01 / What I built
Less checking. Less starting from scratch.
I wanted AI to handle useful work between conversations. My eBay operation gave me a practical place to build that: research to gather, listings to prepare, offers to review and orders to organize.
Doing all of that manually meant checking the same places, repeating the same instructions and rebuilding similar outputs. I used Muse, with an assistant I called Astro, to turn those repeated tasks into workflows I could run and improve alongside my other responsibilities.
My role
I designed the workflows, set the rules, tested the outputs and adjusted the system as I used it. I used AI to help with research, preparation and supporting tools, while keeping final decisions and publication with me.
02 / How it worked
Background work, with a clear handoff to me
- 01Watch & gatherScheduled checks collect relevant updates
- 02Filter & prepareApply my rules and build a reviewable output
- 03 / My decisionReview & actI approve changes, replies and publication
- 04Improve & repeatTurn corrections into reusable instructions
Monitoring ran on a schedule. Research and preparation ran when an item, batch or decision needed them.
Monitoring without constantly checking
I set up sale and offer checks every two hours, plus a separate monitor for potential sourcing opportunities. That moved routine checking into the background while I focused on other work.
The monitors used filters and previously seen messages to avoid repeating the same alerts. Sourcing checks had specific criteria, so a quiet run could stay quiet. The aim was to bring me something worth reviewing.
Researching a batch instead of one item at a time
For a 26-listing audit, the research ran across five parallel batches and came back as one combined review. I could work through the findings together rather than manually repeat the same searches for every item.
I set the comparison rules and checked the recommendations. The assistant gathered and organized the evidence; I decided what to do with it.
Preparing listings without rebuilding each one
I used a reusable format for titles, descriptions, item details and shipping information. The workflow produced 34 HTML listing packages and helped prepare drafts directly in Seller Hub using my photos.
That gave me a consistent starting point to review, rather than a blank page each time. I checked the item details and published the listings myself.
Keeping context and building practical tools
I gave the assistant reusable instructions for my preferences, packing constraints and approval rules, so each session could pick up where the last one left off. I also used AI to build supporting tools, including an address-label app, packing-slip templates and local Python and Playwright automation.
I wanted the setup to fit how I actually work. Local tools used double-click launchers on my Mac, making them easier for me to run and manage.
03 / How I improved it
I kept refining the system as I used it
The first version was a starting point. Running the workflows against real items and real tasks showed me where the instructions needed to be more specific.
Better research inputs
When product comparisons mixed up boxes and tins, or different collectible variants, I tightened the matching rules. Accurate research started with identifying the right item.
More consistent outputs
Listing reviews exposed differences between the description and the form fields, including shipping and handling details. I added those cross-checks to the review process so the same issue was easier to catch next time.
Control over what kept running
I paused a Discord monitoring workflow after a rate-limit concern and disabled advertising checks when they were no longer useful. Managing the setup included deciding which automations still deserved to run.
Background monitoring had room to operate on its own. Publishing, account changes, offer responses and buyer messages came back to me for approval. That division let me delegate repeated work while staying responsible for the outcome.
04 / What I bring
I can turn a recurring task into a working AI workflow
This project shows how I work with AI beyond individual prompts: identify the repeated work, define the inputs, connect the steps, test the output and keep improving the instructions.
I built and managed background monitoring alongside batch research, browser-based preparation and reusable tools. The practical change was moving repeated checks and first-pass preparation into the system, so I could spend my attention on review, exceptions and decisions.
That is the same approach I bring to marketing and operations: understand the task, make the process easier to run and stay close enough to the output to improve it.
Quick questions
How I used AI in this project
What did I automate with AI?
I set up recurring sale and offer monitoring, prepared listing packages and browser-based drafts, and used AI to organize research and shipping materials. These workflows handled repeated checks and preparation while I reviewed the decisions that needed my input.
What ran in the background?
Scheduled monitors checked for sales, offers and potential sourcing opportunities every two hours. They used filters and previously seen messages to keep the updates relevant. I kept listing publication, account changes and buyer communication under my control.
What was my role in building the system?
I defined the workflows, set the rules, tested the outputs and improved the setup as I used it. I used Muse for research and preparation, and AI assistance to build supporting tools. I managed the operation alongside my day job.
Personal project, September–October 2026. Counts describe prepared outputs and the audit scope; the two-hour interval describes the monitoring schedule.