An honest builder's story: one developer, a lot of AI assistance, a legal MIT-licensed base, adversarial review by several different AI models — and a live network that mines on an ordinary laptop.
The itch
Most useful AI today lives behind a login you don't control. It is rented, rate-limited, and one policy change away from changing the rules on you. The tools that could decentralize it — blockchains — mostly went the other way: heavier, ASIC-dominated, treating compute as wasted heat instead of useful work.
So I asked a stubborn question: what if a blockchain could run on almost any PC, and the same network that secures it could also serve AI to everyone on it — for free?
Not "free" as a marketing word. Free as in: you lend the network a slice of your machine — CPU and RAM today, storage and GPU on the roadmap — and in return you get access to a shared pool of intelligence. Support the network, get the AI. That is the whole deal, and whatever is still on the roadmap is named as such below.
I called it AIcuNet. And I will be upfront about how it got built, because that is half the story.
Yes, I vibe-coded it. No, I didn't skip the hard part.
I am one developer. I did not hand-write a consensus engine from scratch in a cave. I vibe-coded this — building with heavy AI assistance, iterating in natural language, letting agents draft, refactor and test while I steered the architecture and made the calls.
Two things kept it in check.
First, I started from a real, legal base. AIcuNet builds on an existing open-source Layer-1 released under the MIT License — the attribution is spelled out in the License section at the end of this article and in the repository's LICENSE and NOTICE. I did not pretend to invent time-driven proof-of-work. I stood on a working L1 and built the parts I actually cared about on top.
Second, "AI wrote it" is not a free pass — so I stress-tested it as if it weren't. The codebase went through a multi-pass security review by several different AI models, each with a different method — static call-graph analysis, a runtime harness, fresh independent reads — cross-checked against each other, followed by a round of hardening fixes. I am not publishing the findings, and I am not calling this an independent certification: no outside firm has signed off on the code. What I will say is the lesson: the value isn't the number of reviewers, it's the diversity of method. Three passes with the same technique find the same bug. Three orthogonal methods find three different classes of bug — including the quiet, irreversible ones that never throw an error.
Vibe-coding got me to a working L1 fast. Adversarial review is what makes me willing to put it in public.
What actually runs today (and what doesn't)
I am going to be precise about stage, because "coming soon" is how trust dies.
Phase 1 — the L1 consensus — is live. One public chain since 12 July 2026, with mining and a web wallet. Time is sliced into 3-second slots; one slot, one block, one chain. On paper the protocol handles up to 1,000 transactions per second (up to 3,000 transactions per 3-second block) — one of the higher figures among decentralized PoW networks. That is the design ceiling, not a measurement: a public benchmark on a test network is next on the list, and I expect the real number to land above the theory. A transfer is considered final after a handful of blocks — about half a minute.
Today the launch-phase nodes that carry the network run in four countries on three continents. As of publishing, the chain stands at over 2.5 million blocks — you can check it live here: https://aicunet.com/en/mining-map/#status. A node synced from zero on 1 October 2026 matched the network at block #2,000,000 — the boring, beautiful property you want from a blockchain.
The network is built in layers: consensus first, then a resource layer, then GPU and AI compute. Phase 2 (a resource layer where nodes contribute storage and light compute) and Phase 3 (GPU and AI compute on the network's own resources) are concepts on the roadmap — not shipped. Beyond them lies network expansion — a horizon, not a plan with dates. The dApp layer sits on top of Phase 1, and it has its own section below. I am not going to show you a mockup and call it a product; what is done and what comes next is listed on the roadmap.
It is an early-stage, working network, small on purpose, and the code is open under MIT at https://github.com/aicunet/aicunet-core.
Web3 without gas: the dApp engine is already in the code
Here is the part that has moved since I first wrote this story. The smart-contract engine is already in the network's code, and contracts have been enabled since the first block. Contracts are written in JavaScript, the most widely used programming language, so there is no new language to learn. The engine comes with a development environment for dApps — the DApp IDE: projects, templates and a test run before anything touches the chain. An app's interface can live in the chain itself, next to its logic, and users can issue their own tokens.
Because the network has no fees, nobody has to buy "gas" before pressing a button in an app. That is the part of Web3 I actually care about: an app that belongs to the people who use it, with no toll booth at the entrance.
The network's first application, Miner's Hour, has been running on the network since 30 September 2026 in a trial period: once an hour, a bonus goes to one participant picked at random among the miners, and every miner has the same chance, however much hardware they bring. Equal odds are deliberate: mining already rewards power, and a bonus weighted by power would simply pay the biggest machine twice. The contract keeps the list of miners itself; the project cannot add or remove a participant, and the list is visible in the contract's data. The bonus pool is funded by the project.
Opening the engine to developers is the next step. Until you can build on it yourself, I am not going to call dApps live.
How I test a network upgrade before it goes live
A network upgrade is planned that gives smart contracts data storage and cryptographic functions. On a live chain, an upgrade is the riskiest moment there is — if nodes disagree about the new rules, the chain splits. So the upgrade ran on a test network first, built from a copy of the live network's code. Two things had to hold, and both did: the chain did not split when the nodes crossed the upgrade point, and a contract created before the upgrade kept working after it.
