Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Back to Podcast Digest
PwC’s Dallas Dolen: AI Hardware’s Advantage, TokenMaxxing, Automation’s Impact
TL;DR
AI has flipped the old software-vs-hardware value stack — Dallas Dolen says margins have rotated from SaaS back to hardware, especially chips, as AI compresses software moats and makes it easier to build CRM, ERP, and marketing tools faster and with fewer engineers.
The trillion-dollar AI buildout is real, but supply constraints will decide who wins — Dolen cites forecasts of roughly $7 trillion in data center spending over 10 years plus nearly $1 trillion more in telco, while warning that chips, workers, copper, trade friction, and power shortages mean many announced projects still may not happen.
Enterprises aren’t hitting compute ceilings yet — they’re hitting ROI ceilings — PwC has had to shut off teams for overspending on tokens, and Dolen says the practical constraint today is not “we can’t run the model” but “we don’t want to pay for this use case anymore.”
PwC is treating the model market as a multi-vendor race, not a winner-take-all bet — Dolen says demand is coming in roughly a third-third-third across major providers, with different hyperscalers and frontier models winning on security, engineering workflow, or specific industry momentum.
Agentic AI is already useful when the job is clear, bounded, and annoying — Dolen’s own workflow uses Grok to scan X, Gemini to fill gaps and clean formatting, and Microsoft tools to distribute a 5:12 a.m. global briefing, while client traction is strongest in finance back office, sales/marketing customization, and legal review.
The biggest long-term shift may be org design, not just automation — Dolen expects companies to move from the classic pyramid toward a “spindle” or narrow “sail,” with fewer people at the base, but argues costs, human preference, and change management will make the transition gradual over the next two to five years.
The Breakdown
The trillion-dollar AI factory boom, with a Vegas-sized caveat
From the floor of Google Cloud Next, Dolen frames the moment as a massive buildout of “AI factories” — basically data centers — and says forecasts point to about $7 trillion over 10 years, plus almost another $1 trillion for telco. His basic message: yes, the spend is real, but “invest at your own risk,” because not every bet will hit and a lot of projects may stall on chips, labor, copper, and deployment bottlenecks.
Hardware is the new software
The sharpest idea early on is that AI has inverted the stack: value has shifted from SaaS back to hardware. Dolen says AI is compressing software moats because companies can now build chunks of CRM, ERP, and marketing themselves, while chipmakers and infrastructure providers are sitting on the high side of the cycle and capturing the margin instead.
Enterprises will pay more for AI — until they demand proof
Alex brings up Dolen’s earlier audience poll where roughly 30–40% said they’d pay up to 5x more for the AI services they already use today — from $19.99 to around $100 a month. Dolen says that’s great news for OpenAI, Anthropic, Google, and others, but the enterprise eventually asks the harder question: where’s the ROI?
The real bottleneck isn’t compute yet — it’s waste
Dolen says most enterprise customers are not failing because they can’t get enough tokens or compute. They’re failing because they’re spending too much on the wrong things, and PwC has literally turned off teams after deciding certain AI usage wasn’t worth the token bill; yes, there’s even some internal “token maxing” and gamified leaderboards, but with guardrails.
Nvidia vs. accelerators? For now, everyone gets a seat
Prompted by Jensen Huang’s “I didn’t wake up a loser” line, Dolen says PwC isn’t picking one ecosystem yet. They work across hyperscalers and frontier model providers because different environments demand different tools, and because the market itself is still split roughly into thirds; the longer-term edge, he thinks, may go to whoever controls the underlying infra, pipes, and power when structural limits really hit around 2027 to 2030.
What an agent actually is, and the eight Dallas uses himself
Dolen gives the cleanest metaphor in the interview: in Hollywood, an actor’s agent has authority to act on their behalf in a defined domain. That’s how he thinks about AI agents too, and he says he runs about eight daily; one lives on Grok to scrape news from X, then passes through Gemini for missing headlines and cleaner links, before landing in Microsoft tools for distribution to about 1,000 PwC partners.
Where agentic AI is actually working in enterprise
The strongest use cases he sees now are not sci-fi. They’re back-office finance workflows like source-to-pay and procure-to-payroll, front-office sales and marketing customization, and legal work like research, contract comparison, redlining, and summarization — all areas where agents can already do 90–100% of the routine flow before a human reviews.
Work won’t disappear overnight, but the org chart will change
On automation, Dolen says the classic corporate pyramid is likely shrinking into something more like a “spindle” or even a boat’s sail, with a narrower base over the next two to five years. But he pushes back on instant-doom narratives, arguing that economics still matter — if an agent costs $12,000 a month, a person may still be cheaper — and that change management, not model capability, is the hardest part.
Shadow AI, plumbers, and the people who “see the way”
The closing stretch gets philosophical and personal: Dolen tells a great story about his nearly 80-year-old dad using Enterprise GPT and YouTube to troubleshoot a toilet before finally needing a specialist plumber, which becomes his argument that humans adapt fast and expertise still matters. He ends by praising the “shadow AI” builders inside companies — like a PwC partner who spent five weeks of nights and weekends coding a Claude-powered workflow that gets to 99% of the desired output — as the people who “see the way” and pull old organizations forward.
Share
Keep Reading
Two Ways to Follow
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.
