It's all moving so quickly!

MMaximilian Schwarzmüller
Computing/SoftwareBusiness NewsAdult EducationInternet Technology

Transcript

00:00:00it's probably not super shocking to you when I tell you that everything's moving super fast with
00:00:06AI. And I don't even mean the model releases where we get new models almost every week and we got a
00:00:12new best model every week of course. I'm specifically referring to developer tools here or these agentic
00:00:20engineering tools. It's only been a couple of weeks ago in January that I recorded and released a video
00:00:26where I shared my views on Cloud Code, Open Code, Cursor and GitHub Copilot. Now back then I only
00:00:35used Codex for I don't know a few weeks and only here and there but I have been heavily influenced
00:00:44by Peter Steinberger who we of course all know from OpenClaw which is an amazing story but it's
00:00:49a different story because back in January already I saw posts from him time to time where he mentioned
00:00:56that Codex is really amazing and my understanding wasn't is that he's referring to both the model and
00:01:04probably also the AI coding agent so an alternative to Cloud Code and that is when I started playing
00:01:13around with it too but Cloud Code was still my main driver back then and in 2025 I've mostly used
00:01:20GitHub Copilot and Cursor and Open Code so yeah I'm really switching a lot because of course I
00:01:26yeah I really like playing around with these different tools and seeing what they can do,
00:01:30which features to offer. I mean I'm the kind of guy who liked having those different JavaScript
00:01:35frameworks back in 2019 so I guess I just like learning new stuff but that's why I also enjoy
00:01:41diving into these tools. Codex by OpenAI is a really great agentic engineering tool I have to
00:01:50say and to be very clear right away I wouldn't say it's better than Cloud Code or Cursor and
00:01:57that changes all the time anyways but it's great and there are a couple of reasons why I think
00:02:02that Codex is a great tool and might be worth a closer look. One reason is the Codex AI models so
00:02:08yeah the models have the same names as the tool. Here I'm talking about GPT 5.2 Codex and GPT 5.3
00:02:16Codex by the way fun fact GPT 5.3 was released in early February and I became like a heavy Codex
00:02:24user in late January so most of my time has been with this model and GPT 5.2 was only in December
00:02:32so yeah those model releases are really getting close to each other and the obvious reason is that
00:02:38of course OpenAI wants to compete with Anthropic and the Cloud models Opus so yeah there is that.
00:02:45But I will say the Codex models really are great and to me it feels like 5.3 Codex often takes a
00:02:52bit longer but tackles certain tasks more thoroughly than Opus but then again really I'm not in the camp
00:03:01of telling you that one model is so much better than the other or that there are such huge
00:03:07differences between these tools but what I do like about Codex the tool now not the model is of course
00:03:15that it's powered by the models but also that it has different form factors different ways you can
00:03:20use it for example it has that in my opinion really amazing desktop app at the point of time where I'm
00:03:27recording this that only exists for Mac OS but I think a Windows version is planned. By the way I
00:03:33did create a new course about Codex and I used Codex combined with Cloud Code to build this
00:03:41MaxDraw app which I'm using in all these videos which I've been using for the last weeks and
00:03:45months already so that is built by me with those AI tools and I got a lot out of that that is why I
00:03:52basically condensed all my learnings all my experiences and how to use skills how to work
00:03:57with work trees how to offload tasks to the cloud I put that all into a course and if you're interested
00:04:03if you want to learn that you find a link to it below this video of course. So this app this native
00:04:09desktop app is really amazing in my opinion and I would say it gives us an idea of how
00:04:16agentic engineering could look like in the future because in this app you don't primarily look at
00:04:23your code there there are no files here instead here you have your sessions and you can basically
00:04:30send your prompts to Codex and have it work on your tasks and then you can always review changes
00:04:39in here. So here I have a conversation in a different project where I have changes quite a
00:04:45lot of changes here actually and you can then open this mode here where you can dive into the
00:04:51different files and see what changed compared to your last commit and my workflow has definitely
00:04:58changed such that I don't write a lot of code myself anymore I still do here and there but I
00:05:04mostly architect the software and I steer the AI and I try to do that very carefully and intentionally
