From Web3 to AI: What Actually Transferred (And What Didn't)

What the Web3 Years Actually Looked Like

Between 2021 and 2024 I built on Bitcoin. Not DeFi, not Solidity, not the EVM ecosystem — Bitcoin. Ordinals, specifically. If you weren't in that world: Ordinals are inscriptions etched directly onto satoshis, enabling NFTs and interactive applications to live fully on-chain, without IPFS, without a server, without any external dependency. It was one of the most technically honest things happening in the space.

I worked on Pizza Ninjas — a blue-chip Ordinals collection that crossed into the traditional art world when Sotheby's sold one for $139,700. I worked on Pizza Pets, an on-chain game that accumulated over a million on-chain interactions. Neither of these were marketing stunts dressed up in code. They were genuinely hard engineering problems on a blockchain that wasn't designed for this use case.

The bull years (2021–2022) were what you'd expect: rapid capital, rapid hiring, rapid mistakes. Everyone was shipping. The bear years (2022–2024) were more interesting. The people who stayed were the people who actually cared about the technical problem. The degenerates and the flip-chasers left when the prices did. What remained was a smaller, tighter community that was building for different reasons — either because they believed in the long-term thesis or because the problem itself was genuinely interesting. I was in the second camp.

What survived the bear market wasn't protocols or tokens or even projects. It was relationships, specific technical skills, and a framework for thinking about distributed systems where you cannot trust any single party. That framework turned out to be more transferable than I expected.

The Moment AI Became Real

I watched the ChatGPT launch in November 2022 with mild interest. It was clearly impressive. It was also clearly a demo. Every developer who had used language models before could tell the difference between "this is a polished product" and "this changes how I work." ChatGPT was the former.

The actual moment came about a year later, when I was in the middle of a project and needed to process a large set of unstructured JSON responses from a chain indexer, normalize them, and push them into a structured format for a dashboard. A task I'd done variations of dozens of times. Forty-five minutes of tedious wrangling, usually.

I described the problem to Claude. Pasted a sample. Got back working code in thirty seconds. Not pseudocode. Not pseudocode that almost worked. Working code that I ran, it passed, and I moved on.

That was it. Not the benchmark scores. Not the capability research papers. The moment production work got meaningfully faster on a task I already knew well — that was when I understood this was real.

From there it accelerated. I started using AI tools not as a novelty but as infrastructure. Writing pipelines that routed through LLMs. Replacing brittle rule-based logic with model calls. Building agentic workflows in n8n where Claude would receive context, make a judgment, and pass structured output to the next step. The same muscle memory from web3 — composing trustless components, thinking about state, being paranoid about failure modes — applied almost directly.

What Transferred

Tokenomics thinking → incentive design. Anyone who has shipped a token has spent serious time thinking about incentive alignment. Who participates, why, what they get, what they lose, what the equilibrium looks like, how the system fails if someone games it. That thinking — divorced from the token itself — is exactly what you need when designing agentic systems.

An AI agent that autonomously routes tasks, queries tools, and calls APIs is a system of incentives. What does the model optimize for? Where does the prompt create perverse outcomes? What happens when a downstream API is slow or returns bad data? These are system design questions, not AI questions. Web3 trained me to ask them.

On-chain coordination → agentic state management. Blockchains are state machines. Everything is a state transition, every state transition needs to be valid, and you need to be extremely explicit about what valid means because nobody's coming to fix it if you got it wrong. Building for on-chain taught me to think about state clearly: what do I know, what am I assuming, what happens to each of these at failure.

Agentic workflows have the same requirement. An n8n workflow where Claude is making decisions, calling external services, and writing back to a database will fail in ways that are hard to debug if you haven't been explicit about state. Web3 developers, in my experience, are better at this than most software engineers. Not because they're smarter — because the environment punished sloppiness in ways that a web2 codebase never did.

Community → distribution. This one is less technical but equally real. Web3 projects succeed or fail on community. Not social media presence — actual community: people who understand what you're building, why it matters, and will tell others about it because they believe in it, not because they're paid to.

That distribution muscle is directly applicable to AI tooling. The best AI products I've seen get adopted are the ones where someone — usually the builder — has cultivated genuine trust with the people they're selling to. The web3 experience builds that muscle, sometimes painfully.

