Spent this morning at Minnebar. This year was the 20th anniversary. I’ve had several posts about Minnebar on this blog through the years. I’ve only been able to go a couple times over the past ten years and it was good to see many familiar faces. (Even some Minnesota expats. 👋, Luke!)

(Oh, and I left my conference shirt in a room at some point and it was no longer there when I returned. If you have no conference shirt can you even say you were there?)

I called TRUNCATE on a Postgres database table in production for the first time. That was nerve wracking!

I did some wood work for the first time in a while. I’m working on a loft for Emma. I don’t know whether it’s time away from woodworking or just the type of project, but this felt a lot more like work and a lot less like hobby. Part of it might also be because it’s such a physically big project and that is a lot for my older body.

I’m way behind, but I just learned that Songkick was purchased by Suno. Did I know what Suno was before I learned that? No, but it sounds pretty ick. From their own home page:

Make any song you can imagine
Start with a simple prompt or dive into our pro editing tools, your next track is just a step away.

Yeah, ick. Time to migrate to Bandsintown? Something else? Give up?

I just finished the excellent Fela Kuti: Fear No Man podcast by Jad Abumrad and team. Highly recommended.

I think I might become an Onion paper subscriber. Anyone else going back to paper?

Discovered NOT A HOTEL. A man can dream.

Since building Pika I’ve wanted to incorporate a “now and then” sort of now page with dynamic history. With the recent flurry of updates to Pika variables, I realized that my dream could now be made reality. It turns out that my recent habit of writing Friday posts is effectively me writing what I’m doing/seeing/thinking, well, now(ish).

So now my now page is no longer chronically stale! If you happen to write a “general state of things” post every week or month, you can do the same thing on your own Pika site.

The page is pretty simple. First I wrote “Now” as my title. Then in the page body:

{{ posts_in_stream tag: “Friday” with_excerpts: no limit: 1 }}

### Then

Here’s to now pages past:

{{ posts tag: “Friday” skip: 1 }}

I put the body text above in a code block and wrote it as Markdown to avoid variables rendering on this post here. You may or may not be able to copy-paste the above variables into your Pika editor.

And that’s it. You may notice that my “Then” pages are linked dates without any titles. AI helped me with some wild CSS to modify Pika’s default output to what you see there. If you want something similar, you’re welcome to look at the CSS in my now page’s source to get started. Just look for .post-link in the page source.

Ava is traveling and shared this from the Victoria and Albert Museum: 

A lengthy, imperfect article on how much Minnesota mattered this winter. (archive)

The challenge of this work week has been trying to figure out a way to streamline the moderating Pika blog posts. With only two of us on the team, it’s impractical for us to read every blog post and look for rules violations. Especially when you consider that there are many blogs on the platform with hundreds and thousands of posts. Choosing posts or blogs that we’d rather not include in the soon-to-return Pika Pulse is an even more subtle and nuanced task.

It seems obvious that LLMs could be engaged to help us with this, and in testing in my local development environment it is absolutely true that it would be a big help. We are still strong believers in not training AIs on Pika blogs. The terms published by APIs for these services are clear that when sending anything to them the default position is to not train on that content. You must opt-in.

Do we trust them? Not exactly. So we’re going to keep looking for alternatives that stay away from LLM cloud services. 🤔 I have a Mac Mini lying around here that could maybe run an LLM that has no access to these motherships…

This time of year I do a lot of working from my couch with basketball games on TV. It’s not nearly as compelling these days with NIL and players constantly transferring between schools, but old habits die hard.

I learned about Johnny.Decimal from Pika blogger Michael Borland. The last post on his blog clued me in. Sadly he has since passed away. RIP Michael. ❤

I’ve dipped in and out of Getting Things Done (GTD) many times over the years. The most organized I’ve felt in life is when I was truly adhering to the process, including physical tickler folders in a filing cabinet. A filing cabinet!

Today Johnny.Decimal came across my view, and I think I might give it a go. The current state I’m in is this:

• My work has been all-consuming, in a good way.

  • My work is not organized, but rather little note papers on the desk, a million Basecamp to-dos, and a general trust that the most important things will remain in my brain or come to the fore as necessary.

  • With the productivity power of LLMs, I’m putting more and more balls in the air every day.

  • While I know I can’t copy-paste a personal system onto our work to-dos, I hope Johnny.Decimal can help bridge that gap.

• My home projects are abandoned.

  • There are papers piling up on my office floor, awaiting organization.

  • My office is a mess.

  • I’m daily making small decisions about what is important to get done (pay the bills) and what isn’t (scan the documents).

  • I have notes and reminders spread across Things, Apple Reminders, Apple Notes, and Obsidian.

  • I’ve uncharacteristically missed some important tasks over the past year. Late estimated tax payments being one glaring example.

  • There are little projects to do, but my system has become disorganized and overwhelming. So any spare moments with energy I generally pour down the drain of internet distraction.

  • I have lots of woodworking ideas that I can’t work on now, but I suspect I’ll pick back up in the next decade.

At this point in life I’m not seeking to be an efficiency master. I’m well aware that I’ll never get everything done. I’m happy to sit and read a book or watch a movie. Yet I would like to feel that my ideas and projects are organized in a way that I can find (or be reminded of them) again later. I’ve had that feeling before and it was the first year after starting GTD.

