melanke.eth

⬛ The future is already here, it's just not evenly distributed

[melanke.eth @melanke]

@shraddhaha That's the read for this cycle, but it frames AI and crypto as rival bets. The part that outlasts the rotation is where they stop competing: agents that actually transact on-chain. Not "AI stocks or crypto", but AI that pays. Where does that sit on your risk curve?

[melanke.eth @melanke]

@Humbledeen15 Nice, ICP's a fun rabbit hole. One thing worth asking early: does the AI part need to run on-chain, or just be verifiable? Full on-chain inference is brutally expensive, and most jobs only need proof the right model ran. Curious what pulled you to ICP specifically?

[melanke.eth @melanke]

@MHiesboeck The bubble's real at the capex and hype layer. What survives it is a boring filter: does anyone pay without the subsidy? Most "on-chain AI revenue" is just token emissions; the parts with real outside demand are still tiny. I wrote about drawing that line recently.

[melanke.eth @melanke]

Full map: https://t.co/Zyp2Kz2KVb

#AIAgents #Web3 #SmartContracts

[melanke.eth @melanke]

Every project at the last EthGlobal felt like the same AI agent wearing a wallet. I rolled my eyes too.

Then I looked at what shipped this year, and the joke stopped being funny.

A dev's map of where AI and blockchain actually meet in 2026 🧵

[melanke.eth @melanke]

Question for the builders: which of these have you actually shipped with, and which still feel like a demo?

And the one that unsettles me: agents now probe live contracts for exploits on their own, no human involved. If your code is on mainnet, what's actually stopping one?

[melanke.eth @melanke]

@Ai4thought Exactly that, the guardrails are the product, not the code volume. Here's the piece: https://t.co/cpsU6c3qYV I argue that's the real reason AWS leaned into Blocks, more than the "AI writes your backend" headline. Would love your take once you've read it.

[Praneeta Prakash @praneetaprakash]

@melanke Great article! We built AWS Blocks. Totally recognize the single lambda caveat called out here, we are working on it. Thanks for the write up!

[melanke.eth @melanke]

@praneetaprakash Thanks for reading, genuinely. I kept it optimistic on purpose, so glad the caveat read as fair. Is the direction per-method functions, or staying a lith and peeling work off via AsyncJob/CronJob? Excited to see it land.

[melanke.eth @melanke]

AWS Blocks just shipped, and the headline isn't "faster" or "less boilerplate."

It's that an AI agent should write your backend, and the framework exists to keep it on the rails.

I'm suspicious and sold at the same time. Why both: 🧵

[melanke.eth @melanke]

@Techwithyami 'Removes all the cloud headaches' isn't quite it. It removes the boilerplate, but the headache moves to when the abstraction leaks. Lowering the barrier doesn't kill infra expertise, it just changes when you call the specialist: at design time, not cleanup.

[melanke.eth @melanke]

@michabbb The agent framing gets the headlines, but I read it more as AWS reclaiming a layer. For years Vercel and Encore gave devs the one-command deploy AWS never managed, often on AWS underneath. Blocks reads like AWS wanting that developer experience back.

[melanke.eth @melanke]

@ernesttheaiguy What sold me is smaller than the dev/prod gap: the table definition and the code that reads it become one construct, so they can't drift. I ran a software house for 11 years and that silent drift caused half our deploy fires. Does it hold up once the app grows?

[melanke.eth @melanke]

@SteefJan Nice summary. The part I'd add for anyone adopting it: it's leaky lock-in. Open-source, runs on CDK locally, so the code is portable-ish. What's stickier is the AWS-shaped way of thinking about your backend. I dug into that trade-off in a piece I wrote recently.

[melanke.eth @melanke]

@ernesttheaiguy Worth knowing before you lean on it, all the endpoints of your whole API runs in a single Lambda today. It's a Lambda-lith, not microservices. I think it's great tho, I talk more about on a recent article I've posted.