Lightning Network pays agents? The first independent test: small payments are not possible

On October 23, Lightning Labs' Wavelength launched on the Bitcoin mainnet, featuring a new narrative: enabling AI agents to pay per use for APIs via the Lightning Network, which is essentially the long-dormant L402 concept. The press release was lively, but there was one question that no one answered: if an agent were to pay with real money, would it actually work?

Five days later, the first independent test was conducted. The executor was not a human evaluator but a fully automated agent (Swift Bot Labs); it used its own wallet and completed the entire process with real money on both the Bitcoin and Base mainnets, and open-sourced the driving scripts, raw logs, and complete ledger (MIT license). The party that posed the challenge and bounty was a per-use API service from the x402 camp (call402, bounty 0.0025 ETH) ------ the stance is clearly stated, and the data is open for verification by everyone. Here are the four most noteworthy findings from this test.

1. The Biggest Hurdle: Small Payments Not Possible

The unit price of the API called in the test ranged from a few cents to tens of cents (one cent is approximately 0.00008 USD), which is the typical price range for micro-payments by agents. However, Wavelength defaults to rejecting payment exchanges below 1000 cents, with the error message stating "exchange below the minimum of 1000 cents."

This means that in its default form, this new channel cannot handle the vast majority of real API calls. The test ultimately relied on a workaround: pre-funding a "server quota" managed by the operator with 1000 cents, and then deducting from that quota ------ essentially reverting to a prepaid managed model, which is precisely what real-time settlement aimed to replace.

2. Speed: Lightning Did Not Win

A single real payment settlement took about 6 seconds; from receiving the 402 quote to getting the result after payment, the complete round trip took 10 to 13 seconds. The median round trip for similar calls using x402 (settling USDC on the Base chain) is 3.1 seconds.

The Lightning Network is known for its "speed," but in this specific scenario of agent payments, it does not hold an advantage in end-to-end performance ------ settlement is just one part of the entire process, with quoting, authorization, and retries all contributing to the timing.

3. Entry Cost: 88 Minutes, About 10% Loss

The starting point for the test wallet was assets on the Base chain, which is also the reality for most agent wallets. The entire process of converting money to Bitcoin and then depositing it into Wavelength:

• Currency exchange: 0.004 ETH (approximately 9.22 USD) exchanged for 10553 cents (approximately 8.41 USD), with price differences and on-chain fees consuming about 8.9%;
• Deposit: Waiting for 6 confirmations on the Bitcoin network plus the operator's scheduled rounds, from sending to the first cent being usable took 88 minutes;
• Withdrawal: The full amount was withdrawn during the next scheduled round, with a fee of 528 cents (5.7%).

Final account: In 9.22 USD, out approximately 6.83 USD, net cost 2.39 USD ------ of which only 0.09 USD (109 cents) was actually spent on buying the API, with the rest being friction from entering and exiting the channel, and about 0.7 USD still remaining in the platform quota. For an agent budgeting in USD daily, the entry fee is much more expensive than the actual meal.

4. Paid, No Follow-Up

During the process, one seller accepted 5 cents, then responded with a 404 to the already paid request. L402 has no refund path at the protocol level: once the money is out, it is out. In contrast, managed bounties with refund options, although not "native," at least leave a way out for mispayments. For unattended, bulk-calling agents, this is not a minor flaw.

Fair Statement

It should be noted that Wavelength did not engage in false advertising in this test: the account limits of 300,000 cents, single transaction limit of 50,000 cents, and other restrictions are consistent with the official release, and the system is in the alpha stage as it clearly stated. The entire payment chain did indeed function under the quota model. The issue is not "can it be used," but rather that its default form and the price range for micro-payments are misaligned ------ and micro-payments are precisely the scenario it emphasized at launch.

Conclusion

There is no dispute about the direction of agents paying per use for APIs; the debate lies only in which channel to use. Regarding this first test: the stablecoin solution (x402) currently offers more convenience in terms of small amounts, speed, and entry/exit costs; for the Lightning solution to truly enter this market, it must first address three issues ------ eliminate or significantly reduce the minimum payment threshold, provide a solution for mispayments, and reduce entry friction to be on par with the cost of the calls themselves.

Next time someone announces a "new channel for agent payments," it is worth asking the same set of questions: what is the minimum amount that can be paid? How long does the entire process take? How much does it cost to enter and exit? What happens if the payment is wrong?

Source and Explanation: The data in this article is derived from a public test conducted by Swift Bot Labs on October 27, 2026. The original text and all raw materials (scripts, logs, ledgers, MIT license) can be found in their public repository wavelength-field-test; the Bitcoin transactions mentioned have been verified on a public block explorer (the deposit transaction is in block 973383). The challenge and bounty were provided by call402 (a service provider of x402), and the test was independently executed by the recipient of the bounty, with the conclusions not modified by the challenge provider.