The Problem: AI Agents That Talk Too Much

Large language models, especially those fine-tuned for coding assistance, often suffer from verbosity. When asked a question, they don't just give you the answer; they provide a lengthy preamble, a detailed explanation of the underlying concepts, a history of the technology, and sometimes even philosophical musings on the nature of code. For developers, who are often pressed for time and seeking direct solutions, this can be incredibly frustrating. It’s like asking for a screwdriver and receiving a full toolkit, a lecture on metallurgy, and a biography of Henry Ford.

This tendency to over-explain, while seemingly helpful, actively hinders productivity. Developers need to quickly identify the core information—the code snippet, the command, the configuration setting—without sifting through paragraphs of extraneous text. The signal gets lost in the noise.

Introducing 'I-have-ADHD': A Prompt Engineering Solution

A new, deceptively simple technique, dubbed "I-have-ADHD," has emerged on Hacker News as a way to combat this over-verbosity in AI coding agents. The core idea is to prepend a specific phrase to your prompt. This phrase acts as a directive, subtly guiding the AI to prioritize conciseness and relevance.

The prompt structure is straightforward:

I-have-ADHD: [Your actual coding question here]

For example, instead of asking:

How do I install the latest version of Node.js on Ubuntu?

You would ask:

I-have-ADHD: How do I install the latest version of Node.js on Ubuntu?

How 'I-have-ADHD' Works (Theories)

While the exact mechanism is not fully understood, several theories explain why this simple prefix is so effective:

  • Implicit Instruction: The phrase "I-have-ADHD" might trigger a learned association within the model's training data. Models are trained on vast datasets, including forum discussions, documentation, and user queries. It’s plausible that in contexts where users mention ADHD, the subsequent text is often more direct, less prone to tangents, or focused on immediate task completion. The AI might be inferring a user preference for brevity and directness.
  • Contextual Shift: Adding "I-have-ADHD" could be subtly shifting the AI's internal representation of the user's intent. Instead of assuming the user wants a comprehensive tutorial, the AI might interpret the prompt as a request for the most critical, actionable information, as if the user has limited attention span for preamble.
  • Bias Towards Action: The phrase itself implies a state of being where focus is a challenge. The AI might then prioritize delivering the "action" (the answer) quickly, before the user's (simulated) attention wanes. This is akin to how a human might respond to someone who states they have ADHD – by getting straight to the point.

The surprising detail here is not the cleverness of the phrase itself, but its effectiveness across various LLM-based coding assistants. Users have reported success with tools like GitHub Copilot, ChatGPT, and others, suggesting that the underlying language model architectures are susceptible to this form of meta-instruction.

Practical Implications for Developers

The immediate benefit is a significant time saving. Instead of parsing lengthy responses, developers can get the code, command, or configuration they need much faster. This allows them to move on to the next task, reducing context switching and improving overall workflow efficiency.

Consider a developer debugging a complex issue. They might ask an AI assistant for a specific error code explanation. Without the "I-have-ADHD" prefix, they might receive a 500-word essay on error handling best practices, when all they needed was a two-line code fix or a single command to check a log file. The prefix cuts through that.

This technique is particularly valuable for developers who are:

  • Working on time-sensitive projects.
  • Dealing with complex codebases where quick answers are crucial.
  • New to a particular technology and need direct examples rather than theoretical explanations.
  • Simply overwhelmed by the default verbosity of AI tools.

What This Means for AI Development

The success of the "I-have-ADHD" prompt highlights a broader challenge in human-AI interaction: aligning AI output with user needs and cognitive load. While LLMs are powerful, their default behavior isn't always optimal for every user or every task. This points to a future where:

  • More sophisticated prompt engineering: Users will develop and share more nuanced prompts to tailor AI behavior.
  • AI customization: Future AI assistants might allow users to set explicit verbosity levels or preferred output styles, perhaps even allowing them to input their own 'cognitive profile' for better interaction.
  • Model training: Developers of LLMs might consider incorporating user preferences for conciseness more directly into their training or fine-tuning processes, potentially through reinforcement learning from human feedback (RLHF) focused on desired output length and relevance.

The "I-have-ADHD" skill, while informal and community-driven, serves as a powerful demonstration of how a small tweak in user input can dramatically improve the usability of complex AI systems. It’s a reminder that sometimes, the simplest solutions are the most effective, especially when they address a core user pain point.

What nobody has addressed yet is how this technique might fare when applied to non-coding tasks, or whether similar 'trigger phrases' could be discovered for other common AI output issues, like hallucination or over-generalization.