𝗖𝗮𝗻 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁𝘀 𝗕𝗲𝗵𝗮𝘃𝗲 𝗟𝗶𝗸𝗲 𝗛𝘂𝗺𝗮𝗻𝘀?

I spent 12 hours testing StoryCaptcha. It is an experimental CAPTCHA. It does not ask you to find traffic lights. It asks you to write a story based on a prompt. It then checks your behavior to see if you are human.

I used Playwright, VS Code, GitHub Copilot, and Chromium. My goal was to see if an AI agent can mimic human patterns.

My first run failed with a score of 56/100. The AI was too efficient. It copied and pasted text. It acted immediately without hesitation. It followed perfect patterns. Automation is efficient, but it is not human.

StoryCaptcha shows you the metrics it tracks:

  • Typed vs Pasted
  • Keystroke rhythm
  • Cognitive pauses
  • Mouse path curvature
  • Backspace usage
  • Correction behavior

I began to adjust the agent.

First, I stopped the agent from pasting. I told it to type character by character. The score went up.

Next, I fixed the typing cadence. Humans do not type with perfect timing. We pause to think. We speed up and slow down. I added random delays between keys. The score went up again.

Then, I added mistakes. Humans misspell words. We use backspace to fix them. I instructed the agent to make errors and correct them. The score improved further.

I also changed the mouse movement. Humans do not move mice in straight lines. They hover and explore. I added curvature to the mouse paths.

One metric stayed at 0% even when my score rose. It was key-overlap. Humans often press the next key before releasing the last one. Most automation sends perfectly sequential signals. This is a very hard pattern to fake.

My results:

  • Initial score: 56
  • Intermediate score: 60–70
  • Optimized score: 76–77

The agent eventually passed. But the real lesson was not the score. Most people focus on browser fingerprints or IP addresses. This experiment proves that behavior is a powerful signal.

It is not just what you do. It is how you do it.

As AI agents grow, behavioral detection will become vital. The hardest part of being human is not the task. It is the imperfection.

Source: https://dev.to/shahraan_hussain_b42640e7/can-an-ai-agent-behave-like-a-human-a-12-hour-experiment-with-storycaptcha-1661

Optional learning community: https://t.me/GyaanSetuAi