Two AI labs launched the same feature on the same day. The crypto market didn’t notice. That silence is the signal.
Anthropic dropped “Record a skill” into Claude Cowork. OpenAI pushed an identical “Record a skill” button into Codex. Both let users grab screen activity, clicks, keys—even voice—and package it into a reusable automation workflow. A non-technical analyst can now teach an AI to produce a weekly DeFi dashboard without writing a single line of Python.
Most commentary focused on productivity gains for office workers. But for anyone who has spent years tracing liquidity ghosts through the ICO fog, this is not a story about better spreadsheets. It is a story about automation architecture metastasizing into blockchain workflows—and the macro-risks that come with it.
Context: What Recorded Skills Mean for Crypto
At first glance, recorded skills are just a UX improvement. Instead of manually defining a chain of actions, you demonstrate it once. The AI picks up the pattern, stores it as a structured prompt, and replays it on demand. This is the same principle that powers RPA (Robotic Process Automation) but with an LLM brain replacing brittle script logic.
In crypto, the equivalent is composability—linking protocols, oracles, and bridges into a single atomic sequence. Today, most DeFi automation relies on bots, reserved for developers with API access. Recorded skills lower that barrier. A trader could record a session where they shift collateral across three lending markets to chase the highest yield. The skill then becomes a replicable strategy, shareable with a team or sold on a future marketplace.
The immediate beneficiaries are user‑facing dApps. Imagine a wallet that records your typical swap‑and‑stake routine and lets you replay it each week. Or a cross‑chain aggregator where you demonstrate a bridge‑swap‑farm pattern, and the skill handles the multi‑step execution.
But beneath this shiny surface, the structural fault lines are identical to those I mapped during the 2017 ICO boom. Back then, I modelled on‑chain velocity across 500 token sales. Sixty percent of initial liquidity recycled within four hours. The perceived demand was a mirage—liquidity ghosts haunting the same few addresses. Automated skills risk recreating that illusion at a more granular level.
Core: The Stability Mirage of Recorded Skills
A recorded skill is a frozen snapshot of an environment. It captures the positions of buttons, the layout of dApps, the names of pools, the slippage curves. But crypto markets are not static. A liquidity pool rebalancing, a new front‑end version, a sudden DXY spike that dries up stablecoin flow—all break the recorded path.
Consider a skill that automates a yield strategy on a lending protocol. It records the exact UI elements: the “Supply” button at pixel (450, 620), the dropdown for stablecoin selection, the confirmation modal. Now imagine the protocol updates its interface. The skill fails. Worse, if the skill relies on on‑chain oracle feeds (e.g., for collateral ratios), a flash crash or a liquidator cascade will make the recorded actions counterproductive.
During DeFi Summer 2020, I identified a temporal arbitrage between Uniswap V2 and cross‑border settlement times. I modelled a 15% risk‑adjusted yield advantage. But when I tried to hardcode the strategy into a bot, one variable—the Dollar Index—exploded. My script kept buying a stablecoin that was quickly depegging. Recorded skills are the same trap, just wrapped in a nicer UX.
And there is the data problem. Every recorded session uploads screenshots, keystrokes, and audio to Anthropic or OpenAI servers. For sensitive on‑chain operations—private key handling, governance votes, treasury management—this is a non‑starter. The AI giants become the ultimate oracle, seeing every step before it executes. Decentralization advocates should be screaming.
Contrarian: The Bear Case No One Is Talking About
The “omnichain skill” narrative is VC‑manufactured. Users do not care how many chains your recorded routine touches. They care that it works reliably when they need it, without leaking their private keys or costing an arm in gas.
Recorded skills are inherently fragile. They rely on a centralized inference endpoint. If Anthropic or OpenAI throttles the API, the skill stops. If the model misinterprets a screen icon—think of the “Confirm” button being green in one dApp and blue in another—the skill executes a wrong transaction. The error propagation in a multi‑step DeFi workflow can drain a wallet in seconds.
Then there is the macro layer. A recorded skill that worked perfectly in a low‑volatility, high‑liquidity environment will break when the macro tide turns. Post‑Dencun, blob data will be saturated within two years, doubling all rollup gas fees. Skills that assume cheap L2 transactions will suddenly cost more than the arbitrage they harvest. The skill becomes a liability.
I have seen this pattern before. In 2021, I published a paper on NFTs as digital real estate, linking Ethereum gas fees to US CPI. The correlation was tight—until it wasn’t. When the Fed started hiking, the digital land pricing collapsed. No recorded skill could have adapted because the environment itself changed. The skill is a derivative of the macro cycle, not a substitute for it.
Takeaway: The Real Opportunity Is Not Recorded Skills
The lasting value is not in the ability to record. It is in the ability to continuously re‑record, adapt, and verify against on‑chain data feeds. A skill that can detect a UI change and re‑record itself; a skill that checks a DEX’s TVL before executing; a skill that pauses when volatility exceeds a threshold—that is an architecture with survival properties.
We need skills that are aware of liquidity ghosts. That watch the M2 money supply. That decouple from rigid pixel coordinates and anchor to semantic state: “find the pool with the highest available liquidity, then swap into the stablecoin whose peg is strongest today.”
Tracing the liquidity ghosts through the ICO fog taught me that automation without macro awareness is a ticking bomb. Tracing the liquidity ghosts through the ICO fog reminded me that the easiest path—record and replay—is also the most fragile. Tracing the liquidity ghosts through the ICO fog shows that the real winners will be those who build skills that can re‑record themselves on the fly, reading on‑chain rhythms the way a trader reads order flow.
Anthropic and OpenAI gave us a button. The market gave us a test. The question isn’t whether you can record a skill. It’s whether you can keep it alive when the macro tide turns.