Original video in Chinese.
Key Takeaway
- Codex Record & Replay is not an upgrade to RPA, but its replacement: RPA is an “instruction-based” dead script that only records pixel coordinates and mechanical actions; once the interface changes, it breaks and cannot handle exceptions or unstructured content. Codex, by contrast, is a “goal-based” Skill that can understand operational intent, context, and verification logic, and it is editable and shareable, completely solving RPA’s core flaws.
- Figma-style network effects are happening again: Figma turned a professional design tool into a collaborative platform that everyone could use (one person using it brought in the whole team); Codex is turning RPA, which only IT engineers could use before, into a Skill that ordinary employees can “learn by showing it once,” where one recording can become the team’s standard SOP, creating extremely strong network effects.
- OpenAI’s real ambition is the full enterprise automation stack: Record & Replay is only the entry point. OpenAI is building a complete AI-native enterprise operating system that includes Goals (goal setting), Automations (scheduled tasks), Heartbeat (process monitoring), and Skills (reusable workflows), with the goal of replacing traditional RPA and becoming the infrastructure for enterprise automation.
Codex recently got a god-tier new feature: Record & Replay. With it, you don’t need to describe things in text. You just do it once, and Codex will learn it, then turn your operation into a Skill for easy reuse later.
That was the topic of my last video. After it was posted, a lot of people said in the comments: isn’t this just RPA? Is this really worth hyping?
On the surface, this new Codex feature and RPA are doing the same thing: recording actions and repeating them. But their underlying logic is completely different.
What does RPA do?
Its core logic is to record your mouse and keyboard operations, then mechanically repeat them.
For example, which coordinates you clicked, which field you copied, which window you opened—it records all of it and turns it into a script that runs automatically later.
That sounds very useful. But the problem is that it runs a “dead script” with no understanding at all. That creates three fatal flaws:
First, any UI redesign causes it to break.
If the UI changes even a little, the script throws an error immediately. So enterprise RPA teams have to spend a huge amount of time maintaining scripts instead of creating new automations.
Second, it cannot handle unstructured content.
For example, if the content isn’t standardized, RPA simply doesn’t recognize it.
Third, when exceptions happen, it can only stop.
Any situation beyond the preset path means RPA can only stop, report an error, and wait for human intervention.
That’s why, even after so many years, RPA has remained only a “tool for IT teams” and has never truly spread to every ordinary employee.
Codex’s Record & Replay is different.
Codex also records, but its purpose is not to capture which pixel you clicked—it is to understand what you are doing and why you are doing it.
After recording ends, what it generates is not a “dead script,” but something called a Skill. It is used to record where data is stored, how materials are organized, how to verify whether the result is correct, and so on. And all of this can be edited and adjusted.
In one sentence, the essential difference is:
RPA is instruction-based. Codex is goal-based.
So I think this is not “smarter RPA.” This is RPA’s replacement. In the end, the RPA market will definitely be eaten up. I’ve seen this logic play out once in tech history.
In 2022, Adobe spent $20 billion trying to acquire a company called Figma.
You should know that Figma’s annual revenue at the time was only $400 million. Why was Adobe willing to pay a 50x premium?
Because Adobe realized that Figma was disrupting the entire way design collaboration worked.
Adobe’s tools were aimed at professional designers: complex, expensive, and completely unusable for product managers and engineers.
Figma’s logic, however, was: let everyone involved in the product collaborate on the same canvas.
It turned design tools from “professional software” into a “team collaboration operating system.” Instead of files being passed around, it became a simple link, with multiple people able to work on the same screen in real time.
More importantly, Figma adopted a “pay for editing, free for viewing” strategy. Product managers, engineers, clients—all the “non-designers”—could join in and collaborate at zero cost. That gave Figma a natural network effect: one designer brings it in, and the whole team follows.
Figma eventually captured 77% market share in UI design.
Now, Codex is doing the same thing:
Turning RPA, which “requires writing scripts to use,” into a Skill that you can “use after showing it once.”
Its target users are expanding from IT engineers to every ordinary employee.
A Skill recorded by one person can be shared with the whole team. The way the person best at a particular task does it instantly becomes the team’s standard SOP, without needing to write any documentation.
That is exactly the same network-effect logic as Figma: one person comes in, and the whole team is brought along.
And I think OpenAI has even bigger ambitions.
RPA is only the entry point.
Codex has already built a complete enterprise automation system:
It has Goals, so you can set what outcome you want to achieve.
It has Automations, so you can set scheduled triggers for tasks.
It has Heartbeat, so you can continuously monitor whether key processes are running properly.
It has Skills, which are reusable, shareable workflow units.
Put these four things together, and this is no longer just an automation tool. This is an AI-native enterprise operating system.
Just like I said in the Knowledge Planet, OpenAI is targeting the much larger “enterprise automation” full stack. This is super, super valuable!
If you are an entrepreneur in this direction, think carefully about my logic, especially the four things I mentioned just now. They will definitely give you inspiration.
OK, that’s all for this episode. If you want to learn AI, want to become a super individual, and want to find like-minded people, come join our newtype community. See you next episode!