No matter what your take on genAI, I think we can all agree that the version described by its fans lands somewhere between "a far-off possibility" and "a constantly-shifting goalpost that charlatans sellers use to lure more marks buyers." Which means it's hardly a good reason to spoil the environment, or push people out of their homes, or gut the org chart, or commit any of the other foolishness that's being done in favor of this technology.

genAI may eventually become A Big Deal For Real™. Maybe. Just maybe. But with each passing day it's increasingly unlikely to create widespread, positive ROI. We're getting closer to accepting – or, better yet, being forced to accept – that it will take its place in history as A Humble Technology Tool™. And a fad.

In other words … genAI's become quite a shitshow, hasn't it?

You may point out that it has always been a shitshow. I would disagree! It started off well, as a technology – a natural progression after mathematical modeling and then machine learning. But then it became a movement.

That movement is the shitshow.

Being clear

People love to treat "AI" as a shorthand for just the generative variety. I don't want to sully the name of other technologies that fall under that umbrella, so throughout this piece I'll type the extra three characters to emphasize that I'm talking about large-scale, LLM-backed systems. The kind that landed in late 2022 and have been hogging the spotlight ever since.

This is also a good time to draw a barrier between genAI-the-hype-wave and genAI-the-tool. The hype wave is that cacophony of techbros trying to convince everyone that their magic machine is our lord and savior. Their fans keep singing along with them, despite the fact that the due date for their rapture keeps pushing into the future.

The tool is … well … a tool. Like any other tool, it does some things well and other things poorly. I encourage individuals and businesses to apply this tool to tasks it can perform well. Because contrary to what the naysayers want you to believe, genAI isn't completely useless. There are at least tens of small-scale, individual use cases – touch up an email here, create an image there, that kind of thing. Businesses have also found a few small wins specific to their needs, like touching up emails or creating images.

But genAI needs large, widespread, bigger-ticket uses to make good on the collective excitement and investment. And thus far it has just three contenders:

1/ Software development. The world's seen some noticeable gains here. Some software developers are experiencing significant productivity gains and that's a reason to celebrate. Emphasis on "some software developers," though – I'll get to that in a moment.

2/ Sleaze. Everything from outright-crime (fraud) to not-technically-illegal-but-still-bad (nudification apps, propaganda, cloning other people's work). Terrible people have always been terrible, and genAI has made it easy for them to prompt their way into new levels of harm.

3/ Creating interest in genAI. It doubles as its own hype-man, a self-reinforcing mechanism that keeps the myth alive. For now. Should genAI's fortunes turn, this interest could become an unwelcome spotlight.

Sharp-eyed readers will note that there's just one legitimate use case in there. Relative to the time, money, effort, and hype poured into spreading the genAI gospel, even software development isn't enough to create lasting ROI.

Hence my earlier clarification about "some" software developers. Agentic code tooling has proven a force multiplier to that small segment of experienced, professional developers who have the skills to operate at a high level and are able to delegate. But as you step away from that group, the benefits are more about image and excitement than actual workplace gains. The fallout from vibe-coded amateur apps and corporate tokenmaxxing initiatives may shift the software development ROI to a net negative.

As such, the software development use case rests atop a shaky foundation. What happens if genAI loses the developer crowd? (More on that in a future piece. One which is mostly written, and which is due to land in late August or mid-September.)

Hell, meet handbasket

The core issue isn't that genAI is being advertised as far more powerful and capable than it truly is. It's that its biggest fans have bought into this idea wholesale and they are dragging us along for the ride.

The genAI die-hards have hijacked our reality. We are unwilling passengers careening through their fever dream, forced to go along with a farce that is held together by positive vibes and usage mandates. On their side of the looking glass, genAI actually does everything they claim. The "AI race" is an existential threat to companies that doubles as a thorny matter of national security. Those datacenters? They form a bulwark against loneliness, business inefficiency, and the PRC. (That last group commits the ultimate sin of applying genAI to practical use cases.)

I was about to call this genAI hype wave the OceanGate disaster for the world economy. But unlike the Titan submersible, we are well aware that this is a rickety, oversold heap steered by a PlayStation controller. And we want off.

Even worse, the genAI salesbros can't tell us what it's good for. They preach that it's the end-all be-all, yet it's our problem to figure out why. What's that Chris Rock line? The one about falling for the antics of a trickless magician? That worked great as a comedy bit. But I'm uncomfortable with it as my reality.

This special blend of excitement-without-purpose has created a new ring of workplace hell. genAI-pilled execs have issued usage mandates to turn every employee into a miner of use cases. They offer no guidance beyond "here's a key for Claude/Gemini/ChatGPT; have at it." It's the Infinite Monkeys approach to finding practical applications of the technology – slow, inefficient, and prone to downside risk exposures. Like, say, bots deleting critical files. Or dishing out bad customer service guidance. Or claiming to have completed work they haven't actually touched.

It's also expensive. Ask any tokenmaxxing CEO how they felt when the bill finally arrived. (Or ask the employees who simply wasted tokens just to keep the boss happy.) Some companies have gone as far as to trim headcount in order to fund their genAI drug habit gambling addiction transformation. All this for a tool that has yet to prove itself. They're confident that genAI bots will serve as cheaper replacements for the human employees, even though the major players have finally dialled back their "genAI job-pocalypse" rhetoric. Something that the rest of us already knew was rubbish, since we could clearly see that the bots couldn't do the job.

Did I mention, by the by, that genAI is just a tool? A tool and nothing more? Because it's looking more like an object of worship to this crowd.

