Everyone's jumpy about AI right now—and for good reason.
The hype has been massive. The investment has been astronomical. But where's the actual return?
In this episode, Azeem Azhar, founder of Exponential View and advisor to tech leaders and governments, breaks down why it’s make-or-break time for AI. Companies need to prove there's real ROI, not just prototypes launched and tokens spent.
Note: This interview was recorded well before last week’s "SaaSpacolypse" (big market drop); his analysis only more relevant now.
We cover:
Who can build a real moat with AI—and why the winners will likely come from unexpected places, as they have in previous tech transformations
What hard evidence would actually prove AI is working (hint: it's not usage metrics)
The physical constraints nobody wants to talk about: chips, data centers, power grids, and whether America's infrastructure is up to the task
Why OpenAI's "ubiquity strategy" might be spreading too thin (and what Anthropic is doing differently)
The "pragmatic addicts" problem: we're dependent on AI even though we don't trust it
How Azeem and his team use AI to be more productive, how they automate whatever they can, and why individual contributors are acting more like managers (of AI)
What else we should be talking about! Listener Heather Myers wants Azeem and me to step away from the AI! And in so doing, Azeem surprises me with an answer that invokes two great sages: Beavis & Butthead
Links & Resources:
Azeem Azhar’s Exponential View
Their Boom or Bubble dashboard
More on the “MIT Study” claiming 95% of AI projects fail that Azeem and I both found to be poorly done, but that is nonetheless is quoted by everyone: Here’s Azeem tearing the study apart with data; and here’s me riffing with Kwaku Aning on it. You know why Azeem liked my take? Because I actually read the thing, unlike ~95% of the writers out there who just quoted that 95% number
Azeem's New York Times piece on America's electric grid challenge
Our listener question came from Heather Myers, who runs Spark No. 9, which specializes in rapid experimentation
Chapters:
01:51 Why the next 18 months are the crucible for AI
04:09 What hard evidence would actually prove AI ROI (not token counts!)
06:55 Why it's so hard to measure AI's real impact
09:55 Who can build a moat with AI? Winners will be in "odd places"
12:56 Structural data advantages: why Waymo's edge is hard to replicate
14:34 Coding agents and whether developers will become disillusioned with them
18:21 Physical constraints: chips, data centers, power, and America's grid problem
21:25 How the Gulf countries became an unexpected AI hub
28:02 "Pragmatic addicts": why 75% of Americans distrust AI but use it anyway
31:45 The narrative of AI can be very unappealing: heaven on Earth or dystopia
34:36 How Azeem's team uses AI: augmentation vs. automation
40:06 What should we be talking about besides AI? — A question from a listener: Heather Myers
43:46 Sounds like science fiction: What Azeem can't believe is real and here today
If you don’t already get Future Around & Find Out emails, go ahead:
Thanks!
~ Dan
P.S. I have an upcoming interview (my second!; here’s the first) with Khan Academy’s Chief Learning Officer, Kristen Dicerbo. If you have a question for her on the future of education, AI, Khan Academy’s approach, etc… please reply to this email or email me directly at [email protected]. I’d love to include your question, as I did Heather’s in today’s episode.