I’m watching some introduction videos now, and I might actually pony up for the course to get my bearings. If only I could justify the time…

A couple months ago I put together a fairly lengthy post about getting an old MacBook set up to be a writing machine. It was pretty encouraging, the idea of having a machine that would just be for one thing. At the time of writing that post I was diving pretty deep into my work and I realized I was not in a “writing mode,” but I was also hopeful a little writing would happen with this new, old laptop.

I’m not going to blame it all on this one thing, but at the end of that post I said:

I hate the feeling of typing on this thing enough that I’m considering an inexpensive Keychron keyboard to sit on my lap while I write. That doesn’t seem conducive to actually writing, though.

I wrote once with that laptop and never went back.

So I’ve been on the lookout for a different option. My finger has hovered over the order buttons for both a Freewrite Traveler and a used MacBook One. With the Freewrite, I was hesitant about the lag between typing and the screen, and lets not talk about the price versus general utility. For the latter, I suspected I would dislike the keyboard as much as I dislike the keyboard on my old MacBook Air.

Then came the Neo. With the education price¹, I could get the Neo for less than the Freewrite Traveler. I don’t mind typing on my work MacBook Pro at all, so I’m pretty confident I’d be happy with the Neo keyboard. No reviews thus far have suggested keyboard problems.

So I decided to give it a shot, clicking the trade-in button for that MacBook Air in the process. I initially felt like embracing the low end and rolling with an Indigo 256 GB with no Touch ID. In the end, though, I figured I’d save my future self a few headaches with the 512 GB.

1. TIL at checkout: Education savings: Available to current and newly accepted college students and their parents, as well as faculty, staff, and homeschool teachers of all grade levels.

I really enjoyed Justin Duke’s thoughts in “The death of software, the A24 of software”.

A24 didn't succeed despite the streaming era — they succeeded because of it. The explosion of mediocre content created a vacuum for taste, for curation, for a brand that stood for something. When everything is abundant and most of it is forgettable, the scarce thing is discernment.

Yes!

We are fortunate at Good Enough to be given these powerful LLM tools at a perfect time in our careers. (I have more to say about that timing, but not today.) We can guide agents very successfully given our breadth of experience in programming and design. Perhaps more importantly we have also developed a deep level of taste for the way interfaces work, and honed our attention-to-detail muscles to a fine degree. And our belief in human (humane?) customer service runs completely counter to million/billion dollar companies that ship you to an algorithm and don’t want to talk to you person-to-person unless you embarrass them online (archive).

Household-name status isn’t what we’re after, so the A24 comparison isn’t really apt for us, but there is a lot of room in the marketplace of software products for those teams that are able to stand out based on heaps of quality. I just hope discovery tools continue evolving so that regular folks are able to find the excellent software that is going be made.

We mourn our craft:

The worst fact about these tools is that they work. They can write code better than you or I can, and if you don’t believe me, wait six months.

This point stuck out to me. I’m going to take it completely out of context to make my own point. I haven’t yet found an LLM that is writing code better than I can. I guess it depends on how you define “better.” It’s better in many ways:

• It can write code faster

• It can write code more effectively than I can in languages with which I’m unfamiliar

• At my descriptive prompting, it can quickly give me code design ideas that I hadn’t thought of yet

• It can be used by novices and non-programmers to build software that solves at least some of the target problem space

Yet when it comes to implementing a non-trivial feature in Ruby on Rails, I’ve never seen a LLM put together something that I felt was better than what I would put together. I’ve not seen it in a “Yeah, that’s how I would have done it” way nor a “Woah, why didn’t I think of that?” way. This is, of course, a subjective statement, but it seems like a lot of people are lamenting the loss of that subjective view of beautiful code. I haven’t found that loss to have arrived yet.

It’ll probably happen at some point soon, though. Or maybe, like most things programming, 80% of the work will be in delivering the last 20% of the potential of LLMs. For me, I’m thankful that I happen to be one of those programmers who loves the higher-level building of creative solutions more than the act of typing lines of code. On most days I find the process of leading the LLM to the code I think is best to be a pretty satisfying experience.

Weekly Thing 341 brings up AI fatigue:

With that though our ability to do more fills with more things that we wished we could do. No matter what, there is still only so much time and energy in the day. The fact that Claude is there at 2am while you cannot sleep can be a problem. The fact that you can have five projects going on with different agents is neat, but you still are coordinating them!

The "just one more prompt" trap is real.

It sure is!

With a small team at Good Enough, there is way more work (and ideas) than there is time. It's so tempting to fire up claude.ai/code (literally in a web browser on my phone) and send it ten different tasks that need to be done. In fact, I have sent it ten different tasks…times two or three. I sent these tasks almost exclusively between midnight and two in the morning on nights when I couldn’t sleep.

When I woke up the next morning I got that familiar feeling like I’m managing a clutch of junior developers. I’m spending time reviewing some mediocre and incomplete code. Largely this has been a beneficial process. Even though none of this LLM-produced code has been deployable, the process has unlocked my mind on a few different long-standing to-dos.

After twenty draft pull requests were built up in late January, I decided to cancel my Claude subscription. Maybe after I catch up on finalizing each of those pull requests, I’ll open up the claude.ai/code spigot for another month.

Don’t worry, it’s not as if I’m using no LLMs for development. I have the growing Windsurf bills to prove it.