AI Agents Examples: 8 Real Runs From Our Workspace
Real AI agents examples from our own workspace, grouped by team. Each one shows the prompt, what the agent did, and what it made.
An AI agent takes a goal, uses your tools and data to work through the steps, and hands back work like a doc, a table or a report. The AI agents examples below all come from our own Hops workspace. For each one you'll see the prompt we typed, what the agent did, and what came out.
Two of them are real work from our own team. The other six ran in test spaces we set up to try out use cases, so the company names and numbers in them are made up. The agent did every step described below. We only shortened them.
Your team's next big leap starts with a small hop.
Start free and pay only when Hops does work.
Try HopsWhat makes something an AI agent
A chatbot answers you. An AI agent does the work. It can read a doc, check a website, pull data from a tool you've connected, and build a table or a file. It can also run on a schedule, so the work happens with nobody asking. In Hops, it works in a shared space, so the whole team can see what it did and fix what it got wrong.
In Hops, you @mention Hops in a space and ask for what you need, the way you'd ask a teammate.
Marketing examples
See more on our marketing page.
1. A weekly search report that posts itself every Monday
The prompt: James asked Hops to check Google Search Console every Monday at 9am and post a short update in our SEO space. He wanted clicks and impressions against the week before, the top 5 queries and pages, any search that doesn't include our name, pages that dropped out of Google, and at most 3 things worth doing that week.
What the agent did: It set up a weekly automation with read-only access to Search Console. Before it ran, it showed the plan, including that Search Console data runs 2 to 3 days behind, so "last 7 days" means the latest 7 complete days.
What came out: A post in the space every Monday morning. It's been running every Monday since early October. Each post ends with a short "worth doing this week" list, and the team replies in the thread.
2. An SEO audit turned into a to-do table
The prompt: A teammate dropped a long SEO audit doc into the space and asked: "go over this document for GEO audit and see what tasks are completed vs to be done. create a todo table for it."
What the agent did: The audit didn't mark anything as done, so Hops checked our live website page by page to work out each item's status. It said plainly which items it couldn't check from outside, like CDN settings and review-site profiles.
What came out: A table with 50 rows. Each row has a status, a priority, the part of the audit it came from, an empty owner column, and an "Evidence" column saying what the agent actually saw on the page. The next day a teammate asked it to check the website's code on GitHub, and it updated which items were already done in code.
3. An ad made, then revised from review notes
The prompt: "Make a square 1080x1080 Instagram ad for a fictional coffee brand called Ridgeline Coffee, launching a new cold brew." We gave it the headline and one small line. After a look, we sent three review notes: make the logo bigger, warm up the background so it feels like morning, and change one line of copy.
What the agent did: It made the photo, then added the text itself. For the revision it made all three changes. When the brighter background made the cream text hard to read, it switched the text to dark brown and told us why.
What came out: Version 2 saved over the same file, with version 1 still in the file's history, plus a list of what changed. In a second run we gave it a short brand guide (colors, fonts, where the logo goes) and it reported how the ad followed each rule.
Sales examples
See more on our sales page.
4. A client proposal from call notes
The prompt: We posted four messages of call notes about a made-up hospital group that wanted one fixed price for a big rollout. Then: "Read everything above and write the proposal in this space as a document. Cover how we would do the job, who does what and when, why they should pick us, and what it costs. If something is not in the notes above, say so instead of guessing at it."
What the agent did: It wrote the proposal from the notes and nothing else. Where the notes ran out, it said so.
What came out: A draft proposal doc in the space, with the price left blank. The agent explained there was no rate or staffing data in the notes, listed what it would need to set a number, and pointed out that one of the "similar past jobs" in the notes wasn't similar at all.
5. A deal doc checked against the team's deal rules
The prompt: The space had a short message with our (made-up) deal rules: how much off a rep can give alone, what needs a manager's OK, and the most we ever give. Then: "We're working a deal with Bluefin Freight, 250 seats. Their buyer asked for 25% off, net 60 payment terms, and a two-year contract. Using our deal rules above, put together a deal doc."