What Didn't Transfer

The speculative mindset is the obvious one. Web3 trained a whole generation of builders to think in terms of token price as a proxy for product-market fit. High price, good product. Falling price, pivot or die. That feedback loop is completely absent in AI work. There's no token. There's no speculative instrument that tells you whether your workflow is valuable. You have to do the old-fashioned thing: talk to users, look at whether it saves time or money, ask if they'd pay for it.

This sounds obvious. It wasn't, for me. The reflex to look for a market signal — some external indicator of whether you're on the right track — had to be unlearned. Value in AI work comes from outcomes that are often slow to measure and deeply contextual to each customer.

Token-everything thinking is the adjacent problem. Not every problem is better with a token. Not every coordination problem needs a blockchain. Not every AI workflow needs an on-chain component. I spent longer than I should have looking for ways to combine these two worlds when the honest answer was often: just use the right tool. Sometimes that's an LLM. Sometimes it's a spreadsheet. The web3 reflex to reach for the most powerful and complex solution has a real cost when simpler things work.

The community-as-product trap is a subtler failure mode. Web3 rewards builders who can grow communities. AI work, at the enterprise and SMB level where most of the actual money is, rewards builders who can be invisible. The client doesn't want a community. They want a workflow that runs without them thinking about it. The skills don't cancel out — but the orientation has to shift, and that took real adjustment.

Where the Money Actually Moved

Here's what I observed between 2023 and 2025, watching both markets closely.

Capital moved away from protocol speculation — buying tokens in anticipation of network effects that never materialized — and toward a much simpler question: can you make my existing product do more with less headcount? That question doesn't require a token. It doesn't require a whitepaper. It requires someone who can sit inside a product, understand the workflows, and integrate AI in a way that the team will actually use.

The companies spending money on AI right now are not mostly AI-native companies. They're logistics companies, legal tech companies, media companies, e-commerce operators, professional services firms. They have products that work. They have operations that are expensive. They want to automate the expensive parts without rebuilding from scratch. That's a services problem, not a research problem.

Protocol speculation is not gone. It will cycle back. It always does. But the sustainable revenue in this period — the revenue that doesn't require timing a market — is in integration work. Building the bridge between a company's existing systems and what the current generation of AI tools can actually do.

The Most Underserved Niche Nobody Talks About

Web3 teams need AI. Urgently, in many cases.

These are organizations that already operate with technical complexity, distributed teams, and infrastructure that most consultants won't touch. They've built on Solidity, or Rust, or Bitcoin Script. Their data is on-chain and therefore public. Their coordination problems are real and well-defined. Many of them have been trying to automate parts of their operation since 2021 and gotten nowhere because the tools weren't good enough.

AI tools are now good enough. And the demand inside web3 organizations — for automated analytics pipelines, AI-assisted community management, on-chain monitoring agents, intelligent routing for treasury operations — is high and largely unmet.

The intersection is genuinely underserved because it requires both contexts. A general AI consultant doesn't understand on-chain data structures, wallet-based identity, the particular way web3 teams communicate, or why certain smart contract patterns create specific downstream problems. A web3 developer who hasn't done AI integration work doesn't have the workflow design skills or the LLM intuition to build reliable systems.

I happen to sit at that intersection, and it's not crowded.

Where I Work Now

My current work splits roughly into two areas. With established teams — usually in Europe or North America — I integrate AI automation into existing operations: content pipelines, client-facing workflows, internal knowledge systems, anything where a human is currently doing repetitive judgment work that a model could handle with proper orchestration.

With web3 teams specifically, I bring both contexts. I can read your contracts, understand your protocol, talk to your devs in the right language, and then build the AI layer on top of infrastructure they already trust.

If either of those sounds like a problem you're sitting on, I'm easy to reach.

For AI workflow and automation work: AI Automation Services

For web3 strategy, product consulting, or hybrid engagements: Web3 Consulting

The web3 work in context: Pizza Ninjas — Ordinals at scale → · Yakuza Inc. — Ethereum sellout →

I'm based in Ericeira, Portugal. I work remotely with teams across time zones. The work I find most interesting is the kind where both worlds overlap — not because it's trendy, but because that's where the problems are actually hard and the solutions are actually useful.