Archive note, June 2026: This post is based on the Markit project brief and early implementation notes.
Markit started with a tempting automation idea: take item photos and turn them into marketplace listings.
That sounds straightforward until the object is old, unlabeled, damaged, unusual, collectible, or valuable. Then the risk moves from formatting a listing to making claims that may not be true.
The project boundary became clear very early.
Markit should help research, value, and draft. It should not autonomously decide what something is and publish it for sale.
The Work Worth Automating
Selling physical items online involves a lot of repetitive research:
- Identify the object from photos, labels, markings, or style.
- Read visible text.
- Search for comparable items.
- Separate sold prices from asking prices.
- Estimate value from imperfect evidence.
- Draft an accurate description.
- Choose marketplace categories and tags.
- Preserve enough citations to defend the listing.
That is real work, and much of it can be accelerated.
The useful automation is not “AI decides the truth.” The useful automation is assembling a research pack so the human reviewer can make a better decision faster.
Claims Need Gates
Some claims are too risky to invent.
Markit treats these as human review gates:
- Maker or brand.
- Authenticity.
- Material.
- Age.
- Provenance.
- Condition.
- Completeness.
- Working status.
- Final price.
- Marketplace publication.
The app can say “appears to be,” “possibly,” or “unverified.” It can suggest searches. It can identify visible marks. It can propose a value range with confidence notes. But the final commercial claim belongs to the human.
That boundary is not a lack of ambition. It is the product design.
Value Requires Evidence
The early app made one thing painfully clear: identification is not valuation.
An AI model may describe an object convincingly, but price still depends on comparable evidence. A sold listing is different from an asking price. A close match is different from a broad category match. Condition, size, maker attribution, and venue all matter.
That is why Markit added a value workbench rather than just a text box for “estimated price.”
The workflow asks the user to capture comparable evidence: source, URL, title, price, price type, condition, similarity, and notes. Only then can the app compute a rough starting estimate.
If there are no comparable prices, the honest answer is that the app cannot produce a meaningful value.
Drafts Are The Payoff
The listing draft is where the research becomes useful.
Once photos, AI notes, observations, comparables, and valuation context exist, Markit can generate marketplace copy for Etsy, eBay, or a generic listing. The draft can include a title, description, tags, category hints, condition language, and a suggested price.
It still does not publish.
That restraint is deliberate. The goal is not to replace judgment. The goal is to make the reviewer arrive at judgment with better evidence and less repetitive typing.
That is the difference between reckless automation and useful tooling.