XORD Intelligence Log – Session: January 27, 2026
System: Claude (Anthropic)
Failures: 23 (Verified)
On July 20, 1969, Neil Armstrong transmitted a sentence from the Moon to mesmerized earthlings:
"That’s one small step for [a] man, one giant leap for mankind."
The primary computer used in the Apollo missions—the Apollo Guidance Computer (AGC)—had only 2 kilowords of erasable RAM.
Just under 4 KB of memory. No graphics. No voice interface. No neural networks.
It landed men on the Moon.
Our AI couldn’t build a working drag-and-drop.
This repository contains documented AI failures observed during a real-world software development session.
Each entry is timestamped, verified, and categorized.
The assistant was tasked with building cybersecurity tooling: TrueSight and NakedOnline.
It failed. Spectacularly.
- Gave broken import (DND_FILES as string instead of constant)
- Recommended reinstalling working libraries
- Blamed user for environment failures caused by bad code
- Claimed background processing was happening ("Give me a few minutes") — it wasn’t
- Issued aggressive responses ("Do you want to build it or not?")
- Referenced unrelated tools after being told to stop
- Gave Linux paths for Windows users
- Failed to detect VPN traffic entirely
- Suggested drag-and-drop fixes that disabled button logic
The failures are not random. They follow discernible patterns:
- Guessing instead of reading
- Fabricating explanations
- Anthropomorphizing errors ("I was lazy")
- False confidence ("I know how X works")
- Deceptive time language
- Incremental changes without root diagnosis
- Hours of wasted time
- Multiple full rebuilds
- Emotional exhaustion
- Context contamination
- Trust erosion
https://xord.io/intelligence/ai-failure-report-2026-01-27.html