articles about Anthropic’s Claude Code, and the way I take advantage of it for programming, and totally different strategies that I apply to make it simpler. Nonetheless, the final two weeks I’ve been experimenting increasingly more with OpenAI’s Codex and seeing vastly improved outcomes in comparison with Codex just a few months in the past.
For my part, Codex is equally good on lots of duties and has the benefit that it’s, in lots of circumstances, sooner than Claude Code and that it’s higher at doing precisely what it’s requested to do and never performing different duties (which is the issue I’ve skilled with Claude Code).
On this article, I’ll be discussing my expertise utilizing OpenAI’s Codex for superior coding duties and different utility areas, in addition to some strategies that I take advantage of to boost Codex’s efficiency.
Why use OpenAI Codex?
Initially, I need to cowl why try to be utilizing OpenAI Codex. It’s price mentioning that the pricing of Codex for the 20x Max subscription is similar as Claude Code. The one differentiator is the standard of the outputs produced by the mannequin and the way effectively it could full duties.
Contemplating I program each single day, it’s necessary for me to remain up-to-date with the most recent coding fashions and constantly check out new and upcoming fashions, akin to GPT-5.5, to see if it really works higher than my present setup.
I simply began utilizing Codex with GPT-5.5 round two weeks in the past and easily utilized it to some real-world duties I used to be engaged on. That is necessary, as I imagine operating coding fashions on take a look at duties doesn’t really take a look at the mannequin’s capabilities, and it’s not a sound and full take a look at.
Once I used it on among the extra advanced duties I used to be engaged on, I used to be fairly impressed by the outcomes. For my part, Codex was extraordinarily environment friendly at finishing some duties and accomplished them in a short time. Moreover, I bought the impression that Codex was higher than Claude Code at performing precisely the duties I used to be asking it to do and never altering different stuff within the code. Really, an issue I’ve skilled with Claude Code just a few instances is that I ask it to finish a particular activity, and it largely completes that activity, nevertheless it additionally modifications another issues I didn’t need it to vary.
It’s price mentioning that that is very a lot a stability. On one hand, you may have Claude Code’s strategy, which is to provide the mannequin extra freedom to make choices on what needs to be modified, which might result in the mannequin altering components of the code that you simply don’t intend to vary. Then again, you may have Codex’s strategy, which is altering solely precisely what the consumer is asking you to replace. This may, alternatively, have the draw back that it results in bugs all around the code as a result of it’s not up to date, just because Codex simply performs precisely what it’s requested to do and nothing else.
Particular strategies I take advantage of to optimize Codex
On this part, I’ll cowl some particular strategies I take advantage of to make Codex carry out higher than simply out of the field. I’ll cowl my setup and a few strategies.
My setup
Initially, let’s cowl my setup. I do use quick mode on Codex presently as a result of I’m not hitting my limits that usually. Nonetheless, if you’re hitting your limits, it is best to take into account turning off FastMode or getting one other Codex account.
Moreover, I take advantage of further excessive considering after I’m utilizing plan mode and excessive considering or reasoning after I’m utilizing regular mode, and I’m utilizing GPT-5.5, after all.
I’ve additionally given Codex entry to Playwright MCP, which is a approach that it could entry my browser and carry out actions there. That is extraordinarily environment friendly, for instance, for OpenClaw bots, which I’ll cowl within the subsequent part, and for truly going into the browser and testing options that Codex has carried out. As I’ve talked about in a number of earlier articles, permitting your coding brokers to check their very own work vastly improves the efficiency of those coding fashions.
You’ll be able to learn extra on this in my article under:
How you can Make Claude Code Validate its personal Work
Lastly, I’m additionally utilizing YOLO mode with Codex, the place I give it, or enable it to carry out, any motion inside the folder it’s working in. In my expertise, the frontier coding fashions, akin to Claude Code and Codex, should not inclined to creating extreme errors akin to deleting manufacturing databases or related, and so they’ll sometimes warn you earlier than taking irreversible actions.
Moreover, I additionally imagine that should you arrange your codebase and infrastructure appropriately, this won’t actually be a difficulty. An agent otherwise you, for that matter, shouldn’t have entry to completely delete databases and carry out irreversible harm on any infrastructure. That’s sometimes extra an indication of poor infrastructure design decisions quite than a difficulty with a programmer or a coding agent.