00:05:11and have small tasks but I then mostly spend my time reviewing or not mostly but a lot of my time
00:05:19is reviewing the code because what I don't do I don't vibe away and just accept it blindly instead
00:05:25I read the plans it makes I change the plans it makes and I do review the code and I really
00:05:32like the look and feel of this this app and how easy it is to work with the agent switch between
00:05:38sessions and also yeah review the code and then there's a bunch of other useful stuff in there
00:05:43into which I also get in my course of course like working with work trees or cloud projects and
00:05:49really great app I have to say I really like it a lot now of course if you don't like that app or if
00:05:55you're on windows and you can't use it yet I will say codex is first and foremost of course also a
00:06:01cli tool a terminal user interface like clod code and it's a great one I will say it has just the
00:06:10right set of features I feel like and what was really great doesn't sound like a big thing maybe
00:06:18but it's written in rust which doesn't matter too much to me I'm not in the rust fanboy camp rust is
00:06:24a great language in my opinion though but it's really good performance wise where clod code can
00:06:31feel a bit sluggish and slow from time to time this codex cli for me worked pretty much flawlessly
00:06:40and again just to be clear clod code cursor all amazing but I will say it's it's a great cli and
00:06:46combined with that native app and the great model I had a great time working with it now what I find
00:06:53interesting is if we take a look at how these ai models developed and evolved I mean how these
00:07:04coding agents developed and evolved over time not the models the agents it's really interesting
00:07:10github copilot started in 2021 and back then it was just as autocompletion powered by a model
00:07:18named codex which has basically nothing to do with today's codex then in 2023 we had cursor which was
00:07:25the big thing and and still is a big thing but to me it feels like now in 2026 clod code and codex
00:07:36are right now at least the more popular or hyped tools now hype is a dangerous thing especially in
00:07:44the ai space and we have seen how quickly things change I mean codex has absolutely been not hyped
00:07:52at all like a month ago I would say so that has only recently changed and it can't change again
00:07:57though again it's a great tool so there is some reason for the hype but right now to me it feels
00:08:03like cursor is almost well not dead but it was the main agentic engineering tool like three months ago
00:08:14and now it feels like it's declining and nobody's interested in it anymore which I'm sure is wrong
00:08:21that's just the x bubble I live in which is probably not the best indicator but still it's interesting
00:08:30how quickly this changes github copilot for example to me now feels more like it's maybe coming back
00:08:37because it was the first then cursor totally ate its breakfast or whatever you say and then github
00:08:45copilot had a really hard time coming back and now to me it feels like they are doing a better job
00:08:51here I mean with the latest version of visual studio code the whole agent interface was improved
00:08:58and modernized and by the way you can also use cloud code and codex from indir if you have the
00:09:04respective subscriptions and tools installed so it feels like maybe this is having a comeback
00:09:10and cursor to me feels a bit like it's going down and who knows in three weeks or months everything
00:09:16might be totally different again which I guess takes me back to an earlier point I'm the kind of guy who
00:09:23didn't worry too much about all the javascript framework wars back in 2019 and no not just
00:09:30because I create courses I have not created courses on all the frameworks we had back then
00:09:34I just genuinely like learning stuff and seeing where I can get the most use out of obviously if
00:09:41you're in a job and you have a certain framework to work with or a certain tool to work with
00:09:46that doesn't bother you at all but for me I like having all this innovation and I'm super excited
00:09:52to see how these ai coding tools will evolve what will be new and trending in a couple of
00:09:59months and obviously I'll try my best to keep my respective courses updated I got courses on all
00:10:04these tools now because I've worked with all of them a lot so I'll of course try to stay up to date
00:10:10there but yeah I will say it's it's a really interesting time to live in it's so fast moving
00:10:16and I totally get that this can also be intimidating but at the same time I feel like it's it's also
00:10:22really exciting to some degree but yeah I know not everybody shares that opinion for sure

Key Takeaway

In the fast-paced 2026 AI landscape, developer workflows are shifting toward 'agentic engineering' where tools like Codex and Claude Code allow developers to act as high-level architects and rigorous code reviewers.

