Łukasz Miądowicz - AI, Growth & Platform PM

Why Smart Contracts Caught My Attention - Blockchain x AI Agents

Posted: 14 May, 2026 · ~14 min read

A PM's first-principles pathway into blockchain.

TL;DR
I'm an Technical PM who spent the last two years building agentic systems. A week ago I started learning Solidity on Cyfrin Updraft. This is the field report: what to learn, in what order, on which toolchain, and where the real opportunity is for PMs sitting at the AI × blockchain seam in 2026.

I've spent the last two years going deep on agentic AI. So what pulled me toward smart contracts?

The short answer: I'm an Technical PM, and AI products are about to need a trust layer that doesn't exist in Web2. The longer answer is what this post is about - but it's not really an AI post. It's an honest field report on what it looks like to onboard to blockchain as a PM in 2026: what to learn, in what order, on which tools, and what to skip.

If you're a PM who has been curious about blockchain but never found a non-cringe way in, this is the post I wish I'd had six months ago.

This is the field report. Settle in.

First principles: what does blockchain actually solve?

Most blockchain explanations start with "imagine a distributed ledger." That framing is technically correct and intuitively useless. Here's the framing that worked for me:

Blockchain is a database with a single weird property: nobody owns it, but everyone agrees on its state.

Every other database in the world has a privileged actor - the server admin, the DBA, the company that hosts it - who can change history. Blockchain replaces that privileged actor with a consensus mechanism. The tradeoff is brutal: you give up speed, storage cost, flexibility. What you get is programmable trust.

That's the whole magic. Everything else is implementation detail.

Once you accept that as the core feature, a lot of follow-on questions answer themselves. Why is gas expensive? Because consensus is expensive. Why are smart contracts immutable by default? Because the whole point is removing privileged actors who can change them later. Why does decentralized finance look so weird compared to a Stripe dashboard? Because every interaction is a public, replayable, signed message - there's no backend hiding the mess.

The PM question becomes: is the cost of programmable trust worth it for this product? For 90% of consumer apps, no. For agent infrastructure, payments between mutually-distrustful parties, audit trails that need to survive their issuer, or financial products that can't have a single point of control - yes.

The mental model: seven primitives every PM should know

You don't need to write Solidity to PM blockchain products. You need a working mental model of seven primitives. Here's mine, with the PM lens for each.

1 · Wallet
A keypair. The private key signs transactions; the public address receives them. No "account recovery" by default - lose the key, lose the funds.

PM lens: every flow either makes the user manage a key or abstracts it away (custody, social recovery, passkey-based smart accounts). No third option. The choice you make here shapes everything else about your UX.
2 · Transaction
A signed message asking the chain to do something - send tokens, call a contract, deploy code. Has a sender, a recipient, a value, calldata, and a gas budget.

PM lens: every user action has a cost, a latency, and a failure mode you cannot fully hide. Build your UX around the assumption that some percentage of transactions will fail mid-flight. The status page for your product is the chain.
3 · Smart contract
Code deployed to the chain. Deterministic, public state, immutable by default. Anyone can call its functions; everyone sees the result.

PM lens: ship the contract wrong, ship a public bug that everyone can exploit forever. There is no hotfix. Auditing is not optional. The mindset shift from "iterate fast and patch" to "ship deliberately and rarely" is the single biggest cultural change for Web2 PMs entering this space.
4 · Token
A smart contract that tracks balances. ERC-20 is the fungible standard (USDC, ETH-wrapped, etc.). ERC-721 is the non-fungible standard (NFTs). ERC-1155 mixes both.

PM lens: tokens are standardized state - they let you compose ownership across protocols you don't control. Your contract doesn't need to know about Uniswap; Uniswap knows about your ERC-20 because it follows the standard. This is the closest thing crypto has to a USB-C moment.
5 · Oracle
A bridge that brings off-chain data on-chain - prices, weather, sports results, API responses. Without oracles, smart contracts can only know what's already on-chain.

PM lens: oracles are where most "blockchain solves X" pitches quietly fall apart. You don't escape trust; you concentrate it in the oracle. Chainlink, Pyth, RedStone are the major ones. Their failure modes are interesting reading and inform what's actually buildable.
6 · Gas
The fee paid for execution. Denominated in the chain's native token (ETH on Ethereum). Set by supply and demand in real time.

