Every few weeks another company ships an API that turns the web into text an agent can eat, and the reflex is to line them up on a spec sheet: extraction quality, latency, price per thousand pages, does it do JavaScript. By that scorecard, Tabstack — launched by Mozilla-Ocho, a group inside Mozilla — is a competent entry in a crowded category. Send a URL and a JSON schema, get structured data back. Ask it to convert a page to Markdown, and it does. Ask a research question, and it reads multiple sources and returns a synthesized answer with every claim cited. Hand it a plain-language task — "find the cheapest direct flight and start the booking" — and it drives a real browser on the live page, with the browser, the model, and the orchestration all running on Tabstack's side.
That's a real consolidation. The interesting thing is what it collapses. Fetching a page, parsing it, coercing it into your schema, and attributing the result have historically been four separate jobs stitched together with four separate libraries. Tabstack folds them into one call, which means the unit of the web an agent consumes stops being a page and becomes a finished answer. If you've wired up a scraping stack by hand — the Firecrawl / Crawl4AI / Jina lineup is the reference here — you know most of your code is glue between those four stages, and most of your incidents happen in the glue.
But feature consolidation isn't the story, because everyone in this market is consolidating. The story is the posture.
The premise most of this market is built on#
The web-data-for-agents business has a quiet foundational assumption: that the site on the other end doesn't get a vote. The entire toolchain of rotating residential proxies, headless-browser fingerprint spoofing, and CAPTCHA-solving services exists to reach pages that are actively trying not to be reached by machines. It is, structurally, an evasion business. The product is "we can get in anyway."
Tabstack is built on the opposite assumption. It complies with robots.txt by default. It processes ephemerally and states it will not train on your data. Mozilla's framing is that the pages you extract, the answers you research, and the tasks you automate stay yours. Read as a feature list that sounds like boilerplate privacy copy. Read as a strategic bet, it's the whole point — Tabstack is choosing to live on the sanctioned side of the web, and betting that's where the durable business is.
The rest of the category asks how to get past the wall. Tabstack asks to be let through the gate — and is betting the gate is about to be the only way in.
Why that bet has teeth now#
A year ago, "we obey robots.txt" would have read as leaving performance on the table. What changed is that the web started building the wall for real.
In 2025, Cloudflare became the first major infrastructure provider to block AI crawlers by default and launched a Pay-Per-Crawl marketplace where publishers set a price and AI companies decide whether to pay it. That policy has only hardened: from September 15, 2026, new sites default to allowing search but blocking training and agent use on pages carrying ads, and mixed-use crawlers that won't declare their purpose get filtered. Cloudflare sits in front of a large fraction of the web. When that layer flips from "open unless blocked" to "blocked unless permitted," the ground under an evasion-based pipeline shifts from technically hard to economically metered and legally exposed.
In that world the two approaches don't just differ in ethics — they differ in failure mode. A stealth pipeline meets the permissioned web as a wall: rising block rates, per-crawl invoices, the constant maintenance tax of staying one fingerprint ahead of detection. A compliant pipeline meets the same web as a gradient: it degrades to whatever the site is willing to sanction, and keeps returning something instead of a 403. The same shift is visible from the identity side, in efforts to give agents a declared, verifiable way to announce themselves — see web-bot-auth — which only matters if being a known, welcome bot is worth more than being an anonymous one. Tabstack is betting it is.
The caveats worth keeping#
None of this makes the choice free. robots.txt compliance means there are pages a stealth competitor will fetch and Tabstack won't, and for some jobs that gap is the job. A hosted API that runs the browser and model on its own infrastructure is also a dependency and a trust boundary you don't control — the browser-automation options you can self-host trade convenience for exactly that control. And "we don't train on your data" is a promise, verifiable today mostly by reputation; Mozilla has more of that particular currency than most, which is likely the point of Mozilla being the one to ship it.
Judge Tabstack on the spec sheet and it's one more capable box to check in a busy category. Judge it on the bet, and it's a wager that the agent web is about to stop being a free-for-all — and that when it does, the infrastructure that was designed to be welcome will quietly outlast the infrastructure that was designed to win.
