Hello and welcome to the 12th issue of AI Powered Engineering.

I decided over the summer that I was going to stop writing this newsletter, and sent a final issue titled “Fin”

After sending that, I received a surprising amount of email from subscribers telling me how much they enjoyed the newsletter, and encouraging me to keep writing it, even if it was in a scaled down version.

What ended up being the tipping point for me is when one of our investors, and an entrepreneur that I really look up to, Brett Hurt, told me how much he liked the newsletter, and that it inspired him to invite me on his podcast to talk about some of the core topics I cover in here.

I really wasn’t expecting this level of feedback after announcing the end of this newsletter. And well, it’s hard to ignore that kind of feedback, so I decided to continue writing AI Powered Engineering, just in a more limited format.

Since I am insanely busy, and don’t have that much time to write, I’m going to keep this newsletter short and sweet, and also not commit to any set schedule. My goal will be to send this monthly(ish) and I’ll likely include 2-3 key topics in each issue. And I’m not going to have any sponsors or have any paid plans, free to write what I want, and free for anyone to read.

This week was a big week for the AI world, and more than that, a big step forward for the AI powered Engineering world. I was so excited about some of the things OpenAI shared about Codex on Monday that I wrote a little Medium article about it, you can give it read here if you’d like:

I am incredibly excited about the shift we’re seeing in the engineering world. This week at OpenAI’s Dev Day, the company announced that over 90% of their engineers now use an AI powered Engineering workflow, which means an agentic coding assistant and agentic PR review. This is no longer the future, it’s the now.

And with that, I’m back. Thank you to my readers who wrote in and told me to keep going, and special thanks to Brett for giving me the final nudge I needed - welcome to Issue #12, let’s rock 🤘

OpenAI is overtaking Anthropic in coding goodness

Lately I’ve been noticing a trend on X, more and more of my dev buddies are talking about Codex, and I’m seeing people even going as far as announcing they’re moving from Claude 4.5 Sonnet over to Codex.

Rewind a few months ago and it felt like it was Gemini 2.5 Max and Claude 4 Sonnet duking it out for coding perfection, and OpenAI, while solid, just wasn’t in the same conversations.

Then OpenAI released ChatGPT-5 and with it GPT-5 Codex and well, they’re definitely in the conversation now, and starting to dominate it from what I can see. But it looks like it’s not just me noticing it because when I was drinking my coffee this morning, reading The Information, I saw this headline ⬇️

In the article, The Information confirms what I’ve been seeing with some pretty interesting data from Modu.

“New data show OpenAI’s Codex coding assistant has pulled ahead of Anthropic’s Claude Code assistant in certain coding capabilities. Codex usage among developers is also catching up to Claude Code’s.

Developers approved 74.3% of code written by Codex, slightly higher than Claude Code’s 73.7% success rate in terms of approving code pull requests, according to data from more than 300,000 pull requests collected by startup Modu. (Pull requests are proposed changes to code.)“ (Source - The Information)

Of course, I don’t think this means developers are going to stop using Anthropic models for coding, Claude 4.5 Sonnet absolutely cooks. What I do think is happening is that OpenAI clearly has put more focus on moving into the coding space, and well, they’ve got some insanely smart people there that are making it happen.

What I think we’ll see evolve over the next few months is greater specificity, and by this I mean, more specific use-cases for each model. We might find that Python devs end up really jamming with Claude 4.5 Sonnet, but Codex can output the best damn Typescript anyone has ever seen.

Personally, I think this competition is great for developers, the stakes are high, companies are really being pushed to innovate, and engineers win. Oh and we’re still in early innings here so we’re far from being able to crown a champion. Let the games continue.

I’m digging Zed, a Cursor(ish) competitor

Okay, so now let’s talk about Zed. If you haven’t heard of this new AI-powered IDE, then I’m excited to be the person to introduce you to it. Part of my goal with this newsletter from day one was to share new tools with fellow engineers that might not have been on their radar before.

So what’s so cool about Zed? I feel like it’s a bit more bare bones, streamlined version of Cursor in many ways, while of course not offering every little bell and whistle that Cursor offers.

Zed is written in Rust so it’s crazy fast, and it has a supercool multiplayer coding feature that works a lot like Google Docs where multiple people can edit the same file. This is pretty rad for remote teams, and beats pair programming over Zoom where typically one person has a file open and drives.

F22 Labs wrote a great article comparing Zed and Cursor, so if you want to really do a deep dive, I would give it a read. Or, if you’re too lazy to read the article but just want one sentence that might just be the biggest nugget in there, here it is:

Why I think every code review should have agents in the mix

Okay, so in past issues I’ve shared my love for Graphite, which IMHO is the most badass solution out there for agentic code reviews. And no, I have no sponsors so nobody is paying me to say that, we use Graphite, my team loves it, and that’s about it.

But what I want to talk about right now is why I think no matter which tool you use, why I think you should be using some agentic code review tool, period.

Let’s go back a year ago. Someone on your team submits a PR, and now a software engineer on the team, jumps in and reviews it. And let’s just assume that software engineer is a freakin’ genius, I mean they’ll catch just about anything. They find a few issues, submit feedback, boom - better code gets to Production.

But here’s the thing. Even your best software engineers are only human, and humans not only make mistakes, they also can’t possibly stay on top of every single new library that comes out, deprecated method, etc. An LLM, especially an LLM fine-tuned specifically for coding, and optimized for your codebase, can and will catch things a human won’t.

And that’s why I say that I think every code review should have agents in the mix, and I don’t say, should only be done by agents. Like OpenAI announced this week - “nearly all PRs are now reviewed by Codex.”

OpenAI didn’t say no humans review PRs, they just said an agent is in the mix, and not surprisingly, their own agent. As for what agent you decide to use, that’s up to you, but I think there are enough companies out there making agentic code review solutions that you don’t have to put together something custom, you can, and probably should use a tool built by a team, that all day every day just focuses on building an agentic code review tool.

Given how quickly this space has moved in the last year, I do think it’s safe to say, a year from now we’re all going to feel like dinosaurs for living in a world where humans did all the code reviews and had no help from a little all-knowing agentic buddy.

Okay that’s a wrap for this issue - thanks for reading and as a quick reminder - I have no sponsors, no eBook to sell you, no course, no podcast, and no set schedule.

If you like my newsletter, the most I can ask is that you share it with others so they can benefit from it. And if you want to give me a follow on X @morganlinton, I often RT nuggets like you’ll find in here, along with likely way too many photos of Lake Tahoe because I can’t help myself.

Thanks for reading and until next time!

Secret Bonus Video 👀

There’s a video I watched this week that is just too darn good not to share. Lee from Cursor gave a stellar talk titled: Context Engineering & Coding Agents with Cursor. You can watch it on X here, go Lee!