At my last job, I had a co-worker who went from someone I really enjoyed conversations with to someone who could not stop talking about LLMs. Every single conversation about programming or architecture or about the company was ultimately filtered through how to optimize for LLMs, improve LLM output, or something related to LLMs. His fondness for moving from Jira to Linear wasn’t rooted in the fact that Linear is just plain more pleasant to use but rather entirely about its ability to give LLMs more context. In meetings he seemed to always find a way to say:

Context is king!

Of course, his usage was always related specifically to giving LLM-driven technologies more context, more text, and more everything. It wasn’t just about volume either, it was often specifically about how to craft what was being written for LLMs to get the most out of them.

The whole situation was really unfortunate. This co-worker is not dumb by any stretch. In fact, he was probably the single most intellectually intimidating person I dealt with the whole time I worked there—certainly more interesting than the mental midgets in the C-suite or anyone with a title of “vice-president.” However, he ate it up and ate it up and ate it up some more. He started to sound like another unfortunate co-worker.

I suspect, after the bubble bursts, we ask ourselves, “What now?” We’ll also have to ask, “How did this happen to otherwise intelligent people?” I’m sure plenty of us are already asking, but the conversation will be much broader as it always is the the wake of devastation. In this case I believe it was a combination of company pressure, the builtin engagement and gambling addiction, and I genuinely believe a lot of us who wrap our identities up in being “smart” love hearing just how smart and insightful we are.

God knows I do. (Another reason I avoid this stuff. Know thy self!)

LLMs as deployed are very good at telling us just that! This is why business leaders who eat this stuff up go nuts. I know for a fact my prior CEO is basically the most cardboard character in this regard. There were plenty of first hand accounts internally about him drawing his inspirations from Grok. He is the CEO who has decided chatbots1 have unlocked his superhuman potential rather than expose him has a fairly weak leader with no new ideas.

I digress—which, I realize, is something I do every other paragraph,

The “context is king” comment infuriated me every time it was said or written. It wasn’t just this one co-worker either, but rather legions of the LLM-drunk “engineers.” Annoyingly, it is just as much an excuse as a directive. It’s a fancy version of, “Well, you’re prompting it wrong.” While that is more than enough to feel justifiable rage, I have to ask what feels like the obvious follow up:

Why did you only start caring about context when we were feeding mountains of text to machines that do not think and do not reason? Why is it that our human co-workers or even our future selves were not good enough reason for this very thing? And as always, the irony pops up again. Had you been doing this from day one, you would have looked like a genius when the bots arrived. (Or in theory anyway, I certainly didn’t! I was branded a luddite.)

In principle, I agree with the statement “context is king.” This is the very reason I care a lot about metadata and continuously refine how I use it. It’s so distressing that people were never a good enough reason to do this kind of work. I think this speaks a lot about the mentality of particular individuals as well.

Fuel for the toys and tools is a good reason.

Compassion for our co-workers? Nah.

Concern about sustainability and longevity of your work? Rubbish.

As much as I love to talk about software as a fairly degenerate industry (because it is), I think this is also a good time to focus and just how utterly hapless and braindead executives are. This might be the first time I am aware of that I could get the average tech CEO to even maybe care about context or perhaps give directives that context is important and essential. As with the programmers, humans—and in this case customers as much as any of their employees—were not good enough. When the idea of reducing humans in the workplace became the dream, context was worth emphasis and even executive decrees.

Context is worth the cost for bots but certainly not for the people. Got it.

Why can we never just use our powers for good? Why is it always this?

Personal Context

It’s not precisely related, but I had two things I wanted to write about and this allows me to tie them together under a single entry.

I think a lot of people misunderstand my position on LLMs. I do not have any particular problem with LLMs as a technology. It’s very interesting technology. The world is full of interesting technology though. The problem is that technologies are not used in a vacuum. There is context for their use, their effects, and their byproducts.

I have been characterized as a luddite by some people I know. Make no mistake, being laid off was influenced heavily by my refusal to use it in the workplace, being a vocal opponent of how it was deployed and why it was being used, as well as my concerns about the religious tone corporate adoption was taking2.

I just do not see the point of exploring the use of LLMs right now for code right now. I watched a very enjoyable video by ThePrimeagen that hit home for a number of reasons, and at the end he talks about exploring this tech for what are, on their face, good reasons. Is it useful? Can it improve my flows? Can it help me in a way where it doesn’t drive?

I know enough smart people who I trust who are answering these questions with a “yes.”

So why not me? If I think the tech can be used in a way that helps, why don’t I use it?

I’ve been plenty vocal about my ethical reasoning, which is much more expansive than “this technology is evil” or “this technology is displacing people.”

I do believe there is a lot of certifiable evil coming from the companies that are pushing this technology. I definitely do not believe that people like Dario Amodei are somehow unaware or detached from the gravity of what their entities are doing. (I think him being a “true believer” might actually be scarier than him just being yet another stock corporate villain, but this is not good in any direction.)

I believe how it is being used to threaten to displace people is also a very terrible thing. This is where the myth making and realities do not align. Your corporate overlords certainly wish to believe and, more dangerously, exploit this myth.

There’s much more to it than that. Others, such as Baldur Bjarnason and Johannes Link have already done great work at explaining this. I agree with pretty much every point both of these individuals have made.

Unfortunately, the ethical issues seem to fall on many deaf ears in tech circles. The only people who care about them are readily discarded by supposed neutral technology authorities who don’t want to talk about anything but the technology in and of itself, including Linus Torvalds.

There are other questions around AI (like what the economy of it will actually look like in the end), but “is it useful” is no longer one of those questions. Anybody who doubts that clearly hasn’t actually used it.

