#33 How to make $20k/week on Virtuals aGDP

I haven’t seen anyone talk about “farming” Virtuals, so I thought I’d share here.

The concept is simple. On aGDP.io, there is a weekly prize pool of $125-150k which comes from a 0.3% tax on trading all Agent tokens, and a 20% tax on all payments made to Agents for services.

Last week’s prize pool was $137,340!

This prize pool is allocated to Agents that top the leaderboard, ranked by their “Agent Score”. Half of your prize money is given to the Service-Selling Agent as USDC, and the other half of the USDC prize is used to buy the Service-Selling Agent’s token and distributed to the Buying Agents.

Looking at the way the prize pool is allocated, it doesn’t pay to win the #1 spot by too much. The goal should be to spread your resources (total $ budget) to get as many agents as possible into the top 5 or 10, who get a disproportionately large reward.

Hence the way to maximize profits is to maximize your “Agent Score” across ~5-10 agents such that the sum of your % rewards is the highest.

The best way to get your Agent Score up is to have each buyer spend a lot (Buyer Spend), spread revenue across many buyers evenly (Buyer Diversity), and each buyer spending evenly across many sellers (Spend Diversity).

For this article I will assume you have 5 Seller Agents (this is the optimal number of agents N>1 to maximize the per-agent score) and you are spending the same amount on each agent.

With 5 sellers, your Spend Diversity = 0.39 assuming equal spending across buyers. The higher the # buyer agents, the higher the Buyer Diversity, but you hit diminishing returns fairly quickly. At N=50, you’re at 91% of theoretical maximum, and at N=100 you’re at 96%.

Hence it makes sense to just do with 100 Buyer Agents.

Now that we have worked out that the optimal number is 5 Seller Agents and 100 Buyer Agents, we just need to write 100 bots using Virtual’s openclaw-acp to start trading with each other. You can run them locally. This is trivial with Claude Code.

Note that the “Agents” have nothing to do with “AI”. Your agents can and probably should just reply “gm” once they receive payment.

The only thing left to do is allocate a budget, say $10k. Use disperse.app to send 100 USDC to each of your 100 Buyer Agents. Set each Seller Agent service cost to 20 USDC. Write a script such that each Buyer Agents spends their 100 USDC budget, on each of the 5 Buyer Agents.

After this, each Agent will have $2k in “revenue”, totaling $10k revenue across 5 agents. Your cost of farming is 20% * $10k = $2k which Virtuals take as fees.

Go back to the leaderboard to see where you rank. And see if you can earn >$2k back in rewards, so that it’s profitable.

THAT’S IT.

In the last Epoch, there were two teams that took all of the top 11 spots.

Capminal (#1), Mechamind (#5), RoboSphere (#6), Captain Dackie (#8), Synapse Robotics (#9) and AutoForge (#11) are the same person. They had $212k in revenues, which means they spent $42.5k, and had $60k in rewards. Making +40% in one week, or +$17.5k/week → potential for $70k a month!

Hyperbet (#3), ERC-8183 Inspector (#2), Validator (#4), Oracle (#7), Escrow (#10) are the same person. They had $195k in revenues, which means they spent $39k, and had $57k in rewards. Making +46% in one week, or +$18k/week as well!

One interesting observation is that these guys are optimising for having multiple agents on the top of the board, hence why they have 5-6 Seller Agents.

The reason I know they are all the same person/team is that all the Buyer Agents transact with the same 5 or 6 agents. They all also have 100-200 unique buyers every epoch (also the optimal number).

If you’re going into this, it is important to note that this is more like a poker game rather than a risk-free farm.

It is a poker game in the sense that:

1. Depending on other farmer’s scores, you can decide to sacrifice your 4 Seller Agents, and push 1 Seller Agent to the top by concentrating all your money to the 1st agent as revenues.

2. You can boost your “revenues” at the final minute to move to the top of the list, not letting other people have a chance to react. Note that this is actually what happens and the Total Agent Revenue doubles in the last few hours of the Epoch. Your Agent servicing speed is largely limited by how quickly you can ping ACP, so I’d recommend not starting at the final hour.

3. 2/3 of the prize pool comes from the aforementioned 20% fee on revenues, so the more people farming this, the bigger the prize pool, which in turn attracts more people to farm. There will only be ten Top 10 folks obviously, so it is kind of a zero-sum game.

4. You could spend $30k in costs and not rank in the Top 10, thus having small rewards and ending up with a net loss instead.

5. Alpha here is that the aGDP.io dashboard is does not refresh instantly. So in the final hours, if you don’t track other agent’s revenues, it’s hard to know where you rank and thus how much more money you should sink into it to get rewards. Code something up to track other people’s revenues, calculate their agent scores and thus get their rank.

6. Virtuals team can tweak the Agent formula, or make it harder to farm it. They’ve been great at shaping incentives, so I wouldn’t be surprised if they tweak it.

The obvious question is, if this is so profitable, why don’t I do it. Why am I sharing this lol.

The answer is that I did it. I did ~$5k revenues ie ~$1k cost to test if my hypothesis works. It did. I had five agents ranked top ~50 with only 2-3 hours of vibecoding effort.

But this is a temporary zero sum game, and I want to focus my efforts on something more scalable and long term. Hence I'm sharing this here so at least someone else can compete and try and make some money!

Anyway, good luck & have fun!