Writing
Essays on AI tooling, plus launch stories from the founders themselves.
We measured the HTML document of all 701 listings — not the images, not the scripts they pull in, just the markup. 114 are over half a megabyte, 35 are over a megabyte, and the heaviest is 2.7 MB. Compression hides it from you and not from the browser.
A title tag saying "Ollama" ranks for people who already know Ollama. We read the title and meta description of all 701 listings — 60 are bare brand names, 125 are too long to show in full, and 41 have no description at all.
A machine found a price on 512 of them. Reading 92 of those by hand found that one in ten figures was not a price at all — a funding round, a benchmark table, an example query. Corrected, roughly 65% publish a price and 35% do not.
GitHub released an update to the Copilot impact dashboard on 7 August 2026, adding a potential return on investment section that links spending to pull request output. The release also includes adoption cohort user counts that reflect users active across the full 28-day window.
The 2026-07-28 revision removes the initialize handshake and the Mcp-Session-Id header. Our server answers both revisions from the same URL and decides which rules apply from what each request carries. One method appears not to exist until you ask in the right dialect.
Baseten has become a supported inference provider on the Hugging Face Hub, allowing developers to run open-weight language models directly from model pages, client SDKs, and agent harnesses. The initial integration brings support for conversational and text-generation tasks using models such as Kimi K3, DeepSeek V4 Flash, and GLM-5.2.
Cloudflare has announced a developer preview of WebMCP, allowing websites to add tool interfaces for browser-based AI agents without changing their origin code. The experimental standard ships in Chrome 146 and uses an edge injection to serve a bridge script and activate browser-based tool packs.
Cloudflare announced the release of the Model Context Protocol 2026-07-28 specification, which shifts the protocol from stateful connections to a fully stateless design. The update removes session headers and handshakes from the core request path, allowing servers to run directly on serverless infrastructure.
97 sites have no og:image at all. Eight point at a URL that 404s. Six serve an SVG that most scrapers will not render. And 18 are using a square logo, which is the one that gets you a thumbnail instead of a card.
Titles pass on 99% of them, logos on 99%, meta descriptions on 96%. The thing that actually stops a tool being listed automatically is bot protection — and the sites doing it are the ones that can afford to.
Five give a do-follow link, four give no-follow, three are mixed. One charges $497 and tags every outbound link no-follow. One does not link to you at all — it sends visitors through a redirect.
Twelve "alternatives" searches, 336 results, 248 different sites. Reddit appears in every one. Not a single general AI-tool directory appears anywhere. And the difficulty score every keyword tool reports for these searches is wrong.