AI in My Workflow: Personal Thoughts

A Word on AI Workflows

You guys know me, I hate AI slop. As in things made entirely by AI that carry no soul.

That is a messy definition, I know, but let’s not focus on that.

Today I wanted to make a quick and “low effort” post on the use of LLMs in my day-to-day job, and what my opinion is on that, like I’m an self proclaimed expert or some shit.

I am also preparing another post that will tackle larger aspects of these changes, including macroeconomic changes as well as the social changes this may provoke.

Ok, let’s get started with this small subjective post.

The Story Behind This Post

Today I had an uncanny experience. I am working on a pretty advanced project where I use my own HOLY CORE as a flight controller for a homemade drone (even the ESC is designed by myself).

It’s an ego trip project, so I can flex on the internet like we (the nerd content creators) always do, so we can feel smart with yet another “I MADE A insert goofy ahh project name. AND IT BOROKE ME.” video or whatever.

Anyway.

Because the HOLY CORE was 100% HANDWRITTEN (perhaps some boring ahh files were done by AI but the logic is indeed 100% handwritten), it has bugs… Duh. It’s a hobbyist project.

I spotted one recently and wanted to fix it.

PROBLEM: the bug was in the data cache subsystem!

For those who don’t know, the data cache is NOT A TRIVIAL MODULE TO WRITE, let alone if you don’t vibecode it like ppl do nowadays. From experience, picking up the module and fixing a bug like that takes ages (as in half a day to 2 full days of work!!!)

So I tried my luck and asked Claude.

Guess what: it found bugs and I even asked it to improve my testbench, which it did! And successfully… in 15 minutes.

15 MINUTES. Now I knew vibecoding was a real thing. I even vibe coded projects myself in embedded systems for some business prototypes we are working on. But these projects (like the HOLY CORE) were THE PROJECTS where I did not like to use AI because… Well, my EGO did not want to let a dumb AI get the credit for MY MAIN LIFE PROJECT from which I learnt so many things. But also because I somehow convinced myself AI was too dumb for that. That it did not have enough context, that the timing / signals requirements are too complicated for an LLM, etc.

Yet, it successfully did it! And I got excited, as well as a lil scared. So I left this note in the testbench code:

# =============================================================================
# TEST 6 (added oct 2026): Asked AI to pick up some gaps in the test and it did find some,
# this test is a summary of those gaps tested.
# I shall specify this code is not handwritten and heavily vide coded
#
# The motivation behind this is that I want to tidy up the cache bugs which
# was a very difficult module to write, harder than the core tbh when you are learning
# And it takes a looot of time to fix bug by hands. Becaus I was working on the drone project
# I needed a quick way to fix a simple bug and more importantly check the module againt more test cases.
# Tbh this AI stuff is so good that I ffel like I won't ever have to write a cache again.
# At this point, I'm just thankful I ever got t learn all that, but today, it feels unrelevant to write stuff by hand
# especially when, like me, you moved on a while ago from a specific piece of code
# and would take 1-2 days to pik it back up and fix the bugs, days which I don't have
# Anyway...
# =============================================================================

It summarizes pretty well my feeling about this, which is that we get to focus on higher level stuff, which is exciting! But at the same time it’s weird because this used to be hard.

The Impact on my Work

Personally, I don’t use Claude Code or “agents”, unless it’s a project where I actually need to prototype fast and where requirements are not rock hard / very complex.

Because of ego and also because I like to understand and still manage the architecture myself, which can be 100% taken care of by AI today btw. But I chose not to.

Also I use multiple software tools (MATLAB, Vivado, GTKWave) which the agents can’t use. So I stick to the web interface for when I need a full driver written in 2 minutes (usually takes DAYS lol), some HAL to run on the HOLY CORE for a specific IP CORE on the FPGA, or if I have a question, read through 40 pages of datasheets in 2 minutes,… yadda yadda.

This allows me to focus on mid-low level work, still have some fun and not have to do the “dirty work” that takes ages.

For now, this is smooth sailing, this method proved very productive, gives me plenty of control, I already did it before by hand so I kinda know what’s going on, and nobody at my job complained about my throughput.

And more importantly, I keep learning! Well, not low level stuff of course, but still! Whereas prompting an agent feels very… Well, lazy. Which I absolutely hate.

Impact on Future Work

Not digging into the macroeconomics impact of AI in this post, I’d rather like to talk about something very subjective that everyone likes to give their opinion about: “what does that mean for our work ? (or workflow)”.

AKA : “I’m dev, what are we gonna do now ?”

Well, my take is this: it’s scary. Because AI will be taking over more and more of the low level stuff. At that pace, we can imagine a future where the only things one theoretically needs to make a project happen are simply capital and a list of requirements.

You first need capital because you need to buy the things AI will tell you to buy in order to get the project done (production means, materials, etc…). And a list of requirements like “Hey I wanna build a cigarette machine that builds cigarettes super fast” or whatever your project is.

This creates multiple problems, but, in the scope of this post, we’ll focus on the fact that it sucks because we don’t get to think anymore. There is no more room for the slow human thinking, which used to be very fun. I love thinking and at this point I’m just thankful I got a couple of years to do so in my pro career in FPGA, control systems and embedded systems in general.

Yeah, we are at the point where I see each year of work as a bonus year right now, one more year where I get to work in a stimulating environment. But the thing is… These years are (probably) running out. There may be like what? 1 or 2 years left before thinking is too slow and not economically viable anymore for a business?

And another thing to conclude my opinion on the work aspect of this much bigger issue is that we will give more control to AI. As nobody will think or check the low level stuff anymore (simply beacause it will be impossible to keep up), any dumbass can come up with a project idea in the form of a PDF FILE, give it to the AI, follow its instructions on what to buy (or better, give it his credit card directly, bruh) and let AI build the product, and the dude just has to put it on the market.

Apart from the UNHINGED capitalism solely limited by energy availability (which I’ll talk about in another post), we get a scenario where AI could create dangerous systems without us noticing.

Now I HATE (not really hate but you get the point) content creators whose only business model is to SCARE people to get views. Be it in politics or tech, this kinda pisses me off. Which is pretty much what all media do but anyway.

BUT if AI gets corrupted, or “”"”bad sentient”””””, or simply loses track of what is considered “safe” or “good/bad” when reading the requirements of a project (e.g. executing a biohazard-inducing project or whatever), it could be VERY VERY BAD because nobody will check / see it until it is too late.

conclusion

That’s all I had to say (on the work aspect) !

The main takeway here is that we all shall try to keep some pride and try to keep on thinking ! If you stop thinking, you stop existing (just as a casual reminder).

And, if tomorrow thinking becomes to slow, do it anyway ! Just as another reminder, the goal in life in not to “create shareholder value”. First things first, not acting like I hold the secret to life or anything, but I would guess that actually living and experiencing / thinking are definitly the basics which are on anyone’s bucket list. So fuck the job if it implies us not thinkin you know what I’m saying ? I’d rather stay unemployed, which counterintuitively, may be more stimulating in a few years because you may get some time off the “press enter” (which I’ll call “doom vibe-coding” or just “doom-coding” from now on).

Godspeed, lads.

-BRH