Stephen Meriwether’s Blog

Trying to learn something new is filled with all sorts of frustrations, but those small, infrequent aha! moments keep me going.

SNL’s new skit: Elon Musk Cold Open is chef’s kiss perfect.

Scarlet Ink is a newsletter by Dave Anderson, and here’s a quote from its About page:

Scarlet Ink educates both technology veterans and newcomers on how to grow their careers, build their leadership capabilities, and level up their interviewing skills.

I read a lot of newsletters and often find myself with more articles to read than time available and Scarlet Ink is often one that I choose to read later. That’s probably because I already spend so much time thinking about work I’d rather spend my free time reading about my hobbies or other interests. But the value to time-spent I get from his articles is incredibly high. Even better, the hit-rate for relevancy in my day is also high. That was again the case today when I read The Six Steps to Saying No — Why Being a Team Player is Not Necessarily Great. The relevant paragraph to me today was:

There’s a popular saying at tech companies. We say that our work is a marathon, not a sprint. That means that we’re out to create value over years, not over months. If you’re expected to do something which cannot be sustained for years, it’s a defect, and it should absolutely be temporary.

Here’s a parting quote that nearly perfectly describes me:

There’s a common illness in the workplace. It’s called being a “team player.” A team player does the extra work that is falling on the floor. The team player stays late to ensure that the weekly sprint completes on schedule. The team player comes in on Saturday to shepherd the new build. The team player also burns out and quits.

A new post by Simon Willison nearly perfectly captures my current thoughts on how to understand the shortcomings of using LLMs to write code.

Hallucinations in code are the least harmful hallucinations you can encounter from a model. The real risk from using LLMs for code is that they’ll make mistakes that aren’t instantly caught by the language compiler or interpreter. And these happen all the time!

One thing I’ve learned building Conductor is compilation errors are the easiest to fix. Most of the time all that’s needed is to give the LLM the error and it will fix itself. If that doesn’t work, I can prompt my way towards a fix. Self-healing or prompting solve 99% of all of these types of errors. Making sure that the code actually does what I intended is an entirely different thing.

Also from Simon:

A general rule for programming is that you should never trust any piece of code until you’ve seen it work with your own eye—or, even better, seen it fail and then fixed it.

This is increasingly less common with the proliferation of LLM-enabled code generation. We’re thinking of ways to “bake in” this process into the systems we’re building but it’s a really hard problem.

In his latest podcast, Bill Simmon’s talks to Susan Morrison about her book about Lorne Michaels, creator and producer of SNL. Conversations with people who are deeply passionate about the topics are my favorite to listen in too and this was that.

John Gruber, at Daring Fireball, is my favorite iPhone reviewer. His reviews are honest but ever so slightly skewed in Apple’s favor. As an Apple fan that’s what I want to read. I don’t want to read how an iPhone compares against a random Android phone because I would never buy that random Android phone and I’d never recommend anyone who asks me to buy it either. When I read an iPhone review I want to read about the iPhone and that’s what John does. So what’s his thoughts on the iPhone 16e, this sums it up best:

The iPhone 16e is an iPhone for people who don’t want to think much about their phone. But they do want an iPhone, not just any “whatever” phone. A just plain iPhone, with a good screen, good enough (and simple) camera, and great battery life. I think Apple nailed that with the iPhone 16e.

Another reason why I enjoy reading his reviews are these little tidbits:

(Also take note of a clever touch: Apple’s default wallpapers for each phone subtly suggest how many camera lenses they have.)

His reviews often have a least a couple and each of them make me smile.

While building Conductor we’ve often struggled to make sure we haven’t introduced regressions when changing system prompts, etc. Manually testing sort of works but we often don’t have the time to test the entire feature set. We have integration tests but they also don’t cover the entire feature set.

I think DeepEval is the solution. We’ve only recently started to implement DeepEval but the vision of the future is clear. DeepEval lets you “unit test” features of LLM-enabled software. It’s open-source and comes with a lot of useful tools built-in. It’s a little finicky to get started but the learning curve is not that steep.

On of my favorite things I’ve read in 2025 is: [My Father Was a Conservative Evangelical Paster. Than I Came Out](My Father Was a Conservative Evangelical Pastor. Then I Came Out.) published in the New York Times Opinion about a father’s journey of reconciling his beliefs against his reality, through his private journal entries.

What I found in my dad’s journals gave me a deep appreciation for what it takes for people to really change their minds — what it means to confront the limits of their faith, to risk their career, their standing in the community, even to question the foundations of all that they believed to be true.

Two things that stood out to me: the unfettered honesty of the journal entries and the process of truly confronting your beliefs.

While building Conductor we’ve often struggled to make sure we haven’t introduced regressions when changing system prompts, etc. Manually testing sort of works but we often don’t have the time to test the entire feature set. We have integration tests but they also don’t cover the entire feature set.

Enter DeepEval. We’ve only recently started to implement DeepEval but the vision of the future is clear. DeepEval lets you “unit test” features of LLM-enabled software. It’s open-source and comes with a lot of useful tools built-in. It’s a little finicky to get started but the learning curve is not that steep.

At the top of my list of favorite things I’ve read in 2025 is this piece published in the New York Times Opinion: [My Father Was a Conservative Evangelical Paster. Than I Came Out](My Father Was a Conservative Evangelical Pastor. Then I Came Out.).

What I found in my dad’s journals gave me a deep appreciation for what it takes for people to really change their minds — what it means to confront the limits of their faith, to risk their career, their standing in the community, even to question the foundations of all that they believed to be true.

The story takes us through a father’s journey of reconciling his beliefs against his reality, through his private journal entries. There were two things that stood out to me: the unfettered honesty of the journal entries and the process of truly confronting your beliefs.