But wait – there's more

That's just inside the workplace. Let's not forget the impact this genAI mania has had on the world outside the corporate four walls: social harms like AI psychosis; people who have been on the receiving end of AI-based crime and harassment (including nudification apps); anyone who has fallen victim to some oversold AI tool (facial recognition and applicant scoring tools); damage to the environment and people in the area (making way for thirsty, electricity-guzzling datacenters); businesses attacked by bots that break containment and go rogue; and several more issues that will have cropped up between the time I write this and the time you read it. It's as though someone took Philip K Dick's fiction as a blueprint instead of a warning.

Plus there are the potential future harms to the financial system, as the genAI companies take on debt to fund datacenters. It's even said that a few of them seek to IPO simply because they need the money to stay afloat.

Even worse is that the economics of genAI hype don't add up. Consider all of the money pouring into the sector, under the assumption (hope) that the dream eventually becomes real. Where's it going? We have the major players creating circular financing deals. Some of them are issuing bonds and taking on loans to finance their massive datacenters, even though there aren't nearly enough use cases to support that amount of computational power. And one even hinted at the idea of a government backstop for their activities (before quickly claiming they meant something else). Is it me, or does this feel like 2008 all over again?

From there, it's not hard to believe the rumors that the foundation model providers have heavily subsidized token prices – taking a loss to gain market share and waiting till companies are hooked before showing them the true cost.

They can pump up genAI for a while, but as it stands it's a perpetual-motion machine. The minute they take their hands off the crank it will grind to a halt. And they know this. Over the past year, major names in genAI have acknowledged that the field is not just a bubble, but a good one. Some have gone as far as to complain about people badmouthing genAI. Because they know they've built a house of cards and are asking people to not breathe on it.

It reminds of a little something uttered by British Prime Minister David Lloyd George, in 1917, describing the horrors of WWI:

If people really knew, the war would be stopped tomorrow.

But of course, they don't know, and can't know.

With that, the genAI companies keep spending. And asking for more money. And pushing for more space, to build more datacenters. And begging for more time, before genAI really does its thing. Even though they acknowledge it's not going as well as they'd planned. Somehow we're supposed to play along.

Did I mention that genAI is just a tool?

Madness

I say this as a longtime tech practitioner. I've been here since the Dot-Com era. I've witnessed early-day e-commerce and cloud and mobile. I was part of that initial cohort of the new-age data wave, a first-hand participant in the iteration through predictive analytics, Big Data, data science, and machine learning. My entire career has been tech-wave (bubble) after tech-wave (bubble) and I can tell you:

This time, it really is different.

This time, it's madness.

If you want the real smoking gun that genAI isn't all that it claims to be, tune out the noise and notice who's not speaking: experienced AI practitioners. I emphasize both "AI" (the wider term) and "experienced" (which excludes the newfound genAI experts who mysteriously changed their LinkedIn profile in late 2022).

None of the actual, experienced ML/AI professionals I know are over the moon about genAI. We neither demonize nor praise it. We just see it as – you're probably catching on to the theme here – Yet Another Tool in the toolbox. An LLM is a larger form of models we've built hundreds of times over. It's a good source of highly-specialized randomness, making it useful for fun side-projects and generating synthetic data. It does what it does, and it doesn't do what it doesn't do. Nothing to write home about.

Since genAI represents the next evolution of our field, we long-time practitioners would be at the forefront of the marketing blitz.

But we're not.

Even people who work inside massive AI companies, like Meta and Google, are tired of this nonsense. They've grown weary of pretending that genAI is bigger than it's said to be. Because it's just a tool.

Still not convinced? Here's the ultimate litmus test:

If genAI were anywhere as good as the fans claim, they wouldn't need to push it so hard.

They wouldn't need to blow cash on massive ad spends and influence campaigns to bring people around. They wouldn't need to buy a podcast so they could control the narrative. They wouldn't need to resort to blatant Sinophobia, or rush datacenter deals through local governments, or try to make these polluting behemoths look nicer, or threaten eminent domain when they don't get their way.

No. If genAI-the-tool were anywhere near as good as genAI-the-hype-wave, we'd all be begging for it.

But right now we're mostly begging them to fuck off.

A way forward, if not a way out

People will accuse me of being rabidly pro-genAI because I don't demonize it, or because I run a business advising companies on (among other things) AI adoption. Others will assume I am fiercely anti-AI because I point out the flaws and don't sugar-coat my message. The remainder can't tell which way I lean, and they have read this far looking for a definitive answer.

The simple fact is that I approach genAI with the calm, nuanced view of a long-time practitioner: it's a tool, and should be treated as such.

Which means my work will continue as usual: searching for meaningful, practical use cases and managing the risk/reward tradeoffs of this emerging technology.

Some other people already do this. I hope that crowd grows. The world needs someone to lay the groundwork for when reality finally, inevitably catches up to genAI and we have to make sense of what remains.

The corporate world is very much on the hook here. We need companies to engage in more deliberate, more strategic exploration of genAI. We need better experiments, and the kind of adoption that treats the technology as a tool.

Because that's all it is. Just a tool.

Or to borrow from the late, great Bill Hicks, it's just a ride:

And it's fun, for a while. Some people have been on the ride for a long time, and they begin to question: 'Is this real? Or is this just a ride?' And other people have remembered, and they come back to us and they say 'Hey! Don't worry, don't be afraid - ever - because... this is just a ride.'

And we kill those people. 'Shut him up! We have a lot invested in this ride! Shut him up! Look at my furrows of worry; look at my big bank account, and my family. This has to be real.'

It's just a ride.

But we always kill those good guys who try and tell us that - ever notice that? - and we let the demons run amok.

In other news …

For more links to recent news, and with a slightly broader scope, I encourage you to check out my other newsletter. It's a weekly, curated drop of what I've been reading.