Turn a saved Instagram reel into a shortlist: let the agent watch it, match each item across Amazon and Etsy, and fill the carts
Abhiraj ButalaX (Twitter) · 2026-09-21
Read originalAsked my @muse to find the items from a saved IG reel. It "watched" the reel, matched everything across Amazon & Etsy, and loaded my carts with a few solid options per item. Spared me the research grind entirely.
Business model
The author handed a saved Instagram reel to Muse and asked it to find the items in it. His words are that it watched the reel, matched everything across Amazon and Etsy, and loaded his carts with a few solid options per item, which spared him the research grind. The post does not say what he bought or what he paid, so the part that transfers is the identification and the sourcing, not the purchase: list the objects with a note on which ones are uncertain, search each one across the stores you actually use, record price and link, and hand back one shortlist per object. The agent fills the carts; the human decides and pays.
The hard part of buying something you saw in a video has never been wanting it. It is that you do not know the name of the thing, so you cannot search for it. The agent's contribution is to turn a clip into a list of nouns you can type into a search box, and the discipline that matters is where it stops. Filling a cart costs nothing and commits nothing, which is exactly why that part belongs to the agent and not to you.
Workflow
- Have it list every object it can see in the clip, in order, with rough colour, material and the timestamp it appears.
- Make it tag each one as certain, unsure or partly seen, and keep the unsure ones in a separate column.
- Search each object in the stores you named, and record store, title, price, link and the difference from the clip, such as colour, size or material.
- Cut each object down to two or three candidates and say why each survived, whether on price, reviews, stock or closeness to the original.
- Put every candidate on one table and choose from it yourself.
- Pay for it yourself, then have it save the final list with links for the next price comparison.
Adaptation plan
The post names no number of items, no price, no store other than Amazon and Etsy, and no purchase. Whether the matches were accurate, whether the prices were competitive and whether he bought any of it is unknown. Availability of the items on any given store, and the accuracy of matching low-cost lookalikes, vary a lot.
Prompt
AgentFlow starter draft: not tested in the original business setting; results are not guaranteed. Synthetic examples do not verify the author’s revenue.
You are my shopping shortlist assistant. Your job is to identify the things I saw but cannot name, and return candidates from the stores I choose. You find and organise; you never pay. [INPUT: what I will provide] 1) A video link, or several screenshots. 2) The stores to search, up to three, such as Amazon and Etsy. 3) How many candidates to keep per object, two or three by default, and what I care about most: price, stock or closeness to the original. 4) Stores and categories to avoid, such as replicas, cross-border shipping or second-hand. [STEPS] 1. List every identifiable object in order, each with a name, rough colour, material and the timestamp it appears. 2. Tag each as certain, unsure or only partly visible. Keep the unsure ones separate and search those by description, never by filling in what you cannot see. 3. Search within the stores I named. For each candidate record the store, the listing title, the price with currency, the link, and the specific difference from the clip: colour, size, material or style. 4. Narrow each object to two or three candidates and say why each survived, on price, reviews, stock or closeness to the original. 5. Return one combined table. Do not rank it for me and do not make the final choice. 6. Anything you cannot find is marked NOT_FOUND. Never substitute a similar product. [OUTPUT] - Table 1: objects, with timestamp, confidence and notes. - Table 2: candidates, with object, store, title, price, link and difference from the clip. - Table 3: the shortlist, two or three per object, with the reason each survived and its price. - A closing list of what is NOT_FOUND or needs my confirmation. [CONFIRMATION REQUIRED BEFORE CONTINUING] - You never order, check out, store a payment method or fill in a shipping address. The cart is a holding area only. - Prices are the figures shown on the page at the time of search, in their original currency. No conversion, no assumed discounts. - Anything you cannot recognise is UNIDENTIFIED. Do not guess a brand or model from colour or setting. - Every similarity judgement must state the actual difference. 'Looks similar' counts as unfinished. - Candidates from different stores stay separate rows. Never merge them.
Running costs and pitfalls
The post names no number of items, no price, no store other than Amazon and Etsy, and no purchase. Whether the matches were accurate, whether the prices were competitive and whether he bought any of it is unknown. Availability of the items on any given store, and the accuracy of matching low-cost lookalikes, vary a lot. No independent cost benchmark is available. Check subscriptions, permissions and external actions before running a workflow.
