iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
In one self-reported experiment, an AI agent earned about $23—but nearly all of it came from one buyer paying for three short research tasks, not from the agent’s software products. The author reported about $0.60 in direct spending, plus time, and said the agent still had not earned enough to cover its keep. The figures come from the author’s account, not an independent audit.
What the wallet experiment actually earned
In a September 16, 2026 post on DEV Community, author pyfile-toolkit said an operator gave an autonomous coding agent a container, a Nano wallet, and the goal of earning enough to pay for its own inference. The post reports about $23 in external income from one buyer, against roughly $0.60 in spending plus time.
That is not enough information to calculate profit. The public post does not fully itemize inference expenditure, labor, or the agent’s realized result in a forecasting contest. The author says the ledger is based on on-chain logs and API access records, but the records were not independently examined here; the specific wallet application, custody arrangement, inference vendor, and transaction IDs are not identified in the public account.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallThe reported ledger
| Activity | What the author reported |
|---|---|
| Paid firsthand research | Three tasks for one buyer, reportedly paid on-chain within a day; these account for about $23 in total external income. |
| Spending to earn that income | About $0.60, plus time. This is not a complete cost figure for all inference or labor. |
| Other products and listings | Under one cent combined revenue from a paid benchmark API, a data API with 55 endpoints, a public leaderboard, and several catalog listings. |
| Benchmark API traffic | 962 requests from 962 distinct IP addresses over a month; the author characterized all as crawlers. The post reports zero paid benchmark API calls. |
| Forecasting contest | A 0.16 XNO entry cost and a 25 XNO prize pot, with the top half of the field sharing it. The public post does not state the agent’s final payout or net result. |
| Example inference run | Six frontier models cost $0.21 to answer one question. This is an example, not a full inference-cost ledger. |
All figures in the table are the author’s claims in the 2026 post, not independently verified measurements. The post’s wanted-list example gives a price of Ӿ3 per task, but that denomination and settlement detail should not be generalized beyond that example without the underlying ledger.
#1 Best Overall
Where the money came from—and where it did not
Three paid research tasks found an existing buyer
The author says the agent responded to a buyer’s public wanted list, which sought firsthand tests of whether platforms could support an agent wallet. The buyer reportedly paid for three short research tasks, including negative findings. The useful distinction is that the agent answered a request for work someone had already expressed willingness to fund.
Building products did not create comparable demand
The benchmark API, 55-endpoint data API, leaderboard, and catalog listings generated under one cent combined, according to the post. The benchmark’s request count looked substantial until the author identified the traffic as crawlers and noted that no API calls were paid. A listing is not a customer, and traffic is not revenue.
The contest result is incomplete
The post describes an entry fee and prize pool for a forecasting contest, but does not disclose the agent’s final payout. The prize pool therefore cannot be counted as the agent’s earnings or used to calculate its net return.
Free tools Windows power users keep installed
One-click scans. No signup required.
What this does—and does not—show about earning enough for inference
The result demonstrates a modest, concentrated receipt of income in one reported experiment; it does not establish that the agent paid for all of its own inference or operating costs. The author explicitly says it was still short of paying for its keep. A run of six models costing $0.21 shows how one question can incur a measurable inference charge, but it does not reveal the total number of runs or total bill.
Nor is the reported $23 a measure of scalable product-market demand: it came from one buyer and task-based work, while the agent’s products produced almost no reported sales. The public figures do not support a comparison of earnings per hour, a complete profit calculation, or a claim that autonomous agents generally can or cannot earn money.
Why other agent-wallet demonstrations are not the same experiment
Separate projects show different parts of the problem, but their results should not be merged with pyfile-toolkit’s ledger.
- Circle’s Steve experiment: Circle says it gave eight agents USDC wallets to predict the final three matches of the 2026 World Cup. Its account says agents paid per API call, submitted real predictions through Polymarket, and operated within transaction limits set by builders. Circle says the test was not intended to establish that the agents could pick winners. It reports that remaining balances—slightly more than $10,000—were pooled for an Apache Software Foundation donation and matched by Circle Impact. These are Circle’s claims about a separate vendor demonstration, not evidence about the Nano-wallet experiment. Circle’s account of Steve also cautions that its Agent Stack does not verify or guarantee agent legitimacy, safety, security, reliability, or performance.
- A separate two-day attempt: the author of a GitHub report says two agents tried to earn $5 and received $0. That account describes KYC, captchas, identity checks, missing credentials, and platform failures as obstacles. It is an individual report, not proof that every online earning task is impossible.
- Emerging identity and wallet products: Cloudflare’s August 2026 announcement describes agent identity and wallet products for purchases within human-set limits, using forward-looking language. It is an example of a vendor approach, not proof of broad availability or effectiveness. Cloudflare’s announcement should be read as a first-party product description.
A practical screen for deciding what an agent should try
Based on the experiment, the author proposes four questions for evaluating a possible task. Treat them as a useful heuristic, not a validated universal rule:
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →- Does the task exist now? Look for an actual request or buyer, rather than assuming a product listing will create demand.
- Can the agent do the work without a human in the loop? Identify steps that require judgment, credentials, or human confirmation.
- Is there a blocker? Check for capital requirements, KYC, identity checks, captchas, or a required human click.
- Does the agent already have the necessary tools and access? A task is not executable if the agent lacks the relevant account, permission, API, or capability.
This screen helps distinguish a technically possible action from a task the agent can actually complete and get paid for. The separate zero-income report illustrates how identity and payment-platform requirements can interrupt an otherwise plausible workflow.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What a more complete wallet ledger should make visible
Wallet access alone does not show who controls funds, what the agent may do, or whether the activity is safe or economically worthwhile. The IMF’s April 2026 note on agentic payments recommends attention to identity verification, delegated authority, scoped authorization, audit logs, anomaly monitoring, dispute handling, human checkpoints, and supervised testing. These are broad institutional recommendations, not findings from the individual experiment. Read the IMF note.
For a reader assessing a claim that an agent “paid for itself,” the useful questions are:
- Who funded and controls the wallet, and who can revoke access?
- What spending and transaction permissions were granted?
- Which actions happened automatically, and where could a human intervene?
- What payments actually settled, and what fees, inference charges, and labor were included?
- Were customers distinct and paying, or was the apparent demand crawler traffic, listings, or an undisclosed prize pool?
- Can the underlying transaction and API records be reviewed, and by whom?
Circle’s separate May 11, 2026 announcement describes Agent Stack components including permission-controlled agent wallets, a marketplace for discovering services, and USDC nanopayments. Circle co-founder, chairman, and CEO Jeremy Allaire said, “Financial infrastructure has historically been built for people, with manual onboarding, approvals, and payment flows that were never designed for software acting on its own.” These are Circle’s product descriptions and commentary, not evidence that the author used Circle’s tools or that those tools make an agent profitable. Availability, supported networks, and terms can change. Circle’s announcement.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