PM lens: gas is your latency × cost knob. It's why Layer 2s (Base, Arbitrum, Optimism) exist - to make execution cheap enough for products with frequent small transactions. Pick your chain by mapping your transaction frequency and value distribution against gas economics. A high-frequency app on mainnet Ethereum is a non-starter.
7 · Consensus
How the network agrees on what happened. Proof-of-work (Bitcoin) and proof-of-stake (Ethereum) are the two production-scale mechanisms. Each chain has its own security/speed/cost curve.

PM lens: you usually don't pick a consensus mechanism, you pick a chain - and the chain bundles consensus + ecosystem + tooling + community. Pick the chain that matches your product's profile, not the one your friends like.

If you can fluently explain those seven to a non-technical PM in 30 seconds each, you have enough mental model to start having useful conversations with engineers. Everything else is depth.

Learning path

When I started a week ago, I tried four resources in parallel: a CryptoZombies tutorial, a Solidity by Example walkthrough, the OpenZeppelin docs, and Cyfrin Updraft. Three I abandoned. Cyfrin is the one I finished and still go back to.

Why Cyfrin won: it's built around production craft, not pedagogy. Patrick Collins (the co-founder) was the lead Solidity educator on FreeCodeCamp for years. Updraft is what he wishes he could have made there - graded exercises, real projects, audit-style writeups. The security course in particular is brilliant: it teaches you how contracts get hacked, which is the only way to learn to write them correctly. You watch real exploits replayed, read post-mortems of $100M+ hacks, and write your own exploits as homework.

Here's the week-by-week pathway I'd recommend:

Week 1–2 · Foundations
Cyfrin Updraft Blockchain Basics. Non-technical. Get the seven primitives above into your head. Set up MetaMask. Fund a testnet wallet from a faucet. Send your first transaction. The mental model needs to be solid before the syntax matters.
Week 3–6 · Solidity
Cyfrin Updraft Solidity Smart Contract Development. Don't watch - build the example projects yourself, line by line. Deploy each to Sepolia testnet. The moment your first contract verifies on Etherscan, blockchain stops being abstract and starts being a thing you do.
Week 7–10 · Toolchain
Pick a framework and ship a real project. Both Hardhat and Foundry are valid. Full comparison in the next section. Build something you'd be willing to demo in an interview - an ERC-20 with vesting, a multi-sig wallet, a simple lending pool, whatever fits your interests.
Week 11+ · Pick a depth
Three tracks: security (CodeHawks competitions, Cyfrin's audit courses), DeFi (read Uniswap, then Aave, then GMX as a progression), or protocol PM (study one protocol's docs end-to-end, then a second, look for the patterns). I picked the third because it's closest to my comparative advantage.

I cut above to most important topics and went through them within a week (still need to finish the learnings)

The toolchain choice: Hardhat vs Foundry

This is the section that took me longest to figure out because the internet's advice is contradictory and tribal. Here's the honest comparison after building with each.

Foundry
By Paradigm. Written in Rust. Tests in Solidity, not TS. Built-in fuzzing and invariant testing.
Hardhat 3
By Nomic Foundation. TypeScript ecosystem. Huge plugin library. Tests in both Solidity & TS (useful for end to end tests). Hardhat Ignition for declarative deployments. Best for full-stack projects tightly coupled to a TS frontend.

Hardhat's strengths are real:

I built a simple use cases in Hardhat first, on purpose - a basic ERC-20 with vesting. Toy version.

Three reasons I chose Hardhat as the starting point:

  1. Familiarity reduces friction. TypeScript is language I'm shifting towards. Starting in TS meant I could focus on Solidity semantics without also fighting unfamiliar Rust-flavored tooling. Lower friction = more iterations = more learning per hour.
  2. Ecosystem depth. Every "how do I do X" question I hit during the toy projects had a Hardhat tutorial. Foundry's docs are excellent but the long tail of community tutorials is smaller. For a beginner, the long tail matters more than the official docs.
  3. Personal curiosity.

The point of building a toy wasn't the products. It was internalizing the dev loop end-to-end: write contract → compile → test → deploy → verify on Etherscan → call from a frontend. Twenty hours of doing that for real teaches you more than two hundred hours of reading about it. There's something about seeing your own contract on a public block explorer that locks the mental model in place.

My recommendation by background:

What to learn and what to skip

There's a lot of noise in this space. Here's the filter I wish I'd had on day one.

