Prompts, Web Applications, & the Death of Accountability

April 15, 2024

I have a theory for where things are going, and I don’t like it.

Web Apps

I could write a big giant post about why I have a lot of problems with web applications. I love the web, but I think from a pure productivity perspective, we’re using it wrong, and asking it to be good at things that it is fundamentally, technically ill-suited to do. And that’s not a knock on the various innovations and workarounds developers have come up with over the last twenty years to make browser-based apps better (and sometimes, borderline really good!) — if anything, those achievements are more impressive when you consider what the browser actually is.

Anyways, I think for the most part, we have web applications for good reasons. It can be extremely useful to jump into an application from a dumb client machine (maybe it’s not yours, for instance). You don’t have to update things (although… you can also CHOOSE not to update things, or at least you used to be able to decide that). Plus, people accepted the browser in a way they didn’t accept Java apps or other write-once run-everywhere environments because the web added a whole crazy level of connectivity to everything that literally changed the world.

All that stuff matters. But the reason most of us spend all day at work in browser tabs using the same 10 applications and dealing with page reloads and clearing our cache and getting logged out isn’t really for any of this — it’s because it’s cheaper to develop one compromised, least-common-denominator app and deploy it everywhere than write multiple dedicated applications for different platforms. That’s it. It’s just money, and time. I should know, because I paid to have a web application developed and I absolutely LOVE native desktop applications. However, Resolution, like most web apps, remains a web app that is accessible to 100% of the total addressable market, but optimized for none of it.

Of course, nobody cares. I mean, obviously I care, but basically no one cares about this, because over time the lack of good desktop app experiences and the improvement of web app capabilities has created a world where the differences are close enough to be ignored for long enough that we stop realizing they exist. Normal people don’t blame JavaScript from a bunch of lame Electron apps when their computer fans sound like jet engines and obliterate their laptop batteries — they blame their computer. But dammit, I blame JavaScript, and yet I can’t get away from it because these are the native apps that get offered to me. The market has decided — or sort of stumbled into this, I guess– by simply not putting up enough of a fight as the coolness of the web (blogs!) slowly took over important, less-cool things (powerpoint!) and boiled us like a frog. Now everything sucks and you need to clear your cache but at least you don’t have software updates (never mind, thanks Zoom, speaking of shitty desktop apps, way to undermine my entire point).


Basically the worst people in technology — not the profit makers, but the profit maximizers — got their way with web apps. We accept worse experiences, and in exchange, it’s cheaper for companies to develop and deploy software. Obviously the web is a wonderful playground for new businesses and ideas, but the fact that information professionals are sitting there with two frigging menu bars on their screen like idiots at all times is sub-optimal but apparently something we’ve all decided to accept.

Now, I’m being told that everything is going to be done with prompts. I type what I want, and the computer just does it. This is the future! Why even have applications at all?

There are reasons, of course, to do this. For all the times that natural language is an idiotic, self-defeating way to interact with a computer, sometimes it isn’t. I too set timers, ask the temperature, and demand that things pause. It’s possible that better linguistic processing could open up some new, useful doors, and that’s cool. But… I don’t want to edit video with prompts. I don’t want to edit photos with prompts. I definitely don’t want to build a website with a prompt, for goodness sake. This is madness — these things are expressions of the human mind; barking your specification (or hand-crafting some bizarre series of hyper-specific instructions like you’re leaving a note for your incompetent spouse to remember to water the plants and feed the kids), hitting enter, and then trying again isn’t how anyone makes anything of any real value.

However, like web apps, there is a secondary benefit to prompts that I’m scared is going to motivate providers to push them into our lives whether they make sense or not. And that’s…


Trying to do something with an LLM can work… sometimes. That always seems to be the situation, doesn’t it? The things that sometimes work continue to expand, but the part about them only working sometimes never seems to change. You know what else always seems to be the situation? That the bad output is my fault, because I wrote the prompt.

If I want to copy and paste something in Photoshop, there are a couple different ways to do it, but those ways are finite. They also work 99.95% of the time, and in the absolutely incredibly unlikely circumstance that they don’t work, Photoshop is wrong and even Adobe would admit that it’s Photoshop’s fault.

It’s hard to understate how financially awesome it would be, in the most cynical possible sense, to run a software company where you were never held accountable for anything your products did in response to user actions, or even how they work. Companies have been trying this for years with concepts like ad-supported, free software that makes money without any of the usual requirements to serve customers — but this would be a whole other level of non-accountability. In this way, the fantasy of “conscious” software (and it really is a total fantasy right now, let’s not kid ourselves) is especially self-serving, because it takes people who design and build software and gives them a bizarre, almost parental relationship to whatever you are paying them money to use. “The software is alive! Yell at it, not us!”

Just like Java, if you rolled this out overnight, people would lose their minds and revolt. You’re seeing it with the stupid Humane AI Pin thing, and you’re seeing it with Microsoft getting out over their skiis and trying to shove CoPilot into anything with a mouse cursor in it. There has to be a tradeoff people will accept, and since it’s probably a bad one, it has to be both very exciting, and occur slowly. With generative AI and prompts-as-UI, that tradeoff is this — we will let you basically launder other people’s ideas and customize them without any skill, and do it cheaply, but you will have to let go of any requirement for granular control, predictability, or reliability. It’s your fault, now.

That’s the tradeoff. I don’t want to make it. But I’d also rather have to remember to upgrade my applications and pick a platform than be stuck running EVERYTHING inside the glorified document viewer posing as an overwhelmed, underpowered application layer. I don’t get to choose that anymore, because other people did make that trade. I just really, really hope enough people choose not to make this one, because I think it’s going to be much, much worse.