I'm not an enterprise buyer, so for me the moment came when I fired up Perplexity's Comet browser and it asked me to import my passwords.
My finger hovered over that button for a long time as I considered my relationship with a vendor that has never given me reason to doubt it, the ways in which I have incrementally given it some rather deep insights into my business, and the great value it has brought me – before spinning up a sandbox to test it in, the gateway of which my most privileged passwords shall never pass.
Because yeah, but nope.
Treating an AI-native browser as a candidate system under investigation will not allow it to prove the value of its novel approach, but I'm not giving it the keys to the kingdom so I can integrate it into the heart of my workflow either. The rewards could be tremendous, yeah, but the risks… nope.
Like I said, I'm not an enterprise buyer. I don't have shadow-AI users demanding better tools right now. I don't have a board breathing down my neck, demanding absurd productivity gains because somebody has let them read the whitepapers again. I don't run the risk of being replaced by a CTO half my age who is considered more of a proactive go-getter because I'm not an AI native.
On the other hand, I am a techno-utopian who hates mundane tasks, so I should be an easy sell.
You don't see a lot of AI hesitancy in the kind of numbers everyone from Oracle to Salesforce are reporting; the pipelines are pumping. There's a disconnect between the vendors and the buy-side C-suits, though, that has me wondering whether we're coming up to a wall.
The sellers keep promising shiny new capability, while the buyers keep asking about security, auditability, lock-in and compliance. One side is talking about trust broadly, while the other (which, let's face it, sells black boxes) only wants to talk about the narrower area of data sovereignty, so they can offer on-premises solutions like that somehow solves everything.
As for the start-ups? When your credentials are limited to the size of your last capital raise, then you might have a problem.
The implications weren't clear to me until Palantir was run out of the UK's Coventry City Council.
Palantir, with its expertise in the “AI-powered kill chain”, has unique reputation issues, and local governments have unique political considerations. What hit me, though, was that while the fight was nominally about AI, nobody, absolutely nobody, was making any noise about Coventry's adoption of Microsoft AI.
It is in part a reprise of “nobody gets fired for buying IBM”, but it is more about incrementalism. Setting loose the latest Microsoft AI tool does not require anything like the trust you need to throw open the data warehouse to an entirely new tool. Add-ons aren't, you know, AI, they're just updates. We need those, for security.
So, Microsoft and SAP win all the business again, no surprises there. But it becomes a wall when the novel approaches are locked out because (a) we're already doing AI, check out this cool paperclip in the corner of my screen; and (b) why would I risk it?
Like the AI as Normal Technology people say: “Benefits and risks are realised when AI is deployed, not when it is developed.” You can't prove benefit if nobody will deploy it, and the imaginary risks are far scarier than anything the real world could bring – but they remain imaginary, and thus super-scary, while nobody deploys it.
The implications for market development would have worried me a great deal less had Perplexity spent a couple of first-run splash screens to reassure me about what its browser would and would not do without my explicit consent, with maybe a bit of verbiage about security and encryption and the kind of oversight it burdens itself with.
Just showing me the shiny thing and asking for my keys? Good luck with that. Shouldn't be long until Google rolls out a big update to Chrome, and I've got the time to wait, thanks.
Share