7 min read

February 2025 bakery

February 2025 bakery

I've become persuaded by the idea that we are the product on most social media apps. The thoughts and links that I would have shared on these hellsites, I now aggregate in a monthly post I call the "Bakery."

Half baked

Some written thoughts, but not enough for their own post.

Cursor vs. Windsurf

Cursor and Windsurf are two IDEs I've started using. Unlike GitHub Copilot, which feels like clunky add on to VScode, AI is a first class citizen in these editors.

Cursor is more popular than Windsurf from what I can tell, but I started using Windsurf first.

What I like about Windsurf is that it is really good out of the box. I didn't have to set much up at all. It also can run commands like creating files and running tests. It is remarkable.

I ran out of credits on Windsurf, which says a lot about my usage.

Without access to Windsurf's best LLM models, I decided to try Cursor. Unlike Windsurf, in my experience Cursor does take some setting up. You need to enable "Yolo mode" (this is real), otherwise the IDE will not run commands or edit your code. You also need to setup rules, which are markdown files that help Cursor align it's prompts to the LLM model with your preferences. Once I got Cursor set up, I was equally shocked and impressed.

I currently prefer Windsurf, but that could be because I started with it or that my Cursor setup needs additional revision. The differences are pretty marginal.

I feel like the barrier to me creating software has fallen dramatically. There is a lot more to say about how I'm specifically using these tools. I have thoughts.

I also hopefully will announce the edtech project I've been working on. So stay tuned or reach out if this is interesting.


This month I started using Rosebud AI, an AI journaling tool.

It is kind of like a therapist. It asks you to reflect on your day or set goals. There are lots of pre loaded prompts and the AI can just start generating prompts based on your previous journals. You can also create your own journal templates, like a monthly review or help me get unstuck journal.

After you respond to a prompt you can choose to either finish the journal or go deeper. Going deeper results in the AI promoting you for further introspection. Sometimes these prompts are quite insightful.

I like this better than therapy for a few reasons. Yes there is a men will do x to avoid going to therapy joke here.

I always feel guilty talking to a therapist, like I'm either wasting their time or not getting enough for the cost. I also don't want a human therapists to know what a crazed lunatic I am, so I tend to downplay some of my more unhinged behaviors.

I have no such compunctions when chatting with AI.

I also like that I don't have to schedule time for my AI therapist journaling tool. I can just do it whenever I want.

The app is not very polished, but the ai functionality is impressive. I often have to force close the app. The speech to text telephone mode is not functional at all.

But the text produced by the AI is read out loud with an ai generated voice that is 100% convincing. I've not once detected an inflection that did not sound entirely human. The prompts do a good job tying in what the tool already knows about you. The ai is supportive and even helpful.

I recommend this app if you'd like to do some interactive journaling. I've found the process of reflection and goal setting clarifying.


My theme for this year is to make things. I want to ship. Most of this will happen at work and will therefore not be something I share on verynormal.info. But, I did spend about a half day putting together an app for tracking my Goose Jam of the Year bracket. I started creating this app with Claude by uploading the image of the bracket and prompting:

Can you make a small html and javascript tool that allows me to interact with this bracket? This bracket is for a band called Goose. Each entry in the bracket is a song the band played live. This is meant to emulate the march madness ncaa tournament bracket, but instead of landing on the best basket ball team, this gets us to the best Goose song performance (jam) of the year.

After a few feedback prompts I had a tool that worked. But I was hooked. I transferred the code into Windsurf, the AI fueled IDE I've been using this year.

From there through my prompting we got to a final product.

This was not my most familiar tech stack - Next.js and Typescript.

After implementing each feature via prompt, I would reuse a prompt for really good results. I'd ask something like:

Please review <files>. Please suggest ways we can improve the architecture, maintainability, and encapsulation.

The result of this prompt is often an improvement to code modularity which then makes it easier for the AI to make small changes to achieve the desired result.

Just as with non-AI fueled engineering, it's important to keep the change set small so the impact of the change can be known.

I had a lot of fun working on this code even though I barely wrote any of it. This begs the question, do I say I wrote this code? Or do I say I created it? Or is there another word we should be using.

I wonder if my professional software work will skew in the direction of product as these tools continue to improve and proliferate. I am starting to call myself a product engineer rather than a backend engineer or software developer. I like how the term focuses on the result and the means are secondary.

Raw

Naked links.

AI, data centers, and power economics, with Azeem Azhar
AI-fueled energy demands in data centers, the changing electricity generation mix which will support it, and more.

AI and energy infrastructure? YES PLEASE.

To avoid being replaced by LLMs, do what they can’t
What LLMs can’t do yet

A computer can never be held accountable therefore a computer must never make a management decision 🐐.

My LLM codegen workflow atm
A detailed walkthrough of my current workflow for using LLms to build software, from brainstorming through planning and execution.

I am not the only one using LLM's to build side projects.

Searls of Wisdom for January 2025
It’s taken me a while to figure out what to write for you that would tie a bow around January 2025. My take on the cultural and political realignment we seem to…

A great essay on legacy.

The “Look What I Can Get Away With” Era
Our Biggest Challenge This Decade: What to Ignore and What to Fight

I'm embarking on my shipping era.

punctuated equilibrium
2024 had hands. trauma dump. memoir. something like that.

This experience was a wake-up call for me — it forced me to think about what really mattered to me and what I wanted to do with my life. 

Be my next co-founder
Who I’m looking to partner with in my next venture

Ron, bubbe! Are we doing this? Call me.

On the Denver tech scene and Turing.

Episode #236: Simon Willison: Using LLMs for Python Development – The Real Python Podcast
What are the current large language model (LLM) tools you can use to develop Python? What prompting techniques and strategies produce better results? This week on the show, we speak with Simon Willison about his LLM research and his exploration of writing Python code with these rapidly evolving tools.

As you can see, I've started to binge Simon Willison. His LLM work is really interesting. Also, in this podcast he touches on the idea of POSSE - "Post (on your) Own Site, Syndicate Elsewhere"