The short version

At AOL 1, the agent has zero context and zero intent. It has no knowledge of your project, your rules, or your decisions. You must explain the context every single time. The output is only as good as what you put in each prompt.

What this looks like

What the agent has
  • General training data from the internet
  • Whatever you type in the prompt
  • Nothing else
What the agent does not have
  • Knowledge of your project structure
  • Your coding style or standards
  • Any decisions you made in previous sessions

The cost

Every session, you spend time re-explaining context that the agent already "knew" last time. The output drifts - the agent makes different assumptions on different days because it has no stable reference point. You are the only source of project memory.

High effort per session
You write long prompts to give the agent enough context to do useful work. This does not scale.
Inconsistent output
Without fixed rules, the agent guesses. The style, structure, and naming change from session to session.
No accumulated knowledge
Every insight, every decision, every fix disappears at the end of the session. Nothing carries forward.

Tools at this layer

Any AI tool used in its default browser state with no project files attached puts you at AOL 1. The tool may be powerful - the layer is defined by what you have given it, not by the tool itself.

ChatGPT
Web interface, no project files. Every conversation starts fresh.
Claude.ai
Browser chat without an attached project. Powerful model, zero project context.
Gemini
Google's chat interface. No file system, no persistent project memory.
Perplexity
Web-first. Great for research and one-off questions, not for ongoing projects.
Any AI chatbot
If you opened a browser tab and typed a prompt - you are at AOL 1.

What it looks like in practice

Some tasks work perfectly at AOL 1. Others fall apart. The difference is whether the task requires project knowledge or not.

Works at AOL 1
"Write a Python script to rename all files in a folder." Standalone task. No project knowledge needed.
"Explain what a webhook is." Conceptual question. General training data is enough.
"Write a subject line for this email." One-shot. You provided the full context in the prompt.
Breaks at AOL 1
"Add a dark mode toggle." Agent invents a solution. Does not know your design system or tokens.
"Fix the auth bug." Agent has no idea how auth is built. Guesses from scratch.
"Continue from where we left off." There is no "where we left off." Every session starts from zero.

When AOL 1 is fine

AOL 1 is acceptable for quick, one-off tasks where context does not matter - writing a single email, explaining a concept, or answering a question. It is not acceptable for any ongoing project where consistency matters.

How to move to AOL 2

Create a CLAUDE.md file at the root of your project. This single file gives the agent a stable starting point. It does not need to be long - even a short description of what the project is and what the main rules are will make a measurable difference.

Next step: AOL 2 - Aware
Add CLAUDE.md. The agent knows where it is and what the rules are. The first real step out of the Friction Zone.
See AOL 2