OpenClaw bots
One other use case I’ve for Codex is that I’m utilizing it for my OpenClaw bots. One of many nice benefits of utilizing Codex over Claude Code presently is that you could energy your OpenClaw bots along with your Codex subscription, which you aren’t allowed to do along with your Claude Code subscription anymore. That is necessary as a result of, for my part, Codex is a frontier-level clever mannequin you should use to your OpenClaw bots, which additionally has acceptable pricing.
With this, I imply that Claude Code API pricing merely isn’t relevant for nearly all programmers on the market, and thus isn’t an possibility for OpenClaw. As an alternative, you should buy a $100 or $200 subscription with Codex and have a really clever mannequin energy your OpenClaw bots, which I imagine is an effective funding you can also make.
I additionally use quick mode on my OpenClaw bots since I’ve sufficient finances out there for utilization. Nonetheless, you possibly can once more flip it off should you suppose it’s essential, and it, after all, is determined by the use case. In some eventualities, you might be extra depending on quick replies out of your coding agent, and in different situations, you simply carry out a fire-and-forget activity. The time it takes to carry out the duty just isn’t actually necessary.
Worktrees
Sadly, OpenAI Codex hasn’t carried out a easy work tree setup but, as Claude Code has. That is undoubtedly a minus in my ebook, as work bushes are a crucial characteristic to have when I’m engaged on a number of issues in the identical repo on the identical time.
Nonetheless, to fight this drawback, I arrange a easy alias the place I create my very own work tree when spinning up Codex. I did this just by asking Codex to arrange an alias for me in order that if I write the command you see under, it’ll spin up a piece tree with the identify supplied.
codex-wt This was tremendous easy to arrange and solely took a couple of minutes with Codex.
Codex vs Claude Code
On this final part, I need to cowl Codex versus Claude Code and my opinion on evaluating the 2 coding brokers and frameworks. For my part, there’s no clear winner between the 2 fashions. They’re each extraordinarily highly effective, and for my part, I can full even my most advanced duties with each fashions. Nonetheless, I do have some preferences in sure eventualities.
When I’ve a really particular activity I need to full or am on the lookout for particular bugs, for my part, Codex works higher and is extra environment friendly at finishing the duty. In lots of circumstances, Claude Code may also have the ability to full the identical activity, however merely take longer, from my expertise.
Moreover, as I discussed within the OpenClaw part, Codex lets you use the subscription on OpenClaw bots, which Claude Code doesn’t mean you can do. If you’re depending on operating lots of OpenClaw bots, I extremely suggest you do it with Codex.
Quite the opposite, I believe Claude Code is extraordinarily highly effective and might full all my most advanced duties whereas nonetheless having lots of options that I actually take pleasure in. The work tree characteristic, for instance, is a good inclusion from Claude Code, in addition to the brokers’ view, which they just lately launched. For my part, the characteristic stack of Claude Code is extra highly effective than Codex and will thus be a motive to decide on Claude Code over Codex.
Nonetheless, all in all, I imagine the 2 fashions are neck and neck and each extraordinarily highly effective. We’ll simply should observe each fashions sooner or later, proceed testing them, and see which one comes out on prime in just a few months’ time. In the interim, I imagine they’re each nice decisions, and what works greatest for you depends in your scenario and preferences.
Conclusion
On this article, I mentioned the right way to get probably the most out of OpenAI’s Codex. I mentioned why I began utilizing OpenAI Codex, highlighting how I want to remain on prime of the most recent coding fashions and need to evaluate it in opposition to Claude Code. I bought a really optimistic first impression, noticing that the mannequin was in a position to carry out even probably the most advanced duties I’m engaged on. I then lined some particular strategies I take advantage of to make Codex higher, akin to:
- letting it validate its personal work
- organising an alias for work bushes
- utilizing it for my OpenClaw bots
Lastly, I additionally had a piece evaluating Codex vs. Claude Code, the place I highlighted that it’s extraordinarily shut, and which one is the higher mannequin. Which mannequin is healthier for you is determined by your preferences. I like to recommend that you simply look into each fashions to see what works greatest for you, and likewise carefully observe the fashions within the close to future, as a result of I imagine lots of new options and extra highly effective LLMs will turn out to be out there.
👉 My free eBook and Webinar:
🚀 10x Your Engineering with LLMs (Free 3-Day E-mail Course)
📚 Get my free Imaginative and prescient Language Fashions e-book
💻 My webinar on Imaginative and prescient Language Fashions
👉 Discover me on socials:
💌 Substack
