Everyone is watching the AI-crypto convergence narrative. Tokenized agents. Autonomous treasuries. Algorithmic market makers. But the real bottleneck isn't inference speed or model size. It's output format. A Claude Code skill called 'i-have-adhd' racked up 1,100 GitHub stars in weeks. Not because it enhances reasoning. Because it forces the model to shut up and deliver.
Context: The AI-Agent Economy Is Here, But Clunky
I've been mapping the intersection of AI and blockchain since 2021. By 2026, my models predict a 300% increase in on-chain micro-transactions driven by autonomous agents. But these agents don't just need to be smart. They need to communicate efficiently. In DeFi, a slow or verbose response can mean missed arb opportunities. In DAO governance, a bloated summary can lead to decision fatigue. The problem is structural: large language models default to polite, verbose outputs—optimized for chat, not for task execution.
Enter 'i-have-adhd,' a third-party skill for Claude Code. Its 10 rules are brutal: start with the action, no greetings, no repetition, max 5 bullet points. It strips the fat. The market response—1,100 stars, viral adoption among developers—proves that the demand for 'bot, just tell me what to do' is massive and underserved.
Core: Deconstructing the Plugin as a Macro Asset
I audited this plugin the same way I audit tokenomics. First, the technical layer: zero innovation in architecture. It's pure system prompt engineering. No weight changes, no fine-tuning. Its value lies in the precise mapping of human cognitive constraints onto AI behavior. Rules like 'only keep currently needed info' mirror Miller's Law—the brain can hold 7±2 items. This is not a gimmick. It's applied cognitive science.
Second, the social collateral layer: the plugin's name 'i-have-adhd' is a masterstroke of cultural signaling. It doesn't just describe a feature. It builds an identity bond with users who feel overwhelmed by information overload. That emotional resonance is a form of social collateral—hard to replicate, easy to dismiss. In crypto terms, it's a brand moat.
Third, the regulatory risk lens: Anthropic (Claude's parent) will likely absorb these rules into their official system prompt. When a platform absorbs a third-party tool, the original loses its reason to exist. This is the same dynamic we saw with Uniswap's fee switch—dependency on a platform is a risk. The plugin's survival depends on either staying ahead of Anthropic's defaults or pivoting to a specialized niche.
Contrarian: The Decoupling Thesis Is Wrong Here
Most analysis on AI-crypto convergence focuses on 'agent intelligence'—can the model trade, govern, execute? That's the foam. The real tide is 'agent usability.' The 'i-have-adhd' plugin reveals a truth that the hype machine ignores: the last mile of AI deployment is not intelligence but format. A model that takes 3 seconds to output a 500-word explanation before giving the action will lose to a model that outputs the action in 0.5 seconds.
Crypto is particularly sensitive to this because of latency and gas costs. Each wasted word on-chain costs money. Each verbose API response slows down a trading bot. The contrarian angle: we should be investing in prompt engineering infrastructure—tools, templates, marketplaces—as much as in model weight improvements. The alpha is extracted from chaos, not from intelligence.
Takeaway: The Next Macro Signal
This plugin is a micro-signal of a macro shift: the bottleneck in AI-crypto is not compute or data, but output design. As agents become the dominant users of blockchains, the ability to say exactly what needs to be said—no more, no less—will determine which protocols capture the most transaction value. I'm pricing this risk into my allocation matrix. The signal is silent until the noise collapses. The noise says 'better models.' The signal says 'better outputs.'
Mapping the tides while others chase the foam.
Alpha is not found, it is extracted from chaos.
Culture pays dividends long after the hype fades.