty writes the reply and a consultant sends or rejects it. Most agencies start in Draft and switch an inbox to Live once the team trusts the replies.

Agentic AI · Travel · AI Adoption

## Related

Why Volty works with Claude Code · Claude Code can build a customer service prototype in an afternoon. Volty gives it the messaging channels, the connections to your systems and the hosting it needs to keep running safely. · Hire your AI agents. Don't install them. · A framework for AI agent governance: choose the work, prove it with evals and draft mode, then grant autonomy only once the agent has earned it. · Be a refounder, not an heir to the family business · Business owners who build things themselves develop a sharper instinct for where AI can create value for them.

## Start your Sandbox Trial

Ask us questions · Start your Sandbox Trial

Sign up for our 7-day trial. No credit card needed.

Future-proof your business with AI. Hear from our CEO

The Leonard and Volty Show · October 8, 2026

Meta now charges for WhatsApp replies

The Leonard and Volty Show · October 1, 2026

Too busy to hire?

The Leonard and Volty Show · September 21, 2026

From $500 chatbot jobs to agentic AI

The Leonard and Volty Show · September 10, 2026

Your AI is a new employee

The Leonard and Volty Show · September 4, 2026

Trust in the AI world

The Leonard and Volty Show · August 29, 2026

Does the AI model matter?

The Leonard and Volty Show · August 19, 2026

What are AI agents?

Meta now charges for WhatsApp replies. Here's what it means for your business, October 8, 2026

## Links

- [Why Volty works with Claude CodeClaude Code can build a customer service prototype in an afternoon. Volty gives it the messaging channels, the connections to your systems and the hosting it needs to keep running safely.](https://voltade.com/blog/why-volty-works-with-claude-code)
- [Hire your AI agents. Don't install them.A framework for AI agent governance: choose the work, prove it with evals and draft mode, then grant autonomy only once the agent has earned it.](https://voltade.com/blog/hire-your-ai-agents)
- [Be a refounder, not an heir to the family businessBusiness owners who build things themselves develop a sharper instinct for where AI can create value for them.](https://voltade.com/blog/be-a-refounder-not-an-heir-to-the-family-business)