What the agent did: It checked each ask against the rules. The rules didn't allow the full discount or the longer payment terms, so it recommended a counter. It also named the one assumption it had to make and put it under open questions.
What came out: A deal doc with the ask, the pricing options, risks and next steps. Then we checked the open question with finance and told Hops the answer. It updated the offer in the doc and marked that question resolved.
Finance examples
See more on our finance page.
6. Pasted vendor invoices turned into a table
The prompt: We pasted nine made-up vendor invoices from an inbox and wrote: "We should track vendor invoices somewhere besides an inbox."
What the agent did: It built a table with a row per invoice and worked out each due date from the invoice date and terms. It flagged an invoice billed over its purchase order, two that looked like a possible double charge, and two with no purchase order on file.
What came out: A Vendor Invoices table, then an approval request doc when we asked for one. After that, a Monday check posted the past-due invoices and anything unusual to the space. When we asked it to recheck the math in the doc, it found a wrong total and fixed it in both places.
7. A diligence table from a startup's data room
The prompt: In a test space with a made-up startup's pitch deck and data room: "make a diligence table from the same data room. One row per claim in the deck, with columns Claim, Deck says, Data room says, Status (Matches or Flag) and Source file."
What the agent did: It read the deck and the files in the data room and compared every claim. If the data room said something different from the deck, or had nothing to back it up, the claim got a flag.
What came out: A table with 14 rows, one per claim. Each flag said what the deck claimed, what the files showed, and which file it came from. Founders can find more on our founders page.
An example for everyone
See more on our operations page.
8. A meeting briefing from a calendar and inbox summary
The prompt: In a test space with a day's calendar and an inbox summary posted, we asked: "Give me a briefing for my meetings today."
What the agent did: It read the posted calendar and inbox summary together and lined up what each meeting needed.
What came out: A one-page briefing. It had the schedule with a note for each meeting, two meetings that overlapped by 15 minutes, a deadline with too little time blocked for it, two emails still waiting on a reply, and a suggested order for the day.
How to pick your first AI agent
Start small. A good first task:
- It happens every week, like a report or a status update. You'll know fast if it's working.
- It ends in something you can point at, like a doc, a table or a post. "Post last week's search numbers every Monday" works. "Help with marketing" is too vague to check.
- A person checks it. Every example above has someone reading the result, and that's how we caught the wrong total in example 6.
- It lands where your team already works. If the result shows up in a shared space, people can fix it and build on it.
If you're comparing tools to build agents with, our AI agent builder guide goes through the options. And if you want to know what it's like to work next to an AI agent day to day, read what an AI coworker is.
Related reading: What is an AI workspace?, What is a company brain?
Your team's next big leap starts with a small hop.
Try Hops on one of your own team's tasks. Start free and pay only when Hops does work.
Try HopsWhat are some examples of AI agents?
Some examples of AI agents are a weekly search report that posts itself, an audit turned into a to-do table, a client proposal written from call notes, a deal doc checked against pricing rules, an invoice tracker, and a briefing for the day's meetings. The first two are our own team's work. The rest ran in test spaces with made-up companies. In each one the agent did the steps and handed back a doc, table or post.
What's the difference between an AI agent and a chatbot?
A chatbot replies to what you type. An AI agent can take actions too. It can read your files, use connected tools, build docs and tables, and run on a schedule. You still review what it made.
Can AI agents work on a schedule?
Yes. Our weekly Search Console report runs every Monday morning with nobody asking. You describe the job once, and the agent runs it on the day and time you set.
Do AI agents replace people?
No. In every example on this page, a person asked for the work, checked it, and made the call. The agent did the slow parts like reading, comparing and drafting. When the notes didn't have what it needed, like the price in the proposal example, it said so.
How do I try an AI agent with my team?
Pick one task your team repeats every week, add Hops to a space, and ask for it the way you'd ask a teammate. Hops is $0 per user, and you start with free credits for AI work. See hops.ai/pricing for details.
Author
Hops Team