OpenAI Codex is free right now. That's the trap.

We all understand that free services from a company that is spending billions on computing power won't remain free. It's inevitable. But the reality is we are all already using Codex.

OpenAI recently announced that Codex was made available in ChatGPT for free, but access was limited. At the moment, developers can use code completions, refactorings, or generate entire functions without spending any money, but they have limited usage. And thousands of developers have been already incorporating it into their daily routines.

That's the point.

The Dealer Playbook

This isn't new. It's the oldest growth hack in tech. Give it away until the switching cost is too high to leave.

Slack has done it. Figma has done it. AWS has done it using free tier credits that for some reason consistently run out just when your startup is unable to migrate off them. OpenAI is simply executing the maneuver with a product that interfaces with your real codebase.

Each time you scribe a prompt, each shortcut you memorize to "correct this function," every manual you choose to bypass as Codex provides a prompt response — you're accumulating lock-in similar to technical debt. It builds up quietly and exponentially.

The Quality Tradeoff Nobody Talks About Honestly

Here's what I continue to see in dev conversations: Codex performs quickly. It's faster than other options and people can tell. But the quality of the code it produces? Not as good.

Developers are actively debating Codex versus Claude right now. The tradeoff keeps coming up:

→ Codex gives you speed. You get answers quickly and move on.
→ Claude gives you cleaner output. But it's slower, and it's not free.
→ Speed wins in the moment. Quality wins at 2 AM when you're debugging.

The issue is that bad code that gets there fast gives the impression of being efficient. You ship faster today. Tomorrow, you debug even more. However, at that point, you have already integrated Codex with your muscle. 🎯

Dependency Is the Product

Codex will need to be the coding assistant you grab like second nature for any OpenAI projects.

Once everyone in your team is using it, when your junior developers are capable of coding with it helping as a tool, when your PR (pull request) reviews consider that the output is generated with AI assistance, when your sprint velocity is taking into account the acceleration this tool provides, at that point, you should start considering the costs associated with using it.

And you'll pay it. Because rewiring a team's workflow is harder than just eating the cost.

I still subscribe to tools that I find somewhat annoying for the same reason. The cost of switching is not the subscription money. It’s the cognitive load. It’s building new habits. It’s the painful week of everything not working. 😅

What I'm Actually Doing About It

I don't love Codex. But free tools are free tools. Boycotting them would be silly.

However, I am multitasking:

→ Keeping my prompts portable. I write them in a way that works across models. No Codex-specific syntax dependencies.
→ Not letting it replace understanding. If I can't explain what the generated code does, I rewrite it myself.

We simply aim to make use of the free tier without being exploited as a product of the free tier.

The Takeaway

OpenAI is actually subsidizing your compute today because it wants the value of your future dependency tomorrow. That's not evil. It's just business. But pretending it's charity is foolish.

Use it. Don't need it. That's the only play that keeps you flexible when the invoice arrives. 💸

I have a question for you. Has your team integrated Codex into its standard workflow yet? And if they have, how do you plan to phase it out when the pricing becomes more competitive?