Highlights

The landscape of AI coding tools is evolving rapidly with new model releases like GPT 5.3 Codex occurring within weeks of each other.

OpenAI's Codex tool features a standout native macOS desktop app that shifts the focus from file-level editing to session-based agentic engineering.

The speaker emphasizes a shift in workflow where the human acts as an architect and reviewer rather than a primary coder.

Codex provides a high-performance CLI tool written in Rust, offering a snappier experience compared to alternatives like Claude Code.

There is a visible shift in the 'hype cycle' where Cursor's dominance is currently being challenged by the resurgence of GitHub Copilot and new entries like Codex.

Modern AI engineering involves using 'work trees' and offloading complex tasks to cloud projects to maintain efficiency.

Timeline

The Rapid Evolution of AI Coding Tools

The speaker opens by highlighting the incredible speed at which AI developer tools and agentic engineering platforms are currently moving. He references his previous comparisons of tools like Claude Code, Cursor, and GitHub Copilot, noting how quickly his own preferences have shifted. A major influence on his current direction is Peter Steinberger and the success of OpenClaw, which led him to explore the OpenAI Codex ecosystem. This section establishes the context of 2025 and 2026 as a period of constant experimentation with different JavaScript-style framework wars now applied to AI. It matters because it sets the stage for why staying adaptable is the most important skill for modern developers.

Deep Dive into OpenAI Codex Models and Performance

This segment focuses on the specific capabilities of the GPT 5.2 and 5.3 Codex models released by OpenAI to compete with Anthropic's Claude Opus. The speaker notes that GPT 5.3 often takes longer to process but provides a more thorough analysis of complex coding tasks than its competitors. He clarifies that while 'Codex' is the name of both the model and the tool, the integration between them provides a seamless engineering experience. The discussion covers the rapid release cycle, with new iterations appearing in December, January, and February. This technical breakdown is crucial for developers choosing between the thoroughness of OpenAI and the speed of other LLM providers.

Agentic Engineering and the Desktop App Experience

The speaker introduces the Codex native desktop app for macOS, which he considers a glimpse into the future of agentic engineering. This app moves away from traditional file explorers, instead focusing on 'sessions' where the user steers the AI through high-level prompts. He shares his personal success story of building the 'MaxDraw' app using a combination of Codex and Claude Code. The workflow described involves carefully architecting software and reviewing AI-generated plans rather than 'vibing away' and blindly accepting code. This section emphasizes the importance of manual code review and intentional steering to ensure high-quality software output.

CLI Tools, Rust Performance, and Market Trends

Beyond the desktop app, the speaker evaluates the Codex Command Line Interface (CLI), praising its performance due to being written in Rust. He compares this favorably against Claude Code, which he mentions can sometimes feel sluggish or slow during heavy operations. The narrative then shifts to the broader market, charting the timeline from GitHub Copilot's 2021 origins to the 'hype' surrounding Codex in 2026. He observes that while Cursor was the dominant tool just three months ago, its popularity appears to be declining in certain developer circles. This observation serves as a warning about how quickly 'hype' can shift in the artificial intelligence space.

The Resurgence of GitHub Copilot and Final Reflections

In the final section, the speaker discusses the potential comeback of GitHub Copilot following significant improvements to its agent interface within Visual Studio Code. He notes that users can now even use Claude Code and Codex directly within the VS Code environment if they have the right subscriptions. Despite the potential for 'tool fatigue' or intimidation caused by the speed of change, the speaker expresses genuine excitement for the ongoing innovation. He concludes by promising to keep his educational courses updated to reflect these frequent industry shifts. This conclusion reinforces the theme that while tools change, the fundamental joy of learning remains constant for passionate engineers.

Community Posts

View all posts