Category: Thinking Machines (AI, ML)

Blog posts about AI, ML and adjacent technology.

  • Just Use AI Bro!

    An extra leg. A sixth or seventh finger. A cat wearing boxing gloves fighting a dog. A political interaction between an eastern and wester leader that seems just… a little too romantic and sexy to be real. Eager, annoying, affirming enthusiasm for your half baked idea you thought you’d research and ended up saying “yes, I am a genius. I knew it”.

    Coming across unconsented AI created content has started to elicit a response of anger, frustration, annoyance, and disappoitment in people who come across its slop in the corners of the internet. Or it leaves you patching up the holes made in your life after convincing you absolutely should – nay – MUST start that knitted plushie business after quitting your current job. Follow your dreams baby, we’ll make it real! Thankfully in my case, it often just leaves me refactoring/patching up code instead of poorly made plushies.

    I’m a software engineer. They call us “SWE”s in the biz these days. It’s a term I’ve heard specifically AI and big-tech CEOs use disproportionally more than other folks in the tech industry. Perhaps it’s yet another way of separating the actual level of skill and human ingenuity and intuition required for the job. It makes it easier to talk about us as a disposable, replacable “units of work capability” that can be replaced by the equivalent promise of an AI agent.

    Engineer is too human of a term to give to “SWE”s. Its connotation is too skillful. Heck, I think of an engineer, I’m thinking of someone who designs and calculates a bridge that spans a 400ft gorge, or a rocket builder, electric car designer… Not some monkey at a keyboard bashing out lines of nearly inevitably “buggy” code for a decent salary. It’s playing to my imposter syndrome. In reality, whether I like it or not, some of the stuff I build is mission critical for businesses, their compliance and safety. The decisions I make, how well I work with clients and the rest of the team, my discipline, intuition and experience surely gives a LITTLE credence to the term “engineer” right?

    Maybe it’s just my fragile, work-incompatible, messy, unpredicatble and expensive human ego talking.

    We get juicy speeches from the like of Anthropic’s CEO, and Open AI every 6 months spouting that in 6 months human “SWE”s will be unneeded. In fact we can probably just pack up our things now, because in the AI HQ, they tell us their own AI is already writing most its own code, so yeah. Why do I need to be paid a salary to get dorito crumbs all over the keyboard? Why even try?

    Remember the phrase “1000 monkeys at 1000 typewriters will eventually write the entire works of Shakespeare”? Infinite Monkey Theorem? That’s what comes to mind with the fabled “country of geniuses in a data centre” that one CEO likes to say on repeat. Hell of a lot easier to write Shakespeare if you’ve got say, ALL OF SHAKESPEARE THAT’S ALREADY BEEN WRITTEN available to you to copy. Also more possible if you’re harnessing the power of the sun, or maybe the heat of all humanity stewing in vats of life-supporting fluids living in a virtual “just imperfect enough to stop us for suspecting something’s up” world… They should make a movie about that.

    Ironically (look I NEVER know if I’m using that word right – would it be ironic if this is the correct use?), while getting mad at these people threatining the relevance of my being and value to the workforce, I’m probably “tabbing” away to AI auto-complete suggestions, or writing a new prompt for AI to go run with scissors and see what magic or mayhem it might pull off.

    There are studies that suggest that AI only makes you “feel” faster rather than actually increasing a SWE’s output – but by jeeves if I don’t FEEL like it’s made me faster at my job! I mean, luckily for me I have a pretty good benchmark. It’s a complicated formula:

    How much shit I get doneTime\frac{How~much~shit~I~get~done} {Time}

    There are at LEAST a dozen more subtle variables that have been simplified out of this formula to help non-engineers and non-SWEs understand it, but that’s pretty much it. If I run that through my calculator, it’s pretty clear that I’m… pretty much exactly as fast as before. But my feelings!

    Perhaps my prompt engineering is lacking. Perhaps I’m lacking in vibes. Maybe I should be using Cursor instead of CoPilot, or better yet be running my own LLM locally, the $10k+ investment will easily pay off in a year. I mean, the internet told me there’s this one guy that’s shipped like, 20 AI built apps in a month and is raking in $100k in revenue! Damn I should get on board that HYPE!

    Why the heck do I use AI then?

    Hold on to your Doritos because you may be surprised to hear it – I actually like using AI and I think it’s a fantastic tool. Didn’t I just say it doesn’t make me faster? YES. But what I didn’t include, is how much it’s made me BETTER. Here’s the summary of what AI has done for me:

    AI has helped me look at nearly every issue that I tackle through a different lens, and provides me with a multitude of ideas and patterns that either didn’t come naturally to me, or were outside of my current knowledge, or I was too lazy or under too much time pressure to research and learn by my own volition.

    Using AI over the last couple of years, my ability to write code and solve problems more thoroughly, with less technical debt within the same or increased time pressures, has increased.

    ⭐️⭐️⭐️⭐️/5

    So really, like the internet, cars, spatulas, screwdrivers, rocks, cats – AI, to me, is a tool to help me get my work done in a more efficient way. Efficiency being “what you get out vs what you put in”. And the REAL formula is more like:

    How much shit I get donelog(Quality of said shit)TimeNumber of refactors\frac{{How~much~shit~I~get~done} * log({Quality~of~said~shit})}{{Time}*{Number~of~refactors}}

    Now the mathematicians in here might say “hang on, doesn’t the term ‘Quality of said shit’ partly INCLUDE ‘Number of refactors’. Yes. Yes it does. An inverse relationship. # of refactors goes up, quality of said shit usually (but not always) goes down. I said it was complicated ok?

    What’s critical to me, is there’s an experienced human SWE behind the steering wheel, who knows how to drive. Sure there are self-driving cars, but what the heck’s a toddler going to do if the car suddenly decides to freak out and request human intervention? It might make for a hilarious Vince Vaughn movie moment, but you’d probably prefer an experienced driver to be there? They can see and interpret the meta-situation – seeing how the car got there, where it needs to be, the best way to get there, and will happily re-engage the autopilot once we’re back on track.

    I don’t know if I’ve made a point with this ramble, but I’ll put this through AI for polishing over my empty Doritos bag!