Jim Gardner

I'm giving up on programming related professional development in the evenings. I've started to learn Mandarin, instead.

Yes, this is driven by disillusionment with the industry, in large part thanks to AI. Professional development in evenings feels pointless. For 6 years coding has been my main special interest, and because I discovered programming in my 30s, I've also felt a pressure to catch up, learn, fill in gaps.

I like the process of writing code. I like the literal typing out of the words and symbols. I enjoy 'thinking by doing'. It was the same when I did lots of observational drawing as an animation student. There's a connection between mind and fingers that stimulates my brain like crazy.

That specific feeling of creating something is NOT the same as iterating with an agent or plain old LLM. I'm not saying using them makes you stupid, and I like what @grimalkina@mastodon.social has been doing with her Claude skills in that respect (you can listen to her [podcast ep](https://www.changetechnically.fyi/2396236/episodes/18692591-you-can-learn-with-ai) about it), I'm just saying they're qualitatively different experiences for me.

I don't enjoy coding with agents very much. It's kind of hollow. I'm still in charge and have to do most of the planning, yes, so I'm still thinking about stuff. But again, it's not the same kind of thinking. It's missing the part that relies on the doing.

The other thing is, and this may get me crucified because I'm saying it aloud on the fediverse, but, models are way, way, way better than they were even 18 months ago. They were pathetically bad. A waste of time. Now, if you work in Python and all you need is something a mid-level developer could understand and work with, an agent in your IDE using one of the latest models can probably do it (with lots of guidance).

I also think it's also reasonable that the outcomes of using them will get even better, if for no other reason than clever people will improve how we interact with them and add extra tooling about important things, like correctness of code outputs.

I'm a very valuable programmer to have in a team, but I'll only ever be mid at the actual programming. So all told, the magic is gone and I no longer feel an impetus to keep up.

Maybe one day I can be mid at speaking Mandarin, as well as programming.

Edging, but it’s the first step of doing jigsaw puzzle

Doing a short course on making AI agents, which of course means I have to pay Google for API usage.

My soul is screaming in agony because Google, Anthropic, OpenAI, they're all fucking shit, and I hate everything surrounding this technology. I think it is incredibly unethical.

But if I don't engage I'll struggle more and more, professionally, and I honestly don't know how I'd make a living doing anything else at this point in life.

Argh.

Was doing some professional development which required me to get set up with Google AI Studio. I've just seen that the Gmail of that very same Google account... put an email from AI Studio into the spam folder 🙃

Music I loved aged 10:

- The Spice Girls
- Pizzicato Five

https://youtu.be/t4Jo26CQfR0

Pet hate in the workplace as an autistic person:

Being given verbal instructions of some new work to do, then told to go write those verbal instructions as Jira tickets.

It’s the wrong way around.

If you can already describe what you want through conversation… YOU are ready to write the tickets. Before we talk.

THEN you bring the tickets to me, so that I can ask the 200 clarifying questions that will turn the tickets into actionable pieces of work.

Next time you’re reading a LinkedIn post, imagine the author getting their head flushed down the toilet every time there’s a blank line.

@adam I think in this case it's not laziness, rather that they want to provide autonomy. But it's related to autism for sure because of the needs for

1. Clarity. Vagueness is frustrating and most people (including myself) are often at our vaguest when speaking aloud

2. Knowing what's going to happen in advance/context. Another thing we're all kinda bad for doing to others, TBH. In this case, if "we're going to write tickets" actually means "you're going to write tickets after we talk", those are two different things and I would prepare for them accordingly

3. Externalised information. We should either be given some written info to start from, or the time to do the writing ourselves during the conversation

MCP = Manky Computer Perversion

Man... LLMs keep producing Python code that uses `dict.get("key_name")` all over the place instead of `dict["key_name"]`. I've even written in copilot-instructions.md to strongly prefer the latter, so that key errors get thrown if the expected key isn't in a dict... and they only follow this behaviour sometimes. The result is that I can run something and be like, oh, that actually got to the end! Then I check the output and see missing or placeholder values all over.

#NineSols is awesome. It's rekindling my long-standing desire to learn Mandarin

People are now also unironically selling their programming language/package/framework by saying “Every function and class is documented with lots of examples! Perfect as context for agents!”

Dudes. Do you know who else might benefit from all those examples?

Is it really the widespread use of fucking chat bots that’s going to drag documentation kicking and screaming into the modern era??*

* My outrage here is a little bit OTT. There’s pretty good documentation in the R package ecosystem, Pandas has got pretty decent, Polars is good, etc., but it’s still an issue. https://chaos.social/@frederic/116453183275715578

“TAKING A SHIT IS DEFINITELY THE WAY FORWARD”

Graffiti from the animation department toilets at Edinburgh College of Art, mid-2000s