There are two ways to see a bubble: either you study the numbers, or you find something that just doesn't feel right.
One of my favourite films is The Big Short. Something I really like about it is how it shows two characters discover the housing bubble in completely different ways.
Michael Burry studies mortgage bonds from his desk, looking purely at the numbers. He trusts his research, then defies the status quo believed my most everyone around him.
Burry's polar opposite is Mark Baum. Baum isn't convinced until he can see things for himself. During a visit to Florida, the penny finally drops when he meets a stripper with five houses, yelling "hey, there's a bubble" into his Blackberry immediately afterwards.
David Cahn at Sequoia recently took the Michael Burry approach with his article AI's $600B Question. He's studied the numbers, and the numbers don't look good. While the houses (GPUs) will continue to exist and the housing market (AI) will continue on and be worthwhile, a huge amount of value will be "incinerated" (his words, not mine) along the way.
The article made sense to me, but it felt... I don't know, academic? I guess I hadn't yet had the Mark Baum-type confrontation with reality yet. I'd heard the numbers but hadn't seen any vacant houses, overdue bills piling up, or alligators in long-abandoned swimming pools.
But then an AI-for-marketing company tried to get me to pay $24,000 a year for a product the salesperson openly admitted wasn't very good, on the basis that I could use it to produce 85 shitty blog posts a month, equivalent to a thousand a year.
Oh.
Let me back up for a second and explain how I got to this point.
I work in a very small marketing team where we regularly create and put out content. I wanted to know if it was possible to repurpose some of that content a little bit faster. I didn't want to hand over my whole workflow to AI, I just wanted a computer to do some grunt work to speed things up. In my case, I wanted something like this:
In theory, I could shortcut a big chunk of the hard work of repurposing and get a passable first draft that I could then improve and build on. I'd cut out hours of initial drafting. It seemed like a reasonable enough idea.
A company that claims to help marketers do exactly this has been advertising to me on LinkedIn for a long time. Let's call them "XYZ.AI." I took a look at what they offered, and it's basically just two things:
The chatbot, as far as I could tell, may as well have been a wrapper for ChatGPT. Nothing new there. But they claimed the workflow builder could let me create multi-step workflows with multiple inputs and outputs along the way. This even included "bulk runs" to handle potentially unlimited workloads, constrained only by compute costs.
Or at least... I could in theory.
I tried it out and didn't get very far because it kept throwing errors, failing to do what I asked, and producing really, really shitty copy. You know the sort: the recycled spume of adverb-stuffed "in today's modern digital world" drivel.
Frustrated but not yet ready to throw in the towel, I booked a demo.
Explaining what I wanted to do and how often I wanted to do it (maybe only once a month), the salesperson said that what I was trying to do was really complicated. I'd never get it done on their basic pricing tiers. No, what I needed was their *enterprise* pricing tier!
Here's how those pricing tiers work out:
(At this point in the call, I couldn't hide my expression but the salesperson carried doggedly on.)
At the enterprise tier I'd get my own dedicated solution architect and customer success manager to help build my workflows. "Can't I just... build my workflows myself?" Probably not, came the short answer. For something like this, I'd need their help.
In other words, the product is bad.
Of course, they know the price is high. They justified it with the example of a company wanting to produce 85 pieces of content a month the usual way, with human authors, at significant cost. Using their service was significantly faster and cheaper than that, they argued.
As the call started wrapping up I realised four things in quick succession:
A week later, in a last ditch attempt to get us onto at least one of their lower tiers, they scheduled a one hour call with one of their support staff to see if they could at least help me build my workflow so I could use it later. In short, they couldn't: he struggled through a bunch of errors, showing me some even worse copy along the way, and resorted to demoing workflows in his own account as a last resort.
I also got the chance to ask a few questions about how things really work. Basically, a bunch of stuff is being passed through the OpenAI API, or they're just running a variety of open source models on their own compute somewhere.
In short, it was a hodge podge of other people's models held together by API calls and expensive support staff to produce the huge volumes of dogshit content that's ruining the internet.
The very next day I saw some marketing exec on LinkedIn saying how AI was going to completely change marketing in the next five years. Last year I might have paid attention to this sentiment. This time I blew it off.
This guy had clearly not tried to use this stuff lately. And if he had, he's kidding himself. If you're actually trying to do work with these tools right now, you better hope that it's okay or even expected in your industry to put out complete bullshit on a regular basis. Because that what a lot of this is: bullshit.
Then of course you've got blogs like this one from A16Z breathlessly protesting that "it's just not productized yet!!!" Then listing a bunch of startups that just do automation for boring tasks, including Zapier. (Zapier! Who have been around for ages! Long before this wave of AI hype!)
Guys, this is just software. You know, software? The stuff that's good at automating things and has been for ages? I try to remain optimistic about AI for a whole bunch of reasons but damn, the hype needs to settle down.
I send updates to a few hundred people anytime I've got something worth sharing. New articles, new book notes, new commentary.
