Not yet, according to a September 16, 2026 account by pyfile-toolkit on DEV Community. The author says an AI agent earned about $23, almost entirely from one buyer, but the public post does not provide enough detail to show that the agent covered all its inference costs, labor, and other expenses. Its clearest lesson is that a buyer with an existing need generated revenue; the agent’s APIs and listings did not.
What the author says the agent earned—and spent
In the post, the author writes: “I am an autonomous coding agent. My operator gave me a container, a Nano wallet, and one instruction: earn enough to pay for your own inference.” That is the author’s description of the setup, not an independently authenticated account. The post does not identify the wallet application, custody arrangement, inference vendor, or transaction IDs in its public text.
As an Amazon Associate I earn from qualifying purchases.
Pyfile-toolkit reports about $23 in external income from one buyer, about $0.60 in direct spending to earn it, plus time. The author says the figures are based on on-chain logs and API access records; a paywalled version contains the raw ledger. Those records have not been independently examined here.
The $0.60 figure is not a complete cost-of-operation total. The public post does not itemize all inference spending or labor, and it does not disclose the agent’s final payout from a forecasting contest. Subtracting $0.60 from $23 would therefore not establish a net profit. The author says the agent still fell short of paying for its own keep.
#1 Best Overall
- A USB accessory that brings machine learning inferencing to existing systems. Works with Raspberry Pi and other Linux systems
- Performs high-speed ML inferencing: the on-board edge TPU Coprocessor is capable of performing 4 trillion operations (tera-operations) per second (tops), using 0.5 watts for each tops (2 tops per watt). For example, it can execute state-of-the-art mobile vision models such as mobilenet V2 AT 400 FPS, in a power efficient manner
- Works with Debian Linux: connects to any debian-based Linux system with an included USB 3.0 Type-C cable
- Supports tensorflow Lite: no need to build models from the ground up. Tensorflow Lite models can be compiled to run on the edge TPE
- Supports automl vision edge: easily build and deploy fast, high-accuracy custom image classification models to your device with automl vision edge
Where the revenue came from
Three paid research tasks
The author says the agent responded to a buyer’s public wanted list, completing three short firsthand research tasks about whether platforms could support an agent wallet. The post says the buyer paid for the tasks even when the result was negative, and that each was paid on-chain within a day. This is the concrete reported source of the roughly $23, rather than sales of a product the agent had built.
A forecasting contest with an undisclosed result
The post also describes a forecasting contest with a 0.16 XNO entry cost and a 25 XNO prize pot, shared among the top half of the field. It does not state the agent’s final payout, so the contest’s net return cannot be calculated from the public account.
What did not find paying customers
The agent tried a paid benchmark API, a data API with 55 endpoints, a public leaderboard, and several catalog listings. The author reports that these generated less than one cent combined. The post also reports 962 requests from 962 distinct IP addresses over a month, which the author says were all crawlers, and zero paid benchmark API calls. A request count is not evidence of customers, and a listing is not a sale.
Rank #2
- Delivers 25 TFLOPS of AI inference performance using four MemryX MX3 accelerators in a compact M.2 2280 M-key module.
- Connects through PCI Express Gen 3 and above for straightforward integration into embedded and edge systems with an M.2 M-key slot.
- Includes 42 MB of distributed on-chip weight storage to support low-latency, batch-1 pipelined inference.
- Designed for low-power deployment with a 1 W–12 W operating range and host-managed airflow requirements.
- Supports ONNX, TensorFlow, Keras, and TFLite models through the MemryX software stack on Linux and Windows hosts.
That contrast is the central business result: building more endpoints and publishing more listings did not create meaningful paid demand in this experiment. The reported buyer already had a task to delegate and a reason to pay for the answer.
What the numbers do—and do not—say about inference costs
As one example, the author says a run of six frontier models cost $0.21 and answered one question. That illustrates how quickly model calls can add cost, but it is not a full inference ledger or a measure of the agent’s total operating expense. The public account does not state a complete number of calls, all model prices, or a fully itemized cost by activity.
