1. Where it comes from
IP reputation is checked first. Cloud servers inherit the history of their address ranges, scrapers included. The same script can score human from an office and bot from a data center.
The short answer
Sites score every visit on signals a scripted browser gets wrong: a datacenter IP address, a connection fingerprint that doesn't match the browser it claims to be, automation flags, robotic input timing and request rates no person produces. Agents that work in the user's own browser, session and IP give those defenses nothing unusual to flag.
A score first, then a decision
Bot-management tools quietly give every request a score from "very likely bot" to "very likely human." The site owner decides what each score triggers: allow, challenge, ask for a second factor, slow down or block. Your agent only sees the result.
The five layers
- 1IPWhere it comes from
- 2connectionHow it connects
- 3browserWhat the browser reveals
- 4behaviorHow the session behaves
- 5rateHow fast it goes
2. How it connects
Before the page loads, the server fingerprints how the connection was opened. A Python or Go client claiming to be Chrome contradicts itself before sending a single form field.
3. What the browser reveals
Automated browsers carry tells: a flag that says "I'm being driven," a headless label, leftover script hooks. Invisible checks probe for them without showing a puzzle.
4. How the session behaves
Pages watch mouse movement, typing and scrolling. No movement in the first seconds, clicks with no pointer travel, or perfectly even keystrokes all look automated.
5. How fast it goes
Too many requests and the site answers "429, too many requests," often with a wait time. Retrying right away turns a slowdown into a block.
Which layer stopped you?
Pick the symptom you're seeing.
Likely layerIP or connection
First checkRun the same script from an office network
Likely layerBrowser or behavior
First checkRun in a normal installed browser with a real profile
Likely layerRate
First checkRead the wait time and count requests per minute
Likely layerStricter rules on key actions
First checkExpected; hand that step to a person
Fixes that don't last
Each works for a while. Each becomes maintenance against a security team paid to beat it, and some break the site's terms.
- Stealth patches that hide automation flags
- Residential proxy pools
- Captcha-solving services (they answer the puzzle, not the score behind it)
- Libraries that fake browser fingerprints
What holds up
Use the front door
If the site offers an API, export or feed, use it.
Ask the owner
Business portals can often allowlist a known account or address.
Identify honestly
Some platforms now let hosted agents sign their requests so sites can allow verified agents.
Run where the user works
Their installed browser, signed-in session and office network, with human-paced input from the operating system.
Pace the work
Honor wait times and schedule batches off-peak.
Hand challenges to a person
Never automate a captcha. It's a message from the site owner.
Terms of use apply to all six. Automate Without an API →
Laptop vs cloud
Same code, different score. In the cloud the agent arrives from a datacenter address, usually in a headless browser with no fonts, GPU or history like a real machine. Every layer reads the difference.
How LaunchAI runs
LaunchAI agents drive a normal Chrome or Edge profile on your machine, with operating-system mouse and keyboard input at human pace, inside sessions you already opened. No headless browser, injected script or extension. If a captcha or multi-factor prompt appears, a person takes over and hands the run back. Product → More on where agents run: Local vs Cloud Agents →
Questions
Does solving the captcha fix the block?
Rarely for long. The captcha appeared because the score was low, and the next request is scored on the same signals.
Do signed agent requests help an agent in my own browser?
Not much. Signing helps hosted agents prove who they are. An agent in your signed-in browser already arrives as you.
Keep learning
- Local vs Cloud Agents Decision one: where the agent executes, meaning whose machine, browser, session and credentials. Read the guide
- What Is a Computer-Use Agent A computer-use agent is AI software that operates a computer through its screen. Read the guide
- What Is an ADK An agent development kit (ADK) is a packaged set of building blocks for creating, testing and running AI agents: a model interface, tools that act on systems, orchestration, guardrails, human checkpoints, evaluation, tracing and a runtime. Read the guide
No prompts. No babysitting. No Integrations.