# Marcus Webb, Senior ML Engineer at Fieldstone Labs — read of "Search API for AI Agents", June 18 2026

> 9 years writing backend services, last 2 doing agentic tooling. I've integrated Exa, Tavily, Brave Search, and Serper into at least six different pipelines. I have a 4-year-old who naps from 1:30 to 3:00 PM and that's when I evaluate new tools.

## How I got here

I was Googling "search API freshness comparison 2026" because a client's RAG pipeline keeps pulling in deprecated SDK docs and I wanted to see if anything new had shipped. This came up on page two. The URL was a truncated slug — `search-api-optimized-for-ai-agents-that-returns-cu` — which already made me suspicious that something was off. I clicked anyway because the meta description mentioned "stale docs" specifically, which is my exact problem this week.

## What I clicked first

The spec table pulled me in fast. "SEO-Free. No blog spam, no AI-written garbage." and "Ranked by Freshness. Official docs over Stack Overflow 2019." These are the right words. That's a real problem stated cleanly. I was ready to look for the docs link.

Then I saw the subhead: "The Wishdeal Factory scores every idea against 10 Adoptability axes." And I stopped cold.

## Where I paused

The Fermi math. "$-22,000 Year-1 take-home." That number is right there on the page, above the fold, next to a score of 67/100. I sat with that for a solid 30 seconds. I've never seen a product page voluntarily display a projected loss. I genuinely didn't know if that made me trust them more or if it meant the product was broken before I even tried it.

Then I re-read the page and figured out what I was actually looking at. This isn't a search API I can call. This is a dossier for someone to *build* a search API. The thing I came here to use does not exist. The page took me about 90 seconds to figure that out, and I was paying attention.

## What I distrusted

"Honest disclosure: we don't have live customers on this idea yet. We shipped the strategy package; you ship the customer conversations." I respect the transparency but this is buried after I've already mentally committed to evaluating a working product. The hero looks like a tool landing page. It reads like Exa's marketing. It is not a tool.

Also: the product name is the URL slug. That's not a brand. That's a CMS artifact that nobody caught.

The "buyer clarity: 10/10" score is confusing to me because I, someone who is squarely in the target market for the underlying idea, did not understand what was being sold until I read the page twice. If 10/10 buyer clarity produces this much confusion, I'm skeptical of the scoring rubric.

## What would convince me

If someone had already built an MVP of the actual API and had even 3 developers using it in their agent stack, with one quote that said something like "stopped pulling in deprecated boto3 examples," I'd reply immediately. Not a case study. A Slack message screenshot. A Discord thread. Anything that shows the underlying idea got validated beyond Fermi math.

I'd also want to see what the index actually covers. Does it crawl PyPI changelogs? GitHub releases? Official docs only or also curated community sources? The spec table lists outputs, not inputs. I have no idea what's in the index.

## What I'd ask in an email reply

1. Is there a working API I can test right now, or is the $99 tier where I pay to get the code to build it myself?
2. How does this differ from Exa's neural search with a date filter, or Tavily with `include_domains` set to docs sites? I've hacked together something close to what you're describing with those tools already.
3. The scoring shows "financial upside: 1/10" -- is that because the market is too small, too competitive, or because you think pricing won't hold? I want to know which one before I think about building this.

## Verdict: on-the-fence

The underlying idea is real and the problem statement is accurate. But I came here looking for a tool and found a pitch deck for a business I could build. Those are different products for different buyers, and this page is trying to serve both at the same time without being clear about which one I am.

---
*Memo by skeptic persona, generated 2026-06-18. Studio breaks own self-grading loop.*