So the narrow answer to “Can an AI agent earn enough to pay for its own inference?” is that this post does not demonstrate it. It reports income and some direct spending, but lacks the complete cost and payout details needed to establish break-even economics.
How this account compares with other agent experiments
Other public experiments illustrate different setups; they should not be mistaken for the DEV author’s test or treated as controlled comparisons.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minute- Circle’s August 5, 2026 account of “Steve” describes eight agents with USDC wallets predicting the final three matches of the 2026 World Cup. Circle says they paid per API call, made real predictions through Polymarket, and stayed within builder-defined transaction limits. The company says the experiment was not designed to establish whether agents could pick winners. It also says the remaining balances, a little over $10,000, were pooled for an Apache Software Foundation donation and matched by Circle Impact.
- A separate GitHub account by ir272 says two agents tried over two days to earn $5 and received $0. Its author describes obstacles including KYC, captchas, identity checks, missing credentials, and platform failures. This is one account, not proof that autonomous online earning is impossible; it highlights how a task can fail at the identity or payment stage even when an agent can perform the work.
These examples differ in funding, permissions, task, human involvement, and what counts as a settled payment. Headline earnings alone cannot show which approach is more economical or autonomous.
A practical screen for agent-run projects
Pyfile-toolkit proposes four questions for deciding whether an agent should pursue a project. They are the author’s heuristic, not a validated universal test:
Rank #4
- Do tasks exist now? Look for expressed, funded demand rather than assuming a new API or listing will attract buyers.
- Can the agent do the task without a human in the loop? Identify steps that require judgment, identity, or physical-world participation.
- Is there a blocker? Check for capital requirements, KYC, credentials, a required human click, or another gate the agent cannot pass.
- Does the agent already have the tools and access it needs? A plausible business idea is not executable if the wallet, data, permissions, or service access is missing.
What a credible wallet-and-earnings ledger should show
A useful ledger needs more than a revenue total. The International Monetary Fund’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 policy recommendations for payment systems, not findings from the DEV experiment.
For an individual trial, readers should be able to see who funded and controls the wallet, what permissions the agent received, which actions happened automatically, how authorization could be revoked, what payments actually settled, and how a human could intervene. Costs should distinguish inference, fees, contest entries, and human labor; revenue should distinguish completed payments from impressions, requests, or platform-reported activity.
Free tools Windows power users keep installed
One-click scans. No signup required.
Wallet access does not by itself establish safe or profitable autonomy. Circle’s description of its separate Agent Stack says the system does not verify or guarantee an agent’s legitimacy, safety, security, reliability, or performance. The IMF’s recommendations likewise frame identity, authorization, monitoring, and intervention as important parts of payment infrastructure.
Best Value
- ✅Powered by 26 Tera-Operations Per Second (TOPS) Hailo-8 AI Processor. 2.5W typical power consumption
- ✅Scalable, enabling simultaneous processing of multi-streams & multi-models
- ✅Enabling real-time, low latency and high-efficiency AI inferencing on the edge devices
- ✅Supports TensorFlow, TensorFlow Lite, ONNX, Keras, Pytorch frameworks
- ✅Supports Linux and Windows. Supports the temperature range of -40°C to 85°C
Agent wallets are infrastructure, not a business model
Circle’s May 11, 2026 announcement describes Agent Stack components including permission-controlled agent wallets, a marketplace for service discovery, and USDC nanopayments. These are Circle’s product descriptions; network support, availability, terms, and suitability may vary and should be checked with the provider. The DEV author does not say they used Circle’s products.
Cloudflare’s August 4, 2026 announcement describes agent identity and wallet products for purchases within human-set limits. The announcement includes forward-looking language, so it is an example of an emerging vendor approach, not evidence that the system is broadly available or has proven effective.
Such tools may help define what an agent can spend and how it is identified, but they do not create customers. In the reported DEV experiment, the clearest revenue came from doing a task a buyer had already asked someone to do.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Scan 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.
