3 AI Agent Use Cases Powered by External Data

AI agents can use external data to research a purchase, check business details on an invoice, or turn a shared travel video into an itinerary. These three AI agent use cases show how Glasser supplies information that an application can turn into a useful next action.

The examples below are hackathon prototypes, with demonstrations of the workflow each team built.

On October 17, 2026, our second hackathon outing brought Glasser to Build Something AI Cannot Do Alone @ Cal in Berkeley. Hosted by DiversaTech Consulting × Build SMTH, the event centered on the human side of building with AI: choosing a worthwhile problem, shaping an experience, and deciding what good results should look like. Glasser supported the event as a sponsor, with a tool demonstration and a Best Use of Glasser track. View the event →

Three projects captured different answers to that challenge. Ponder made research something you could arrange in space. Counter Sign brought external checks into invoice review. TripSync turned friends’ shared travel videos into a plan they could shape together.

For each of these AI agent examples, we look at the user’s problem, the external information involved, and what the application does with it.

1. Ponder: Bring AI Research onto a Spatial Canvas

A chat can help you research a purchase, but comparing the results often means scrolling back through messages. Ponder explores a different interface: a canvas where ideas and research can be positioned, grouped, and revisited.

In his demonstration, MinKi Jung introduces Ponder as a medium for thinking and communication. A voice-enabled agent can move between rooms, focus on ideas, and add nodes to the canvas. He also shows a phone-based controller for navigating a canvas displayed on a larger screen, including tilting and zooming.

The Glasser connection appears in the research workflow. MinKi asks the agent to research a projector purchase and put the results on the canvas. The demonstration then shows new items being added, including a projector search under $500. He identifies Glasser data sources as the information behind this research feature. Watch the Ponder demo →

The useful design idea is that retrieved information becomes something users can organize. A buyer could group candidates, move an interesting option next to a requirement, or keep a question visible while continuing the research. Those possibilities give the canvas a purpose beyond displaying a chat answer.

For builders, Ponder suggests a starting question: once an agent has found information, what should the user be able to do with it?

2. Counter Sign: Check Invoice Signals Before Escalating to a Human

An invoice arrives with a company name, an email address, and a request for payment. Those details create questions: does the sender match the business, and does the document contain something that deserves closer inspection?

Akmal Shovkatov’s Counter Sign prototype addresses that review step. In the demonstration, he describes collecting bills from Gmail on a recurring schedule, alongside manual uploads and a CLI path. The application runs checks, combines their results into a score, and decides whether to ask for human attention.

Akmal identifies Glasser as a source of checks involving business information, email matching, and addresses. The demonstration also describes application-side checks for unusual amounts, look-alike details, and document history. Watch the Counter Sign demo →

This is a useful division of work: external information helps investigate a claim, while the application determines how to combine signals and route a document for review. The demo illustrates an invoice-screening workflow; it does not establish a measured fraud-detection accuracy or prove that a flagged invoice is fraudulent.

For a builder adapting this pattern, the interface matters as much as the score. A reviewer needs to understand which detail triggered attention and what remains uncertain before deciding what to do next.

3. TripSync: Turn Shared Travel Reels into a Group Itinerary

Travel planning often starts long before anyone opens a planning app. Friends send each other videos of restaurants, neighborhoods, and things to do. By the time the trip becomes real, those suggestions are buried in a group chat.

The TripSync team starts its demo with that problem and a proposed Thanksgiving trip to Miami. The group creates a trip and adds preferences such as dates, hotel location, budget, pace, and interests.

Next, they paste an Instagram Reel link. The presenters describe using Glasser to retrieve the Reel’s caption, transcript, and media, with their model identifying the places mentioned. Those places become cards that record details such as who suggested an activity and whether other travelers are interested.

The group can flag concerns about price, distance, or effort. TripSync then demonstrates a day-by-day itinerary that groups nearby places, shows estimated travel times, and allows activities to be rearranged. Watch the TripSync demo →

The product idea begins with something the group has already chosen to share. Glasser supplies access to the source material; the application turns extracted places and personal preferences into a collaborative planning experience. The itinerary and travel times shown are prototype outputs, rather than independently verified travel recommendations.

How External Data Supports These AI Agent Use Cases

The three projects place external data at different points in a workflow:

A common design pattern emerges: start with a specific input, retrieve the information needed to work on it, and give the user a useful next action.

For Ponder, that action is arranging research. For Counter Sign, it is reviewing a suspicious signal. For TripSync, it is deciding together which activities belong in a trip. The experience depends on choices made by the builders: what to ask for, how to organize the result, and where a person should remain involved.

Build Your Own Glasser Use Case

Begin with a task you already understand. Identify the information missing from the current workflow, then decide what someone should be able to do once that information arrives.

You can connect Glasser to your agent with this setup prompt:

set up https://glasser.ai/SKILL.md

Then describe one bounded task: research a purchase, investigate a business detail, or retrieve the source content behind a shared link. Use Glasser’s Search → Inspect → Run workflow to discover an appropriate endpoint, check its inputs and current pricing, and retrieve the data.

Explore the available data sources, or see three more community projects from the AdaL hackathon for other ways to bring outside information into an AI application.