Hello and welcome to the tenth issue of my newsletter, and yes - the third name change. Picking a name is tricky and for those of you who have been reading since the beginning, you know I’ve had a hard time picking something that sticks.

The reality is, while I’m exciting about vibe coding, and the impact it’s having by enabling an entirely new generation of builders, I don’t want to just write about vibe coding. As a CTO, I’m finding that AI is transforming our organization, but not through vibe coding, but instead by using AI coding tools like Cursor and Windsurf, exploring AI code reviews with Graphite, and using o3 and Gemini 2.5 to automate some super-manual stuff that LLMs can do so much faster.

Also, as someone who went to school for Computer Engineering and Computer Science, I love coding and think that everyone should learn to code. I also think that’s what makes vibe coding so powerful. People who don’t know how to code, are building software, and in the process, they’re learning how to code. I think The Primeagen says it best:

When I think about how AI is changing the engineering world, it’s not just vibe coding, it’s really a foundational shift in how software is made, and it includes everyone. I realized, what’s really happening here is a shift to AI Powered Engineering.

And so the new name was born, and here we are, on the tenth issue of my newsletter, third name change, and you know what they say, third time’s a charm! As I write this I have 291 subscribers, my goal for this year was to hit 550 subscribers so I’m now officially more than halfway there.

This was a very special week for me and a lot of other people as we partied the night away with the Bolt.new crew and The Chainsmokers, it might just be the best party SF has ever seen. I put together a little article on Medium about the experience so if you want to hear more about it feel free to give it a read.

Oh and one more change. I got rid of paid subscriptions. The reality is, I’m super busy and I don’t have the bandwidth to write two version of this newsletter, also I don’t really want to write this to make money, I’m just excited about how AI is changing the engineering world so I just want to write about it 🤷‍♂️ So it’s free for all, and it’s staying that way. No more tiers, we’re back to being one big happy family.

As for cadence, my bandwidth is so limited these days that I likely won’t be able to write twice a month, so right now I’d say every 3-4 weeks is probably where things will land. As a startup founder, I works days, nights, and weekends, and while I love writing this newsletter, I have to be realistic about what I can actually do!

Okay, phew - that was a long start, but a lot to cover. Now let’s get to the good stuff, welcome to Issue #10, let’s rock 🤘

What LLMs to devs like the most for coding?

It’s a question that has a constantly changing answer as the foundational modeling companies continue to release new models at an insane pace. Still, I feel like now we’re at a time where there’s some stability and many software engineers have picked their favorite models to build with.

Windsurf asked their users this question and shared the results on X, and I thought there were some pretty good nuggets in here that you’d all enjoy.

While I encourage you to read the thread, if you just want to cut to the chase - here’s the two key nuggets from the thread.

Best model for coding - Gemini 2.5 Pro

Best model for debugging - Claude 3.5 Sonnet

And yeah, Claude 3.7 Sonnet looks like it’s a bust so far, it’s rare I come across anyone who hasn’t had a problem with this one. That being said, I do think that both people, and companies, learn the most from failure so I wouldn’t be surprised if Claude 3.7’s shortcomings end up becoming Anthropic’s super power when they release Claude 4…which will be the next version number right?

You can have AI improve your AI

One of the AI coding tools that is seeing a lot of momentum right now is Devin.ai - and for good reason, it’s come a long way and is now becoming a real, contributing member of so many engineering organizations.

Claire Vo, the CPTO at Launch Darkly, and creator of ChatPRD, now has her own podcast called “How I AI” where she hosts interviews about, well, not shockingly - how people use AI.

Claire’s first episode is with Sahil Lavingia, the CEO of Gumroad who shares, in 60 seconds, how he leverages Devin to improve a prompt and add some filtering to upgrade the Gumroad Slackbot.

This is a pretty cool one as it highlights some of the new features that Devin has shipped that I’m guessing many of you reading this right now might not be aware of.

You can click on the image above to view the tweet, so stop reading and start watching! Thanks to Claire and Sahil for sharing this nugget, Devin is getting crazy good, this was new to me so I’m really glad you shared it.

Building MVPs in Bolt, the right way

When it comes to building product MVPs, I think Bolt is probably the most powerful tool out there. At the same time, like just about every tool, using it wrong means that you won’t get to really see what it’s capable of.

Prajwal has been a very active member of the Bolt community and I would definitely call a Bolt power user…which means he’s tried a lot of things and learned the hard way what works and what doesn’t.

This tweet that Prajwal shared below is absolute gold, and that one tip towards the middle, “lock important files to prevent overwrites” is one that has saved my butt more than a few times, so if you’re not doing that now, you might want to add that to your workflow.

Oh and if you haven’t used Discuss Mode yet, I think this is one of the most useful features Bolt as added this year. I also think this is where AI powered engineering is going. It’s not just about writing prompts that generate or debug code, it’s also about talking with AI to come up with ideas and plan architecture decisions.

Reinforcement fine-tuning now available with o4-mini

Okay, this is something I personally am pretty excited about and I think might have flown under-the-radar so I’m guessing a lot of people didn’t even know it happened. If you don’t know what reinforcement fine-tuning is, that’s okay - the OpenAI Developers account on X shared a pretty solid video of it in action, along with a little roundtable discussion with the OpenAI crew about it.

If you’re too lazy to click the link and ready about it on X, here’s a one sentence description on what reinforcement fine-tuning is:

So if that peaked your interest, and you want to go deeper, here’s the tweet ⬇️

Did you know Windsurf has a memory feature?

It feels like pretty much daily now, I come across something on X, that teaches me something about a tool that I use, that changes the way that I use it. And a few weeks ago Kieran did just that when he shared Windsurf’s memory feature which I’ll admit, I had no idea existed!

Here’s the dets:

I would highly recommend watching the video in this tweet as it shows this feature in more detail as it’s not completely intuitive to use out of the gate. Towards the end of the video you’ll also see where you can access the memories that Windsurf saved, which yes - has it’s own UX within the IDE.

Making memories in Windsurf is pretty powerful because you can essentially have Cascade do research, learn something, and then refer to that learning as much as you want in the future…it’s kinda like making Cascade smarter over time.

How to make a font with Chat GPT 4o

Okay, this was just too cool not to include, and yes, like the title says, it’s a complete walkthrough on how to make a font with ChatGPT 4o.

Not much more for me to say here as dudu does a great job walking through the process from start to finish in this thread. So…if you’ve ever wanted to build a font, no more excuses, make one!

Complete step-by-step guide to building AI Agents with n8n

AI Agents are incredibly powerful, and while you’re very likely using AI Agents yourself in your development workflow, you might not be building them yourself…yet, but that can, and should change.

If you haven’t heard of n8n, get ready to have your mind blown because it’s definitely one of the coolest platforms out there for building AI Agents. And Alvaro put together a really solid tweet walking you through the entire process step-by-step.

Massively underutilized trick when coding a frontend from scratch

Okay, just like that Windsurf memory trick I shared above, this is another great example of something I read on X that changed the way I use the tools I use daily. And while it may seem like a little thing, it actually makes a big impact. I don’t want to say much more since this is a pretty quick, but powerful tip, so I’ll hand it off to Matt for the big reveal ⬇️

Okay well I’m now over 1,600 words and I set my goal for these to be ~1,500 words so I’ll leave it there for now. Thanks to everyone who has been reading and supporting my little newsletter, it’s growing, slowly but surely and I’m just going to keep writing.

Now back to your regularly scheduled Sunday morning. Until next time!