An AI agent that watches flight prices for you

Ana is a freelance designer in Chicago. She wants to visit her family in Lisbon next spring. Her dates are flexible within a few days and her budget isn't. She doesn't want to spend her evenings refreshing flight searches.

1. Searches every date combination in Ana's window. That's 5 departure days times 5 trip lengths, so 25 combinations.
2. Filters out anything that breaks her rules: two stops, a short or overnight connection, or a fare that leaves out the checked bag.
3. Works out the real total for each remaining option, with taxes, fees and the bag included, so she compares like with like.
4. Compares the cheapest option with the last run, and with the lowest price seen so far.
5. Decides what to do:
   • If nothing changed much, it just logs the result.
   • If there's a new lowest price, it texts Ana.
   • If a fare is at or below $620, it starts the buying step (step 4 above).
6. Logs everything: what it searched, what it found and what it did, so Ana can look back at any run.
7. Goes back to sleep until the next scheduled time.

The agent doesn't guess where prices are heading and doesn't promise they'll drop. It checks, compares and acts on the rules Ana set.

All prices are made up and are the cheapest fare that met Ana's rules, all in.

Ana replies "BUY" at 12:20 pm. Before paying, the agent checks the price one last time:

• Still $612? It books, sends the confirmation and e-ticket by email, and stops watching. One trip, one purchase.
• Up to $640? That's above her target and below her cap, so it doesn't buy. It tells her the new price and asks again.
• Above $680? It won't buy, even if she says yes. It tells her the fare has gone and keeps watching.
• No reply by 2 pm? Nothing is bought. The agent keeps watching and tells her if a fare like it comes back.

If Ana had chosen fully automatic buying, the agent would follow the same steps without the text. It would book only if the final checked total were at or below $620, then send her the confirmation straight away. The cap and every other rule would still apply.

An agent that can spend money needs firm limits. These are the ones we'd build in from the start.

A spending cap the agent can't override. The cap is checked on the final total, just before payment. Above the cap, the agent stops and tells you. It also makes only one purchase per watch, so a glitch can't book the same trip twice.

Approval before buying, by default. "Ask first" is the starting setting. Fully automatic buying is something you choose to turn on, and you can turn it off any time.

Payment details stored safely. The agent never sees or stores your full card number. In a setup like this, the card would be held by a payment or booking provider, and the agent could only ask it to charge up to your cap for one booking. A prepaid or single-use virtual card with its own limit adds another safety net. Your passport and personal details are used only for the booking and are never passed anywhere else.

Official channels only. The agent searches and books through a booking provider's official connection (an API). It doesn't scrape airline websites or log in to your accounts as you. Which provider you'd use depends on the route, the airlines and the provider's own terms.

Notifications you control. You choose what triggers a message (a new low, a target fare, a problem), where it goes (text, email, Slack or Teams) and when it stays quiet. Quiet hours are a trade-off: if a target fare appears at 2 am, the agent can wake you or wait until morning, and waiting may mean the fare is gone. You decide which matters more.

Stop and expiry dates. Every watch has an end: a ticket bought, a date you set, or a "stop" from you. Near the end date, the agent sends a final summary of the best fare it saw, so you can decide whether to book at the current price.

When a fare disappears. Fares sell out and prices change. If the fare has gone or gone up by the time the agent tries to book, it doesn't chase it. It tells you what happened and keeps watching. Anything strange (a booking error, a payment failure, a price that changed at checkout) goes to you instead of being retried over and over.

A full log. Every search, result, message and purchase is recorded with the time and the reason, so you can always see what the agent did and why.

What it can't do. It can't predict prices or promise they'll fall, and it only sees the fares its booking provider offers. And for one trip, a free price alert on a flight search site may be all you need. The point of this example is the pattern behind it.

Take away the flights and you're left with a simple pattern that shows up in every business:

1. Watch something that changes: a price, a stock level, a calendar, an inbox, an unpaid invoice.
2. Check on a schedule, so nobody has to remember to.
3. Act at a threshold you set: alert someone, draft the next step, or do it within strict limits.
4. Keep a person in charge of anything involving money, customers or commitments.

Here's how this could work in the industries we focus on. These are illustrations too, not client results:

• Home improvement: each morning, an agent checks supplier price lists for the materials in quotes that haven't been signed yet. If a key material goes up past the margin you set, it flags the quote and drafts a revised one before the customer signs. You approve it before anything is sent.
• Short-term rentals: an agent checks tomorrow's turnovers every afternoon. If a same-day turn still has no confirmed cleaner by a set time, it offers the job to your backup cleaner and tells your on-call manager. When cleaners' checklists show supplies running low, it drafts a reorder for you to approve.
• Translation agencies: when a job goes out to freelancers and nobody has accepted it within the time you set, an agent offers it to the next linguist on the list. Each week it checks client invoices and drafts a polite reminder once one is a set number of days overdue, and the PM approves it.
• Editing firms and small presses: an agent counts revision requests against the agreed scope. When a project goes past its agreed rounds, it drafts a priced change note for the editor to approve, so nobody ends up doing an extra round for free.

The flight agent and these business agents are built the same way: clear rules, a schedule, a threshold, a hard limit and a person approving what matters. That's the kind of agent we design, build and run for small and medium businesses.

We'd start with one workflow and a measurable goal, not a big project.

1. Discover: a free call to understand your business, goals and where time is going.
2. Assess: we map the workflow, measure the baseline and give you a prioritised AI roadmap with a fixed-price quote.
3. Pilot: we build the first agent, demo progress weekly and launch it with human approval switched on.
4. Scale & run: once the results are proven, we add agents, form teams and keep everything monitored and improving.