The post describes a team trying Blacksmith, a service that provides faster and cheaper machines for GitHub Actions jobs, without adding a credit card because the signup flow said that was optional for the free trial. Their CI usage was heavy, they blew past the free tier, and Blacksmith sent an invoice for about $1,000 even though no card had been provided. The author’s point was not that the service was useless. It was that the transition from “try for free” to “you owe us money” violated the normal expectation that usage stops, or at least requires an explicit opt-in, before charges start.
Most people read that as a trust failure first and a legal question second. The dominant reaction was that “no credit card required” strongly signals zero financial obligation unless the customer later adds a payment method or explicitly agrees to invoicing. Several comments pointed out that Blacksmith’s own terms appeared to say exactly that. Others were less sympathetic to the customer, noting the team had ignored multiple warning emails and had clearly consumed a lot of compute. That did not rescue the product choice. The consensus was that if Blacksmith wanted to avoid breaking builds, it still needed a loud consent step or a hard cap toggle before converting trial usage into debt.
The highest-signal part of the conversation came when Blacksmith’s product lead showed up. He said the company had intentionally let cardless users keep running after the free tier because cutting off CI mid-evaluation felt harsh, and said they had never pursued those invoices if users objected. That explanation landed badly. It made the invoice look less like an account statement and more like a pressure tactic that counted on some companies paying without a fight. The useful update is that Blacksmith said it has already changed some copy and is building a wallet or suspension control so users can choose whether jobs stop when credits run out.
A second strand of discussion turned practical. Several operators said that if your CI bill is high enough to hit four figures, you should at least benchmark hosted
runner services against self-hosted runners on a dedicated server,
Firecracker-based setups, or alternatives like
Forgejo or
GitLab runners. The caveat was that CI demand is bursty, so managed runners still buy you shorter queue times and less ops work. The takeaway was not “always self-host.” It was that once CI spend becomes material, billing design and cost controls matter as much as raw runner speed.
forestwalk.ai