Vincent Jeltsch (@vinckoj) header
Vincent Jeltsch (@vinckoj) profile picture

I'm building sitefire (YC W26)

Munich, Germany
Joined August 2022
Photo by Vincent Jeltsch (@vinckoj)Photo by Vincent Jeltsch (@vinckoj)Photo by Vincent Jeltsch (@vinckoj)Photo by Vincent Jeltsch (@vinckoj)Photo by Vincent Jeltsch (@vinckoj)Photo by Vincent Jeltsch (@vinckoj)Photo by Vincent Jeltsch (@vinckoj)Photo by Vincent Jeltsch (@vinckoj)Photo by Vincent Jeltsch (@vinckoj)
We checked the top 500 websites for 8 AI agent protocols: MCP, WebMCP, A2A, UCP, ACP, AP2, x402 and MPP. 457 of them show no public evidence of any.
2
2
1
3
140
Really cool to see this WebMCP polyfill released. Makes it easier to build and test WebMCP tools without waiting for widespread native browser support. npmjs.com/package/@w3c/webmc…
2
2
37
We checked 500 popular insurer websites for eight agent protocols. Only 4 offer tools an AI agent can use.
1
3
2
67
What do these MCPs look like? NEXT runs a whole quote through one stateful tool. Insurify splits it into consent, submission and results. Jerry exposes eight operations.
1
10
For all of them, checkout still happens on the website. The tools collect facts, price, compare and return a checkout link. AI agents only prepare the purchase for the user.
9
ChatGPT Business Hub, and Business agents currently require a website. But product feeds and other info can already be uploaded to your profile. ChatGPT no longer has to read your website to recommend your business and answer questions about it. Will more SMEs eventually just advertise on ChatGPT and no longer run their own website at all?
12
Caveat: "no evidence" means nothing on the public surfaces we checked (discovery files, the landing page's browser tools, official developer docs). Private integrations might exist.
2
8
Is Agentic Commerce overhyped? We'll run checks for specific verticals next. cc @sitefireai
13
Two weeks ago we showed that AI slop has a shape. Today we're launching a free tool that rewrites your post out of it: the Structural Deslopper.
We trained a model to predict AI-written blog posts from structural features - and discovered the shape of slop. On posts it had never seen, it told AI and human apart with 98% accuracy, getting only 19 of 1,740 wrong. Paper and code: arxiv.org/abs/2609.15369
2
1
83
/6 It won't turn bad content into great content: it measures structure, not quality. But it removes the structural patterns that make a post read like AI.
1
14
/7 Try it for free (30 checks a day): sitefire.ai/deslop Or let your agent run the check-and-rewrite loop: npx skills add sitefire-ai/slop-skills It's not perfect, we will keep improving it. If a score seems off, click "Seems off?". That helps us train the next version.
19