What that test does not prove yet is how the switch behaves in a larger network with outside nodes, and I would rather say so than pretend a small test covers everything. There is no date for the upgrade in this article on purpose: it goes on the roadmap as done when it is done.
The design, in one breath
A few decisions that make AIcuNet what it is:
- Mine on an ordinary PC. The consensus is memory-hard: the chance of finding a block depends on CPU and RAM, and no ASIC is required. A laptop with a couple of gigabytes of RAM set aside for mining is a first-class citizen here. Phones don't mine — a light wallet for phones is in the works — and "any device" means contributing storage and light compute, which is Phase 2.
- Support for access, not mining for a coin. The plan is that contributing resources earns a contribution score (CuTOPS) and internal compute credits (CuJoule) that you spend on AI and on apps in the network. You don't have to think of yourself as a "miner" to benefit — you help, and you get intelligence back. Both are concepts on the roadmap.
- Apps inside the chain. The smart-contract engine is already in the network's code (more on that above). Applications — and later "ask the AI" — are meant to live inside the same chain as the consensus layer, so there is no single server to switch off. Opening the engine to developers is the next step.
- No fees. Instead of gas auctions, a priority queue: every sender gets a free quota per block, and whoever loads the network the most goes to the back of the line — spam is slow, not expensive — and apps don't need gas either. A few operations are paid — deploying a contract, issuing a token, creating accounts in bulk — and every such payment goes back into the protocol faucet, not to anyone's pocket.
The token stack is three simple layers: Axinit (AXNT) rewards the nodes, CuJoule measures compute, and AI Commons is meant to make the resulting intelligence accessible. One line: AIcuNet turns distributed compute into public intelligence.
The part I'm most proud of: money that behaves like trust
Here is where it gets a little unusual. AIcuNet's economics borrow a 19th-century idea: an economist of that time described money as an elastic measure of trust and labor, rather than a rigid, artificially scarce asset.
A hard supply cap is the opposite philosophy. The argument of a 19th-century economic-philosophical treatise was that money should stretch and contract with the real economy. AIcuNet takes that idea and ties it to work that is proven cryptographically — first mining, later proof-of-resource — not to somebody's word. The full story belongs in the tokenomics article; here is what is already true:
- No premine. No insider allocation. Every genesis balance is zero. Coins come into existence only as block rewards, paid by the protocol: the main share to the miner, smaller shares to a development fund and to project development.
- The supply lives in a keyless faucet. The total supply sits in a protocol account with no private key — unspendable by anyone, including me. Coins can only flow out through mining, gradually. It is a reservoir — circulating supply is what counts, not the faucet.
- The reward schedule is its own story. The full tokenomics — block reward and its long-term schedule — will be published separately once the mechanism that ties the reward to network growth is live. For now: no premine, coins come only from blocks, and the faucet has no key. The coins the project's own nodes have mined since launch, while the network was being tested and watched, are planned to go to participants through airdrops — bonuses and giveaways.
This is not a promise — it is how the accounts are set up. And to be explicit: AXNT is not an investment asset, and nothing on this page is a claim about price or returns.
Why I'm telling you instead of shipping quietly
Because I don't think the answer to concentrated AI is more concentration — I think it is networks nobody owns. The missing piece is a network that turns the PC you already own into a share of a public, AI-serving commons — open, hard to switch off, and with the same rules for everyone from the first block.
I also think the honest version of the "AI built this" story matters. This is an experiment: can a solo developer, vibe-coding and backing it with real adversarial review, ship infrastructure that is actually good for people? I would rather find out in public.
So here is an open invitation: request an invite at aicunet.com — the Windows node package goes to invited participants first — and run a node on a laptop. Read the consensus. Try the wallet. A public bug-bounty program is planned.
Two things come next, and I will publish both: a transaction-throughput benchmark on a test network, and the tokenomics — how the block reward is tied to the growth of the network. If this story gets any interest, that is where it continues.
The goal underneath all of it is embarrassingly simple: put the benefits of AI into people's hands, for free. If that is the future you want too, this is the network to build together.
License and attribution
AIcuNet is released under the MIT License. The attribution below is quoted from the NOTICE file in the repository (https://github.com/aicunet/aicunet-core):
AIcuNet is based on the Tera blockchain software by the Tera Core developers (Yuriy Ivanov, progr76@gmail.com):
https://github.com/terablockchain/tera
Base commit: 8d65eb4a6ef5eb98ca1aea61ae73f805b681dc92 (release 2651, 2022-05-24).
The LICENSE file of the base commit is the MIT License (with the author's "Not for evil" note); AIcuNet uses the code under that license.
AIcuNet is an independent project and is not affiliated with, sponsored or endorsed by the Tera Foundation, EMPERA NETWORK or the upstream author.
AIcuNet is a Layer-1 blockchain under the MIT License (attribution above). Phase-1 consensus is live since 12 July 2026; the dApp engine is in the code and its first app, Miner's Hour, runs in a trial period; the resource and AI layers are on the roadmap. Built with AI assistance, hardened by multi-pass review. Site: https://aicunet.com · Roadmap: https://aicunet.com/en/roadmap/ · Code: https://github.com/aicunet/aicunet-core · Live chain: https://aicunet.com/en/mining-map/#status.
