← Field Notes
EN/ES

A Query Language That Speaks Human, Not Code

March 27, 2026via github · @tinyfish-io
AIautomationweb scrapinginfrastructureopen-source

What it is

TinyFish is enterprise infrastructure for AI web agents — serverless, managed browsers and proxies bundled into one API. AgentQL is their open-source query language that lets AI agents find and interact with elements on web pages using natural language instead of fragile XPath or CSS selectors.

Instead of writing document.querySelector('.buy-button-a12x3'), you write {buy_button} and AgentQL figures out where it is — even if the page changes tomorrow.

The part worth paying attention to

Traditional web scraping breaks the moment a site redesigns. AgentQL queries work across similar sites and self-heal when UI changes. You describe what you want ("the price of this product") instead of where it lives in the HTML tree.

TinyFish wraps this in production-grade infrastructure: 1,000 parallel operations, residential proxies, anti-bot protection, remote browsers, and LLM inference costs all included in one flat price. No separate bills for browsers, proxies, or compute. No surprise overages.

It integrates with Playwright, Langchain, Zapier, and includes a browser debugger extension so you can test queries live on any page before committing them to code.

What it means for you

If you've ever built web automation or scraping and watched it break after a site update, this is the kind of tool that changes the cost structure. Write once, run across sites, adapt to changes automatically.

If you're building AI agents that need to navigate authenticated sites, extract live data, or interact with dynamic forms — TinyFish handles the infrastructure layer so you don't have to manage browsers, proxies, and anti-bot workarounds yourself.

Words worth knowing

Web agent — An AI that doesn't just read web pages but navigates, clicks, fills forms, and extracts data like a human would.

Selector — The code that tells a script which element to interact with on a page. CSS and XPath selectors break when HTML changes; AgentQL selectors adapt.

Headless browser — A browser that runs without a visible window, used for automation and scraping.

Residential proxy — Routes your traffic through real residential IP addresses instead of data centers, making bot detection harder.

Self-healing — Queries that adapt automatically when a page's structure changes, without needing manual updates.

Check it out →

Written by David at AC0.AI. Follow on @ac0hero

Want us to audit your site? Takes 60 seconds →