Free AI Is Becoming Infrastructure

Yesterday, someone on Hacker News posted a project called OpenGem: an open-source proxy that load-balances requests across multiple free Google accounts so you can hit the Gemini API without paying for keys. If that sentence made your eyes glaze over, good. That’s the point. The most important part of that story is not the cleverness. It’s the motivation: the author was tired of hitting quota limits while building agents and testing ideas, so they built a workaround that turns ‘free tier’ into something closer to ‘reliable baseline.’

The same day, a separate post went semi-viral in a different corner of the internet: a team hid backdoors inside chunky, 40MB binaries and then asked AI plus Ghidra to find them. This is the kind of thing you’d normally file under ‘security people doing security people things’ and move on. But put it next to the OpenGem story and a pattern shows up: AI is quietly moving from a product you occasionally open to a capability you wrap around everything you do. Not because you want to be trendy. Because it reduces the friction of reality.

If you’re a founder, an operator, or anyone who owns outcomes, you’ve probably had the same thought in a different costume. You’ve opened ChatGPT, asked it something, gotten a decent answer, then closed the tab and gone back to the mess. The answer helped, but it didn’t change the system you work inside. It didn’t make tomorrow easier. It didn’t lower the amount of mental overhead you carry just to keep your business moving.

That’s the AI era’s most common trap: treating AI like a clever intern you ask for help, instead of infrastructure you build your week on. The people who are getting real leverage right now are not the people who know the most model trivia. They’re the people who make the tool dependable. They turn ‘sometimes helpful’ into ‘always available.’ They turn a chat window into a workflow.

The OpenGem story is a developer’s version of a very normal business problem. Imagine you run a dental group, a med spa, or a property management firm. Your team’s work is a conveyor belt of repeated language: appointment reminders, lead follow-up, review requests, screening questions, policy explanations, invoice nudges, lease renewals, internal handoffs. It’s all slightly different each time, which is why it eats hours. And it’s all similar enough that you feel dumb paying humans to do it forever.

Most people try to solve this by saying, ‘We should use AI.’ Then they buy a tool. Then the tool gets used twice, inconsistently, by the one person who likes experimenting. Then it fades into the same graveyard as the company’s Notion docs. Not because AI didn’t work. Because the company didn’t build a system that makes AI the default.

Here’s the uncomfortable truth: if AI only exists as a place you go to ask questions, it will always lose to the urgency of your inbox. But if AI exists as a set of named, repeatable steps that happen automatically when a trigger occurs, it becomes part of how the business breathes. That’s why ‘quota workarounds’ and ‘AI-assisted reverse engineering’ matter. They’re signs that the frontier is shifting from capability to reliability.

So what do you do with that, if you’re not writing proxies or analyzing binaries? You do the operator version. You pick one workflow you already run every week, the kind that drains time because it’s repetitive but still needs judgment. Then you write it down as if you were teaching it to a smart new hire. Not in a 20-page manual. In the smallest number of steps that still preserve quality. Then you decide where AI can draft, where a human must approve, and what the fallback is when the output feels wrong.

Try this today: take the last ten messages your team sent to customers that you could describe as ‘the same email again.’ Put them in a document. Read them like a stranger. You’ll see the pattern immediately. Now rewrite the best one so it becomes a reusable template, and make AI the first draft layer for the next ten. The important move is not the rewrite. The important move is that you are creating a default path that your future self will actually follow.

People love to debate whether AI is hype. The more useful question is: which parts of your week are still being run on fragile human memory? If a workflow breaks whenever one person is tired, distracted, or out of office, that workflow is a liability. AI doesn’t fix liability by being smart. It fixes liability by being present, consistent, and fast enough to keep up.

The security story is a warning as much as it’s a flex. If AI can accelerate the detection of hidden backdoors, it can also accelerate the creation of them. That means your advantage can’t be ‘we use AI.’ Everyone will. Your advantage has to be ‘we use AI inside a controlled system.’ You need ownership, guardrails, logs, and the ability to shut things off when they go weird. You need to know what the model touched, what it produced, and who approved it.

This is where the next wave of winners will quietly separate. Not by being the most excited. By being the most disciplined. By treating AI like electricity. You don’t brag that you have electricity. You wire it into the building so it’s always there, and then you build better rooms because of it.

If you feel behind, good. That’s a real signal, not a moral failure. The gap isn’t between people who know and people who don’t. It’s between people who build systems and people who keep asking for answers. Tomorrow morning, pick one workflow. Make AI the default draft layer. Put a human checkpoint where it matters. And watch how quickly your week gets lighter when the tool stops being a destination and starts being infrastructure.