Codex

Codex is not just for coders, that sounds obvious once you use it properly.

But most people still file it under “AI coding tool” and move on. That is the miss.

I have been living inside these tools for a while now. Claude Code, Codex, Obsidian, skills, plugins, automations, little local scripts, half-built systems that start ugly and get useful through repetition. The pattern keeps coming back to the same thing.

The best AI work does not start with code.

It starts with context.

A person knows what needs to happen. A founder knows the messy company problem. A marketer knows the brief. A leader knows the decision that keeps getting delayed. An operator knows the recurring task that eats two hours every week. A researcher knows the question that needs structure before it needs an answer.

Codex matters because it is starting to sit in that layer.

Not just “write this function.”

“Take all this scattered work and turn it into something I can use.”

The coding label is too small

OpenAI’s latest Codex update is easy to skim as a product announcement.

New role plugins. Sites. Annotations. More ways to work with the thing.

Fine.

But the useful read is deeper than that. Codex is moving from a tool for software engineering into a workbench for people who can explain what good output looks like.

OpenAI says more than 5 million people use Codex weekly, and about 20% of users are non-developers. That second number is the one I care about. Analysts, marketers, operators, designers, researchers, investors, bankers. People who are not sitting there trying to refactor a React app all day.

They still have work that looks like building.

They build dashboards. They build briefs. They build project trackers. They build customer review packs. They build launch hubs. They build repeatable ways of doing the annoying stuff they previously accepted as “just part of the job.”

That is the part people miss.

Building is no longer reserved for people who identify as engineers.

You still need taste. You still need judgment. You still need to know what problem you are solving. But the gap between “I can describe the workflow” and “I can make something useful exist” is shrinking.

That is a very different story to “AI writes code now.”

Sites make it concrete

The new Sites preview is the part I think founders and leaders should pay attention to.

Codex can now create hosted interactive websites and apps that can be shared inside a workspace by URL. OpenAI describes dashboards, planners, review spaces, project boards, galleries, and lightweight tools.

That might sound like a small format change.

It is not.

For years, a lot of business work has been trapped in documents, slides, spreadsheets, and Slack threads. Sometimes that is fine. Often it means the work is static when the problem is not.

A project plan wants owners, status, decisions, risks, links, questions, and updates.

A customer review wants usage data, open issues, product updates, renewal risk, and next steps.

A launch plan wants messaging, dates, assets, approvals, and a place for people to point when they ask, “Where is the latest version?”

That does not always want to be a deck.

Sometimes it wants to be a small site.

This is where Codex gets interesting for non-coders. You do not need to start by saying, “Build me a Next.js app with these components.” You can say, “Create a site for this customer review. Use these notes, this usage export, this product update, and this list of open questions.”

Then you can refine it.

Move this section. Change this label. Make the executive view tighter. Add a decision log. Turn the messy spreadsheet into a scenario planner. Keep the site updated as details change.

That is not a coding workflow in the normal sense.

It is a work-shaping workflow.

Plugins are the missing bridge

The problem with most AI adoption is not that people refuse to use AI.

It is that the tool sits outside the work.

You ask a model a question. Then you copy something out of a doc. Then you paste from Slack. Then you explain the context again. Then you ask it to make a first draft. Then you manually carry the result back into the place where the team actually works.

That pattern gets old fast.

Plugins matter because they move Codex closer to the work instead of forcing the work to move toward Codex.

OpenAI is launching role-specific plugins for data analytics, creative production, sales, product design, public equity investing, and investment banking. The important detail is the bundle. Apps, skills, instructions, and workflows together.

That is exactly how I think about useful AI systems.

Not one giant prompt.

A set of instructions that knows your standards. A set of tools that can reach the right sources. A workflow that repeats cleanly. A review point where a human still owns the call.

This is why I have spent so much time building skills in my own setup. A good skill is not magic. It is a saved way of doing work. It says: when this task appears, read these files, follow this process, use this output format, do not invent missing context, and leave me something I can review.

That is useful for a developer.

It is just as useful for a team lead preparing a weekly update, a customer success person turning calls into account plans, or a founder trying to keep investor notes, product feedback, hiring loops, and customer conversations from turning into mush.

Automations change the rhythm

The other piece people underplay is automations.

A chatbot waits.

An automation returns.

That difference matters.

Codex automations can run recurring tasks on a schedule. Morning briefs. Weekly reviews. Recent file summaries. Cleanup checks. Project status updates. Missing information checks. OpenAI’s guide frames a good automation as specific, repeatable, and easy to review.

That is the right bar.

I do not want AI doing vague work in the background and pretending it is useful. I want it doing the boring repeatable job with a clear review surface.

Every Monday, read the current notes and help plan the week.

Every Friday, summarise what moved, what is still open, and what needs attention.

Every evening, scan the projects and create a current-state snapshot.

Every week, review recent customer feedback and pull the patterns that need a decision.

None of this requires you to be a coder.

It requires you to understand your work well enough to describe the loop.

That is where leaders should be spending time. Not “which prompt should my team use?” but “which recurring decisions, checks, updates, and handoffs should become reviewable AI routines?”

Anyone can use AI, but not passively

There is a version of the AI story that says anyone can use AI because the tool is easy.

I think that is only half true.

Anyone can use AI if they are willing to get specific.

Specific about the input.

Specific about the output.

Specific about what good looks like.

Specific about what should never happen.

That is the difference between playing with a model and building with one.

The people who get the most out of Codex will not all be engineers. They will be the people who can look at a messy workflow and say:

“This happens every week.”

“The inputs are usually these three things.”

“The output needs to look like this.”

“The human decision is here.”

“This part can be turned into a site, a report, a dashboard, a checklist, or a recurring review.”

That is not technical genius. That is operational clarity.

And a lot of founders, leaders, and operators already have it. They just have not had tools that reward it this directly.

The limitation is still judgment

This does not remove the hard part.

Codex can make a site. It can wire together tools. It can follow a skill. It can run a scheduled check. It can generate a draft, a tracker, a dashboard, or a first pass at a plan.

It still needs someone who cares about whether the thing is right.

The danger is not that non-technical people use Codex.

The danger is that people use it vaguely.

Bad context creates bad work faster. A weak brief becomes a weak site. A lazy automation creates recurring noise. A plugin without clear permissions and review points becomes a mess. A team that does not know its own process will automate confusion.

That is why I keep coming back to systems.

Skills are systems.

Plugins are systems.

Automations are systems.

Sites are a shared surface for the system.

Codex is becoming the place where those pieces can meet.

The opportunity is not to make everyone pretend to be a developer. It is to let more people build the tools, routines, and shared spaces their work has needed for years.

That is the version of AI I care about.

Not magic.

Not hype.

Just capable people getting closer to the work they already understand.

- JC

Thanks for reading! This post is public so feel free to share it.

Share