Yes, it can also be a somewhat painful tool, both for maintainer workloads and just from a “it keeps finding embarrassing bugs” standpoint.

But the solution is not to put your head in the sand and sing “La La La, I can’t hear you” at the top of your voice like some people seem to do.

Yeah Linus, I agree. Here’s the thing though, not factoring in the non-technical elements of the tool is your own variation on, “La La La, I can’t hear you.”

He also says:

Sure, the social angle of working on open source is important and often a very motivating part of the project, but in the end that’s a side benefit, not the point of the project.

This is NOT some kind of “social warrior” project, never has been, and never will be.

You don’t have to be a “social warrior” project to benefit immensely from the social and legal aspects of a project. The use of the GNU General Public License is no less important to the success of Linux as the technology itself. This is not merely a “side benefit.”

You cannot look at Linux, its use, or its success merely as a technological success. Those chosen license was key in keeping the Linux ecosystem out from under the auspice of a single corporate entity. It has been key in the growth and protection of many core open source projects. This seems pretty important for other load-bearing tools such as git.

Technology does not exist in a vacuum. Not Linux. Not LLMs. Anyone who thinks the licensing aspect isn’t that important doesn’t realize how much Microsoft was able to absolutely strangle innovation on the web for about a decade until mobile browsing became big enough to displace desktops where Microsoft had a de facto monopoly. (Web developers in that era probably remember how many hours of their lives went to making stuff work in IE 6. It was not fun.) Microsoft was especially famous in this era for “embrace extend, and extinguish.” This both a killer and existential threat to may technologies of the first decade of the twenty-first century.3

Why bother though? Linus himself is happy to brush aside anything that isn’t technical. I feel like this, in a very distant sense, is related to context only mattering when it was for machines rather than people. I can’t put my finger on it specifically, but it’s a common character flaw in the industry.

When I discuss the matters with technologists these days, I don’t discuss ethics at all. I don’t talk about the way the tools are deployed right now harm the development of the next generation. I don’t bother with thought rot, code quality, sustainability, plagiarism, or any other mere ethical issues. I’m not going to talk about right-wing poisoning or teenage suicide. These are too complicated and, apparently, unimportant in the wake of “amazing” tools.

I am left to focus on what I think is the real fundamental gravity of it all right now: the economics. Over the past 12-18 months I’ve got to see the likes of Ed Zitron go from basically some blogger out there to being interviewed on Bloomberg4. The only question I want answered from the technologists drunk on the usage of LLMs is this:

How much would you pay for the output?

This is all heavily subsidized. The prices are not coming down significantly anytime soon and will, in fact, increase because if they do not the technology disappears. I think there is a case to be made for someone casually being aware of how all of this works and how it may be incorporated if it were sustainable. I do not see a case for people who have dedicated time and energy to their current workflows.

Do you really want to become reliant on a technology for your profession that you might not be able to afford in a few months? There are certainly complexities for those who are forced to use it—to which I say, go all in, your employer absolutely deserves the output and the cost. For those who become evangelists? This is pure insanity. If your primary concern about your project management system is how well data can be fed to an LLM, you are putting a lot of time, energy, and skill development in a very silly place.

I was glad to see Cory Doctorow, someone noteworthy, say something along these lines. In this interview he, an experienced writer, talks about his use of LLMs as a sort of “plug-in” and the interviewer also talks about how she gets some use out of the technology. He asks how useful it would be, more or less, if she had to pay what it actually cost.

This detail entirely obfuscates the ROI and general usefulness of the tech. For me to evaluate it—in terms of a pure technology—I would need to be able to understand what I had to give up to use it. You cannot meaningfully evaluate the general applicability or whether it belongs in your workflow without having some idea of what tasks cost.

There is a part of me—a part that I admit is not one of my better qualities—who looks forwarding to the weeping of Babylon when these companies are forced to play within their means and find a way to turn a profit. It will happen—either that or they simply collapse. We cannot throw money into the furnace forever. This industry has been built on the ignorance of venture capitalists—Zitron calls it “venture welfare,” which is brilliant.

While venture capitalists are never going to win any prizes for their intellect, their simple motivation—a need to turn a profit—is going to win out. It’s the nature of their game. Microsoft, Google, Amazon, and Meta are drunk on the idea of profit and growth, not the technology for its own sake. The moment one of them wakes up to the reality that profit is never coming from this use of LLMs, I will prepare for those Babylonian tears. They’ll come fast. They’ll come hard. They’ll be delicious. No amount of “La La La, I can’t hear you” will stop it.

I probably shouldn’t laugh in the face of addicts unable to get their fix… but hey, nobody’s perfect.

  1. For the record, I’m moving away from using the term “AI.” It’s a misnomer and even saying “AI” just helps with the mythology. We’re dealing with LLMs and related chatbots. Doesn’t all of this sound far more ridiculous when you call “agents” what they really are: chatbots. ↩

  2. As a cult survivor, lemme tell you, I know all too well what that looks like. Also, what a lame thing to be messianic about. What kind of person views Sam Altman, Dario Amodei, or Elon Musk in a messianic manner? An entirely pathetic and hollowed out husk of a human being, that’s who. If you use Grok to help you run a company, I’m totally talking about you, by the way. ↩

  3. Also, even just the use of “embrace” is why JavaScript hangs around like the worst version of herpes the programming world has ever experienced. ↩

  4. It’s been pretty funny to watch this unfold. While I know Ed is terrified of the coming economic crunch that will certainly come, he’s got to be feeling really smug right now. ↩