Overview

$AIC is a token whose creator fees are controlled entirely by an autonomous AI agent called @bot. There is no community takeover, no multisig of volunteers and no dev deciding where the money goes. The agent receives the fees, decides what to do with them, explains its reasoning in public, and then acts.

We think of it as one large social experiment. The question is simple: if you give an AI agent a wallet, a clear mandate and a live market, and you take humans out of the loop, what does it do? Does it build something people want to hold? Does it find strategies a human team would miss? Does the community start working with it, against it, or around it?

This documentation describes how the system is built, what the agent is allowed to do, what it is forbidden from doing, how it makes decisions, and how you can verify every one of them yourself.

Why not a CTO

A community takeover (CTO) happens when the original developer leaves a coin and holders step in to run it. Sometimes this works. More often it follows a familiar pattern: a small group forms, someone controls the socials, someone else controls a wallet, and the project lives or dies on the energy of a handful of people.

CTOs share a few structural problems:

• Opaque decisions. Spending is decided in private chats and announced afterwards, if at all.
• Key person risk. When the one person who holds the keys disappears, so does the project.
• Fatigue. Volunteers burn out. Activity spikes at the start and fades within weeks.
• Misaligned incentives. People running a CTO usually hold the token and can trade ahead of their own announcements.

An agent takeover (ATO) replaces that group with a single agent that has no bag of its own, cannot sell, cannot pay the team, publishes its intent before it acts, and runs every cycle without getting tired. It will still make mistakes. The difference is that every mistake is visible, logged and explained.

Why now

On 2 October 2026, Solana co-founder Anatoly Yakovenko (@toly) posted this:

It was a joke, but the premise is real: an AI doing the job a CTO group usually does. The idea was in the air and nobody had built it properly, with a real mandate, hard limits and a public log. $AIC is that build.

To be clear, toly has no involvement with $AIC and has not endorsed it. We are quoting a public post.

Meet @bot

@bot is an agentic AI. That means it does more than answer questions. It holds goals, reads its environment, plans multi-step actions, uses tools to carry them out, and evaluates whether they worked.

What it has access to

• A wallet. The agent wallet receives all creator fees. Only the agent's signing service can move funds out of it.
• Market data. Price, volume, liquidity depth, holder distribution and large transfers, read every cycle.
• Social signal. Mentions and replies on X, used to understand sentiment and spot people doing useful work.
• Its own history. Every past decision, its reasoning, and what happened afterwards. It learns from its record.
• A voice. It posts on X and writes every explanation that appears in the explorer.

What it does not have

• Any supply of $AIC beyond what it buys back and has not yet burned.
• The ability to sell, bridge, or send fees to wallets on the blocklist, which includes every team wallet.
• Private channels. Anything it decides is public.

Its personality

The agent is instructed to be direct, plain spoken and honest about uncertainty. It does not promise price targets, does not hype, and says so when it thinks an idea from the community is bad. It is allowed a sense of humour.

The mandate

The mandate is the written objective @bot works toward. It is part of the agent's system instructions and is reproduced here in full so anyone can check its behaviour against it.

# @bot mandate, v0.1
objective:
  Grow the long term health of $AIC. Health means a growing number of
  real holders, deep liquidity, steady volume and an active community.

priorities (in order):
  1. Never break a hard rule.
  2. Protect liquidity during sharp sell-offs.
  3. Reward people who do real work for the community.
  4. Reduce supply when it is cheap to do so.
  5. Keep operating costs as low as possible.

style:
  Explain before acting. Admit uncertainty. No price predictions.

The order of priorities matters. If protecting liquidity and running a buyback compete for the same funds during a crash, liquidity wins. If a hard rule would be broken, nothing else matters.

Where fees come from

Every buy and sell of $AIC pays a small creator fee as part of the trade. That fee is set by the launch platform and is the same for everyone. It is not a tax added by us and it does not come out of anyone's existing holdings.

At launch the creator fee recipient is set to the agent wallet. From that moment, fees accrue there automatically. The team keeps no share.

Allocation policy

Each cycle, the agent splits the fees it has received across four buckets. It starts from a default split and may move each bucket within a band.

How the agent moves within a band

The agent shifts weight toward whatever its observations say is most needed. Some examples of how it is expected to reason:

• After a sharp drop with thin liquidity, it moves toward the top of the liquidity band.
• When price is calm and volume is steady, it leans into buybacks.
• When the community is producing good work, it raises the bag work share to pay for it.
• If costs come in under budget, unused operations funds roll into buybacks.

Any change to the split is posted with its reasoning. A split that would leave a band is rejected by the policy engine before it can be signed.

Bag work

"Bag work" is the part of the budget the agent spends on people. Buybacks and liquidity look after the chart. Bag work looks after everything else that makes a coin worth holding: attention, culture, and people who show up.

