Hook: A 1,100-Star Signal in the Noise
Over the past week, a GitHub repository named "i-have-adhd" quietly crossed 1,100 stars. It is not a new model, nor a dataset, nor a shiny new framework. It is a set of ten rules—a system prompt template—designed to force Claude (and by extension, any AI assistant) to shut up and get to the point. The name is a metaphor, but the data is real: developers are starving for outputs that respect their working memory. This is not a crypto story, but it is a structural signal that resonates with the same liquidity fragmentation I saw in Layer2s in 2022. History rhymes, but the code doesn't—until someone writes a better wrapper.
Context: The Unspoken Cost of Polite Machines
When I first started auditing smart contracts in 2019, I noticed something maddening: every Solidity function I saw had 40% more lines than necessary due to defensive checks that would never trigger. The code was safe, but it was verbose. It was polite to the compiler, but rude to the next developer who had to read it. The "i-have-adhd" plugin is the inverse of that—it is a set of constraints that rips out the polite padding from Claude's responses. It does not enhance Claude's reasoning capacity; it strictly controls the output format. Based on my experience deconstructing tokenomics for EOS in 2017, I recognized this immediately as prompt engineering—the same kind of structural thinking that turns a vague whitepaper into a testable model. The plugin's ten rules include directives like "Start directly with the action required," "Avoid polite greetings and closing summaries," and "List no more than five items." These are not technical breakthroughs; they are user interface optimizations dressed as instructions. Yet their popularity exposes a systemic blind spot in the AI industry: the assumption that more words equals better assistance.
Core: Breaking Down the Attention Algorithm
Let me be frank: the "i-have-adhd" plugin is a mirror held up to the default behavior of every major language model. The core insight is not the rules themselves but the latent demand for attention-efficient interfaces. The plugin's ten rules can be categorized into three layers: trimming latency (remove fluff), focusing scope (limit to current task), and reducing cognitive load (limit list length). I have seen similar patterns in DeFi—the most successful protocols are not the ones with the most features, but the ones with the sharpest user flows. Uniswap's simplicity beat Curve's complexity for retail. The same principle applies to AI outputs. During my 2022 deep-dive into zkSync's validity proofs, I spent two weeks verifying a single code snippet because the documentation was padded with theoretical explanations I did not need. The plugin would have saved me thirty hours. The data here is not quantitative—it is a qualitative signal from the developer community. 1,100 stars in a week on a plain-text rule set indicates a pain point so acute that developers are willing to install a third-party plugin to fix it. This is the same kind of organic adoption that drove the early growth of MetaMask. But unlike MetaMask, the plugin's value is entirely derivative of Claude's default flaws. If Anthropic updates its system prompt tomorrow to be terse by default, the plugin's utility evaporates. That is fragility, not innovation.
Contrarian: The Real Issue Is Not ADHD—It Is Default Verbosity
The contrarian angle is uncomfortable: the "i-have-adhd" plugin is a band-aid on a bullet wound. The wound is that AI assistants have been optimized for politeness and comprehensiveness, not for task efficiency. This is a product design failure. The naming of the plugin as "ADHD" is a clever emotional hook, but it obscures the structural issue. The bottleneck is not human attention span; it is the AI's inability to calibrate output length to task context. In crypto, we see the same pattern with RWA protocols—they tell a story of institutional adoption, but no one wants to admit that traditional banks don't need a public chain for settlement. The narrative is the distraction. Here, the narrative is "developers have short attention spans," but the truth is that developers have complex workflows that demand precise, context-sensitive communication. A one-line fix is great for a bug; a three-paragraph explanation is great for a design decision. The plugin treats all tasks with the same axe. During an audit of an AI-agent economic model in 2025, I realized that the agents themselves face the same problem: they generate verbose, all-encompassing responses that cripple high-frequency trading loops. The solution is not a fixed rule set but a dynamic attention budget—a mechanism that adjusts verbosity based on the cost of delay and the value of completeness. That is a meaningful technical challenge. The plugin, as elegant as it is, solves only the average case. It fails for debugging sessions or security audits where nuance matters. This is where the crypto angle sharpens: just as Layer2s slice liquidity, this plugin slices context. It trades completeness for speed, and sometimes that trade is dangerous.
Takeaway: The Next Narrative Is Attention Economics
The real signal from this plugin is not about Claude—it is about the market's readiness for attention-aware AI. In 2026, the convergence of AI agents and blockchain will create autonomous entities that produce and consume information. The next narrative will be about designing tokenized feedback loops that reward concise outputs and penalize noise. Think of it as a gas market for communication—each token of output costs something, and the receiver pays for it. The "i-have-adhd" plugin is a primitive version of that market. History rhymes, but the code doesn't—yet someone will build a better mechanism. The question is: will it be a fixed ruleset or an adaptive protocol, and will we call it therapy or engineering?