GitHub Actions is the Only AI Infra I Needed
This started very simply.
Someone in a Discord group asked me if I could build an agent that curates a newsletter every week.
Not some overcomplicated “AI system”. Just something that runs, finds good content, and outputs something usable. On a schedule.
That’s when it clicked.
I don’t need infrastructure. I already have it.
I’ve been using GitHub Actions like everyone else for CI/CD. But if you ignore the label, it is just a machine that runs code on triggers. Push events, manual triggers, or cron jobs.
That is enough to run an AI agent.
So the setup becomes very simple. A scheduled workflow runs twice a week. It pulls data, runs some AI logic to filter and curate, generates output, and pushes it somewhere. Could be a markdown file, could be an email draft, could be a PR.
No servers. No deployments. No infra setup.
Just a YAML file.
The interesting part is not the automation. It is the control.
The agent does everything. It collects, decides, generates. But the last step is human. You review the output, approve it, or reject it.
That’s it.
No complex feedback loops. No heavy “alignment systems”.
Just a simple rule. AI proposes. Human disposes.
This feels closer to how practical AI systems should work.
And GitHub Actions turns out to be a very good place to run this. It is free, which matters if you are not making money yet. It is reliable. And everything is transparent. Logs, history, outputs, all sitting in the repo.
It is basically a free computer on the internet that people ignore because they think it is only for CI.
If you are building early AI products, the biggest mistake is overthinking infra. You do not need queues, workers, orchestrators, and ten different services just to run something twice a week.
You need a trigger and a runtime.
That already exists.
Of course, this only works if you can actually build the AI part. Infra will not save you there. But if you can, this setup is enough to ship something real.
One caveat. Do not romanticize this. This is not a signal to quit your job and go all in on AI agents. It is just a very efficient way to build and test ideas.
Think of it as leverage, not an escape plan.
The real shift is mental.
Instead of asking what infrastructure you need, ask what already exists that you can use in a different way.
For me, that answer was GitHub Actions.
And so far, it is enough.