> For the complete documentation index, see [llms.txt](https://docs.unbrowse.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.unbrowse.ai/whitepaper/the-problem.md).

# The Problem

The web was built for humans using browsers, not for agents trying to execute tasks programmatically.

That mismatch creates three common failure modes.

## 1. Custom Site-by-Site Integrations Do Not Scale

Building one-off integrations or scrapers per site is expensive, slow, and fragile.

Even when it works, it creates a maintenance burden:

* site changes break the integration
* long-tail site coverage stays poor
* every team repeats the same route-discovery work

## 2. GUI Automation Is Too High-Friction

Browser automation can work, but it forces the agent through the visible interface:

* load page
* inspect UI
* click
* wait
* handle popups
* retry

That is often the wrong abstraction layer for sites whose real logic lives in structured network requests behind the UI.

## 3. Official APIs Cover Only a Small Slice of the Web

Official APIs are great when they exist and when they expose the needed workflow.

But many useful sites either:

* do not have public APIs
* restrict them heavily
* expose only partial workflows
* make them unsuitable for the exact agent task

## The Real Infrastructure Gap

Agents are often smart enough to understand what to do, but they are missing a stable way to use the web at the level where sites actually operate.

That is the problem Unbrowse is solving:

* learn the internal request patterns
* preserve auth and browser parity when needed
* reuse that work across later runs

## Why The Timing Matters

The whitepaper is directionally right that agent demand is increasing faster than web infrastructure for agents.

The current service is built around it:

* the first request is fulfilled in a cloud browser while the route is learned
* later calls replay over HTTP
* a public registry so common reads need no learning at all
* outcomes verified before they count

## Read Next

* [Mental Models](/whitepaper/mental-models.md)
* [How It Works](/whitepaper/how-it-works.md)
* [Paper vs Product Status](/whitepaper/paper-vs-product.md)