A "watch list" agent is the simplest, most useful agent most people never bother to set up. It polls a small number of listings on your behalf, applies criteria you specify once, and alerts you when something matches. Compared to refreshing a search tab manually for two weeks, the agent gives you back the time and catches matches that show up at 03:00 IST when nobody is watching.

This walkthrough covers the two most common variants (apartments and flights) and the rules that keep the agent working without getting blocked, banned, or buried in noise.

What this agent does

The agent runs on a schedule (every 15 minutes for apartments in hot markets, hourly for flights). On each run, it queries the configured sources, normalises results, scores each item against your criteria, dedupes against everything seen previously, and emits an alert for new high-score matches.

It does not book anything. It does not click "Apply" on a rental. The agent's job ends at the alert; the action is yours.

For the broader pattern of read-only agents, see what an AI agent can actually do and AI agent for inbox triage.

Criteria as code

The single biggest lever on signal quality is criteria specificity. Vague criteria flood you with alerts and you stop reading them within three days.

For apartments, specify:

For flights:

Watch sources

Pick sources by API quality first, scraping ease second. For flights:

For apartments, regionally specific sources matter more than category leaders. Some sites publish RSS or API feeds for new listings; others do not. The agent should prefer feeds where they exist and avoid headless-browser scraping unless the platform explicitly permits it.

Cadence and rate limits

Cadence is a function of two things: how fast listings turn over on the platform and what rate limit the platform allows.

The agent must identify itself in the User-Agent header (it is a polite signal that you are not a malicious scraper) and obey robots.txt. For more on agent rate-limit hygiene, see how to handle agent rate limits.

Alerts and noise control

Alert noise is what kills these setups. Three controls keep noise low.

Tune the threshold over the first week. Aim for 1 to 5 alerts a day. More than that and you stop reading; fewer than that and you miss things.

Read the Terms of Service before configuring the agent. Most flight aggregator APIs (Amadeus, Skyscanner) explicitly permit personal use. Apartment platforms are mixed. Universal rules:

For broader safety patterns, see AI agent safety and guardrails.

Common mistakes

Frequently asked questions

Can an AI agent monitor apartment listings and flight prices for me?

Yes. The agent polls a watch list (rental sites, Google Flights, Kayak, Skyscanner) on a schedule, applies your criteria (price, location, dates), and notifies you when a match appears. Most legitimate setups use the platforms' official APIs or RSS feeds where available, and respect rate limits and Terms of Service when scraping is necessary.

What criteria should I give the agent?

Be precise. For apartments: price ceiling, neighbourhood polygon (not just a city), bedrooms, square footage, must-have features (lift, parking, pet-friendly). For flights: origin and destination, date window, max price, max stops, max layover duration. Vague criteria produce noisy alerts that train you to ignore them.

How often should the agent poll?

Apartments in hot markets often warrant a 15-minute cadence during waking hours. Flights are slower; hourly is plenty unless you are watching a fare class that sells out fast. Always respect the platform's robots.txt and rate limits, and prefer official APIs (Skyscanner, Amadeus) over scraping where the API supports your query.

Is it legal to scrape apartment or flight listings?

It depends on the platform. Many flight aggregators publish official APIs (Amadeus, Skyscanner, Kiwi) that are licensed for personal use. Apartment sites are inconsistent: some offer feeds, some forbid scraping. Always read the Terms of Service before configuring the agent, prefer APIs, and respect robots.txt. The agent should identify itself with a User-Agent and obey rate limits.

How does the agent decide what to alert vs ignore?

Match score thresholds. Each listing is scored against your criteria; only scores above a threshold trigger an alert. Below threshold listings are logged but silent. Tune the threshold over the first week so you get 1 to 5 alerts a day, not 30. The agent should also dedupe (same listing reposted does not re-alert).

Three takeaways before you close this tab

Sources