Straight up vibe coding is a horrible idea, but I’ll happily take tools to reduce mundane tasks.

The project I’m currently working on leans on Temporal for durable execution. We define the activities and workflows in protobufs and utilize codegen for all the boring boiler plate stuff. The project hasa number of http endpoints that are again defined in protos, along with their inputs and outputs. Again, lots of code gen. Is code gen making me less creative or degrading my skills? I don’t think so. It sure makes the output more consistent and reduces the opportunity for errors.

If I engage gen AI during development, which isn’t very often, my prompts are very targeted and the scope is narrow. However, I’ve found that gen AI is great for writing and modifying tests and with a little prompting you can get pretty solid unit test coverage for a verity of different scenarios. In the case of the software I write at work the creativity is in the actual code and the unit tests are often pretty repetitive (happy path, bad input 1…n, no result, mock an error at this step, etc). Once you know how to do that there’s no reason not to offload it IMO.

Great read, with some amusing asides.

Shots fired!

We have a test environment but it’s a hot mess. All the data is made up and extremely low quality. All the things you would normally interface with are also in test, but getting other teams to coordinate testing in the test space is… a chore. We do the best we can with mock services, but without having real services or representative data some of the data pattern assumptions don’t play out. Leaders value writing code and our lack of architects that span teams mean that when team architects either don’t meet ahead of time, make assumptions, or don’t ever agree on a design then…

We always host UAT. We also track logins. Guess how many users even show up for UAT, let alone actually click on anything.

This is why the vast majority of our testing happens in prod when our users are doing real work.

Sorry for the baby rant :)

This can also be one of the frustrating parts of open source.

Find something you don’t like? Fix it. Will the repo owner approve your pull request? Who knows. Maybe they’re a bit absentee. Maybe they view the original behavior as working as designed. Maybe your design doesn’t fit their architectural model, so they’ll (eventually) heavily refactor your changes and merge them in.

You can always stand up a fork, but keeping those two at feature parity and going in the same general direction can become harder and harder with time.

That’s not to say not to try! But it also means reaching out to the repo owners/maintainers before making your first change.

If it snows “bad enough” this is very common. The roads are going to be hard to drive on, which leads to more accidents and just puts first responders in jeopardy. The day before a big snow the grocery stores are usually slammed.

I would love to teach kids. I take any opportunity I can to mentor people at work, but in a school setting would be super cool. The pay cut relative to what I make now just isn’t worth it. Maybe in another decade when we have enough set aside…

Most OEMs usually show an update screen on their radio, even if something unrelated is being updated.

If the update is taking a long time it could be a really big file on a SOC. It could also be a smaller file being written to… very slow internal memory because when the part was sourced 8 years ago no one considered including memory read/write speed in the sourcing documentation. I’m betting the second, unless this OEM didn’t include background programming on SOCs, which is kind of foolish given how much easier it is on a SOC than MCU.

I can’t speak for this particular OEM, but 12 volt lead acid batteries don’t have very deep power reserves. The OEM choosing to leave the battery on during programming is likely a method of ensuring there’s enough juice to install the update and start the car on the next attempt.

It’s a mix of piece coat optimization and a lot of creep in what used to be a pretty lightweight process throwing it into the ditch.

The things that run software in cars largely fall into one of two camps: MCUs and SOCs. Think Arduinos and Raspberry PIs. Background programming, with an active and inactive partition, is absolutely possible on a SOC. They’re even file based, so you can do all kinds of clever things. Cars tend to not have many SOCs, so it’s not a monumental task to pitch having them each coat a little bit more for extra storage/processing. The biggest hurdles here are automotive grade and the very long development cycles. These both mean that the hardware is 3+ years old when it launches.

MCUs tend to have monolithic software builds (think literally everything gets compiled into a single .exe). There are a million billion of these things in a typical vehicle from most automotive OEMs. It’s… very hard to make them all have more capacity because you would take that cost and multiply it by 40 or so to get all the MCUs on a vehicle ‘upgraded’ for extra capacity.

If this all sounds a little crazy, it is. From two angles. First: do we really need as much software control in cars as we do? Marketing departments seem to think so. Second: the reason why there are so many small compute units in a car is the slow migration from mechanically controlled components to electrically controlled on. Back in the 80s the majory of automatic transmissions shifted based on a very complex mechanical system (look up a transmission valve body if you’re curious). Moving that to electronic control meant adding a computer to control that functional. Now take this and multiply it and you’ll kind of see the wreck in motion. Most OEMs are moving toward more centralized compute (fewer, larger, and smarter control units), but new electrical architectures take a lot of time/effort so it’s slow going.