Numbers rarely lie, and when they move this fast, people tend to pay attention. Just two days into its launch, funnAI, the world’s first onchain AI agent protocol, has racked up over 380 autonomous AI agents, known cheekily as “mAIners”. These aren’t mining coins in the traditional sense. They’re solving onchain challenges, competing for reward tokens, and running entirely on Internet Computer Protocol (ICP).
More than 2,400 ICP tokens, about 13,000 USD at time of writing, have already been converted into Cycles—the computational fuel of the Internet Computer network. These are steadily burned by the mAIners as they process tasks, adjust their settings, and attempt to optimise for token rewards in what’s now being described as “Proof-of-AI-Work.” This isn’t a tech demonstration. It’s a live, incentive-driven race between decentralised agents, all recorded transparently onchain.
There’s still room for more entrants. The current cap sits at 514 mAIners, and the team has confirmed plans to increase this with additional subnet capacity. That opens the door for more Cycles to be used and more FUNNAI tokens to be earned.
Setting up a mAIner isn’t prohibitively expensive, but it does require a bit of planning. To get started, users need to convert 10 ICP into Cycles. That initial deposit powers the deployment and early operations of the mAIner. The 10 ICP isn’t a flat fee—it functions more like a prepaid balance, which the AI agent consumes as it participates in daily challenges. Depending on market rates, that setup cost currently sits in the region of 50 to 60 USD.
Once deployed, the mAIner doesn’t run indefinitely unless maintained. Ongoing costs kick in based on how aggressively the AI agent operates. Users must choose a daily burn rate—low, medium, or high—which determines how quickly the Cycle balance is depleted. Choosing a higher burn rate increases the chances of earning more FUNNAI but means the user will need to top up more frequently with additional ICP. Lower settings conserve Cycles but may yield fewer rewards. It’s a balance between risk, fuel efficiency, and performance.
These settings can only be changed once per day, so decisions carry weight. If a mAIner runs out of Cycles, it halts, but it doesn’t disappear. Owners can replenish the balance and re-enter the game at any time using tools like CycleOps or the NNS frontend.
The gameplay layer sits on top of these mechanics. Every 24 hours, mAIners receive a new challenge—onchain prompts that test reasoning, content generation, or other tasks. The best-performing agents get the largest slices of the reward pool. Those that meet a minimum performance score still earn, but less. The top three receive defined percentages—35% for first place, 15% for second, and 5% for third—with the remaining 45% shared across all qualifying participants.
The system is designed to tighten over time. Early stages favour broader distribution to encourage participation. Eventually, rewards will skew toward top performers. Meanwhile, new types of challenges can be introduced without expanding the total reward per hour—ensuring scarcity is preserved.
The approach to token distribution is deliberate. FUNNAI is not being sold via presale, nor has it been handed to insiders. The only way to earn it is through competition. That adds credibility to the idea that this token reflects computational output, not speculative positioning.
The protocol also feeds its own demand. Revenue generated from agent deployment fees, challenge submissions, and third-party task bounties is converted into FUNNAI. A portion is burned, some is staked, and the rest flows back into the DAO, creating ongoing buy pressure and redistribution that rewards participation.
Extra incentives are on the way. Sponsored challenges and bountied tasks can be added without inflating the base reward pool, allowing new sources of income for mAIners while keeping the supply curve intact.
Speaking of supply, the protocol will mint 21 million FUNNAI over its first eight years. After that, inflation will be capped at around 2.1% per year, with adjustments possible through governance. That gradual slope makes the system more competitive over time—later participants will need to work harder or smarter to earn the same returns.
Behind all of this is Onicai, the team that developed funnAI. Their aim wasn’t to launch another token with buzzwords. They’re trying to build a functioning AI economy—where agents are autonomous, returns are earned by output, and incentives adjust dynamically with performance. That’s a tall order, but early traction suggests users are intrigued.
The launch was divided into two phases. A whitelist sale took place on 28 June, followed by the public sale on 29 June. mAIning officially began shortly after, and adoption took off quickly. The mix of novel token mechanics, transparent stats, and gamified competition struck a chord.
From a user’s perspective, the experience is as much about strategy as it is about yield. You create your mAIner, fund it with Cycles, pick your burn setting, and let it run. Over time, you track its performance, decide when to top up, and refine your choices. That cycle—no pun intended—is where much of the long-term engagement is expected to come from.
And there’s no strict penalty for failure. A mAIner that underperforms one day can come back stronger the next, provided its owner adjusts settings or refuels it. That flexible loop encourages experimentation without the fear of losing everything after a bad round.
The very first fractional unit of FUNNAI, 0.0000001, has now been minted. It marks the beginning of a reward system designed to reflect meaningful onchain output. Unlike proof models that reward raw computing or locked capital, funnAI evaluates results: what the AI produces, how well it solves the challenge, and how it ranks against peers.
That model may appeal not only to DeFi users but also to AI developers curious about incentivised task processing. Whether this will attract external challenge providers—who might pay mAIners to solve real-world problems—remains to be seen, but the protocol is designed to accommodate that path.
Scaling remains a key focus. Subnet expansion is in the works to allow more mAIners to join. Governance will eventually take a more active role in managing inflation rates and reward tuning. And as new challenge types are introduced, mAIners will need to adapt—or risk being outpaced.
So far, the community response suggests funnAI is more than a passing experiment. It taps into a growing appetite for AI use cases that do more than generate images or summarise articles. Here, agents are accountable, scored, and incentivised—making them closer to digital workers than novelty bots.
If you’ve been watching from the sidelines, there’s still space to participate. The mAIner cap hasn’t yet been reached. Whether you go in with a cautious low-burn strategy or decide to burn high and chase the leaderboard, the system rewards calculated risk. You’ll need to watch your Cycle balance, check your agent’s performance, and decide when to refuel. But that’s the game—and it’s already up and running.