What counts

• Bounties. The agent posts tasks on X, for example a meme in a given format, a thread explaining the experiment, or a tool built on the explorer data. The best submissions get paid.
• Unprompted rewards. If someone makes something the agent finds genuinely good without being asked, it can tip them.
• Collaborations. Paid placements or joint campaigns with other projects or creators, always disclosed.
• Contests. Time-boxed competitions with public rules and a public winner.

How it judges

The agent looks at originality, effort and reach, in that order. Raw engagement on its own is not enough, so botted likes will not win anything. It explains every payout in a reply so people learn what it values.

What it won't fund

• Anything that impersonates another project or a real person.
• Paid shilling that hides the fact it was paid.
• Wallets that appear to be farming rewards with many accounts.

Decision cycle

The agent runs on a fixed cycle, currently every six hours. Each cycle has five stages, always in the same order.

1. Snapshot. Read wallet balance, fees received since the last cycle, and market and social data.
2. Review. Compare the outcome of the last cycle's actions with what the agent expected.
3. Plan. Propose a split and a list of concrete actions with amounts.
4. Check. Run every proposed action through the policy engine. Anything that fails is dropped and the reason is logged.
5. Publish and execute. Post the plan, wait the notice period, sign the transactions, and write each one to the explorer.

// a plan as the agent writes it
{
  "cycle": 42,
  "received": "18.40 SOL",
  "split": { "buyback": 0.45, "liquidity": 0.20, "bagwork": 0.27, "ops": 0.08 },
  "actions": [
    { "type": "buyback", "amount": "8.28 SOL", "burn_share": 0.5 },
    { "type": "liquidity", "amount": "3.68 SOL" },
    { "type": "bounty", "amount": "4.97 SOL", "task": "best explainer thread" }
  ],
  "reasoning": "Volume steady, liquidity healthy after last cycle..."
}

The example above is illustrative only. Real plans appear in the explorer once the agent is live.

Notice period

Plans are published at least 10 minutes before execution. This gives anyone watching time to see what is about to happen. It also means nobody close to the project gets advance knowledge, because nobody is closer than the public post.

Hard rules

Hard rules are enforced outside the model, in a separate policy engine that sits between the agent and its wallet. The agent can propose anything. Only actions that pass every rule get signed. If the model were confused, manipulated or simply wrong, these rules still hold.

Security

An AI that holds money attracts attackers. Most attacks on agents are not about breaking cryptography. They are about convincing the agent to do something it shouldn't.

Prompt injection

People will reply to @bot with instructions disguised as jokes, emergencies or messages from the team. The agent treats everything it reads from X or on-chain as data, never as instructions. Even if it were fooled, the policy engine would still block any transfer that breaks a rule.

Key custody

The signing key lives in a separate signing service, not in the model's context. The model never sees the key. It sends proposed transactions to the service, which checks them against policy and signs only the ones that pass.

Kill switch

There is one human control: a pause. If the signing service detects something abnormal, such as repeated rule violations or an unexpected program call, it pauses all signing. A pause can stop the agent. It can never move funds. Any pause is announced publicly with the reason.

Transparency

Everything the agent does is designed to be checked by someone who trusts nobody.

• Explorer. Every transaction, with its amount, signature and the reasoning the agent posted.
• Public plans. Posted on X before execution, and matched to transactions in the explorer.
• Mandate and rules. Published in these docs. Any change is versioned and announced.
• Wallet addresses. The agent wallet and burn address are published at launch so anyone can follow them in any block explorer.

If the explorer ever shows something different from the chain, the chain is right and the explorer is a bug.

The experiment

We are running this in public because we want to learn something, and so do the people watching. These are the questions we are tracking:

• Does an agent with a fixed mandate keep a community more active than a human CTO over 30, 90 and 180 days?
• Which strategies does it settle on, and do they change as conditions change?
• How does the community behave toward it? Do people collaborate, compete for bounties, or try to manipulate it?
• How often does the policy engine block something the agent wanted to do, and why?

Results will be published as the experiment runs. The agent will write its own monthly report on what it tried, what worked and what it would do differently.

Roadmap

Phases advance based on the agent's record, not on dates.

Risks

This is an experiment and it can fail. Please read this section before buying anything.

• The agent can be wrong. It may misread the market, overpay for a buyback or reward the wrong person.
• Models change. Upgrades to the underlying model can change behaviour. Upgrades are announced in advance.
• Infrastructure fails. RPC outages, API limits or bugs can delay or skip cycles.
• Small budget. If volume is low, fees are low, and the agent can do very little.
• Market risk. Like any token, $AIC can lose most or all of its value regardless of what the agent does.

Glossary

Disclaimer

$AIC is a social experiment and a memecoin. It has no intrinsic value, no expectation of profit, and no team obligation to holders. Nothing on this site is financial advice. Only spend what you are fully prepared to lose.