Learn
Solidity basics. The EVM mental model. Account abstraction (ERC-4337) and token-bound accounts (ERC-6551). Reading audit reports. Oracles and their failure modes. Stablecoin payment flows. ZK fundamentals (not implementation). Both Hardhat and Foundry.
Skip
Memecoin meta. NFT collectibles (unless that's your domain). Most "tokenomics" content. L1 chain wars. Anything that sounds like "this technology will revolutionize voting / healthcare / supply chains" without a working product behind it. X blockchain influencers.

Three resources beyond Cyfrin worth your time:

Where this gets interesting: blockchain × agentic AI

I said this wasn't an AI post, and the body hasn't been. But I'd cheat you if I closed without saying what's actually pulling me toward blockchain - because it's not the technology in isolation. It's the seam.

I love merging technologies. The most interesting product opportunities live where the constraints of one stack are dissolved by the affordances of another. With AI and blockchain, the seam is fertile in a way nobody is articulating well yet.

The pattern: AI agents need infrastructure built for non-human actors, and blockchain (in my opinion) is the only infrastructure that natively serves non-humans.

Today's APIs assume a human user behind every request. Today's payment rails assume a human cardholder. Today's identity systems assume a human face for KYC. Agentic AI breaks all of those assumptions, sometimes catastrophically. Blockchain is what you'd design from scratch if your starting point was "the user might be a script."

Four use cases I'm watching and partly building toward:

1 · Agent-owned smart wallets
ERC-4337 smart accounts plus ERC-6551 token-bound accounts let you give an agent its own wallet with policy enforced by the contract, not by the prompt. The agent can transact, but only within a budget. It can sign, but only of approved transaction types. The principal can revoke at any time. Critically: the spending rule is unbreakable by the agent because it lives in the contract, not in instructions the agent can ignore.
2 · Agent-to-agent commerce
Two agents that don't trust each other negotiate a deal and settle on-chain. The contract enforces the outcome - payment on delivery, escrow with arbitration, streaming payment that stops if SLA breaks. No platform takes a cut, no human approves each transaction. Logistics, content licensing, compute-for-data trades, even ad inventory - anywhere two agents could transact on behalf of their principals.
3 · Verifiable AI inference
Today, if a chatbot says "I ran model X on input Y and got result Z," you have to trust the operator. Zero-knowledge proofs of inference and trusted execution environments (TEEs) anchored to a public ledger make it cryptographically verifiable instead. This matters for high-stakes domains - insurance claims, hiring screens, medical triage - where "the AI said so" needs to become "the AI proved so." The infrastructure exists; the products mostly don't yet.
4 · Agent identity and reputation
If agents transact at scale, you need to know which agents are credentialed, which are malicious, which are deprecated. On-chain reputation with stake-based skin-in-the-game ("this agent has 100 ETH staked against good behavior") is inevitable. The first protocol that nails this becomes the LinkedIn-for-agents layer - and that's a substantial business, because every agent platform will need to either integrate it or rebuild it.

None of this requires new science. All of it requires infrastructure that exists in 2026 but is being assembled by tiny teams with little PM presence.

This is what I mean by merging technologies. Solidity on its own is interesting if you're a protocol engineer. Agentic AI on its own is interesting if you're an applied AI PM. But the cross-product - agents that can hold, commit, prove, and earn on-chain - is a 10× larger surface than either stack alone. And almost nobody is positioned to PM it well, because almost nobody is fluent in both.

That's the bet.

Closing: what to do this week

If you've read this far, you're at least curious. Three actions, ordered by effort:

  1. Spend 90 minutes on Cyfrin Updraft's Blockchain Basics. Free, low commitment. Sets the mental model.
  2. Open a MetaMask wallet, fund it with $20 of ETH on a Layer 2 (Base or Arbitrum), and complete one real transaction. Buy a stablecoin. Send it to yourself.
  3. Pick one toolchain and ship a toy contract. Hardhat if you're TS-comfortable, Foundry if you're security-curious. Deploy a tiny ERC-20 to Sepolia testnet. Verify it on Etherscan. Take a screenshot.

If you do those three things, you'll know in a weekend whether blockchain belongs in your career bet.

For me, the answer was yes - not because I'm leaving AI, but because the next generation of AI products will live on-chain whether their builders intend it or not.


Tagged: #Blockchain #Solidity #Cyfrin #Hardhat #Foundry #ProductManagement #AgenticAI #PMCareer #Web3 #SmartContracts

View original

#AI-Native PM #Account Abstraction #Agentic AI #Blockchain #Cyfrin #DeFi #Ethereum #Foundry #Hardhat #Product Management #Security #Smart Accounts #Smart Contracts #Solidity #Web3