An agentic prediction.

Part 1 of my talk on the Acadian Ventures AI thesis.

This will be the first post of several where I explore the themes in our AI thesis. I presented it at the Acadian Ventures AGM last week in Utah. Thesis is a bit of a presumptuous word here, the talk is really just a bit of history, prediction and conjecture.

I’ll also post a full video of the talk here and on youtube when it’s ready.

First up, making predictions is fraught. I first read about the 1890s manure problem in Daniel Susskind’s book, A World Without Work. It illustrates brilliantly the challenges of prediction.

That said, I’ll make several predictions in the presentation. I’ll explore the first one and draw a parallel from the 1990s here.

In the late 1990s lots of companies implemented ERP systems, in part because ERP was cool (yes kids, it was), and because it would solve for the Y2k problem. Some companies implemented ERP really well, most did it okay, and some were a complete disaster. I remember that Hersey’s missed Haloween, Nike’s share price dropped 20%, and several companies, including FoxMeyer, the pharma wholesaler, filed for bankruptcy. SAP, Oracle, Baan and others all hand their horror stories.

These projects failed for all sorts of reasons, but what interests me here is the impact. Because these projects were genuinely transformational, failure had massive consequences. Success did too.

If AI Agents are even partially as powerful and transformative as many folks believe them to be, then it is pretty much inevitable that an AI Agent is going cause some serious damage.

I’m predicting that in the next 18 months or so, an AI Agent is going to bring a significant listed company to its knees: Chapter 11, or a massive miss, share price pummeling, CEO out the door.

I’m an avid reader of Nassim Nicholas Taleb’s work, both on paper and online. Many of the things people call Black Swans aren’t. An AI Agent messing up is pretty much inevitable, human in the loop or not. It’s not a Black Swan.

Paradoxically, AI’s failures will be evidence of its success.

(I used openAI to make the FT image. I tried hard to get it to spell correctly, but gave up when I realised it was a pretty good metaphor).

So paradoxically, AI’s failures will be evidence of its success.

As I usually do, I’ll end with a tune. Most of the time there’s an obtuse link back to the post. My relationship with the Smiths is complicated, but almost every day a Smiths’ lyric pops into my head and stays there.