Bill Shope’s latest piece on the industrialization of intelligence has been making the rounds in my favorite Morgan Stanley circles, and it’s sparked the kind of conversation I remember most fondly from my trading floor days – the ones where you’re trying to map out inflection points before the market fully prices them in. Except these days, the whipsaw feels less like public market volatility and more like private markets racing to figure out what AI means for labor composition and industrial structure.
Bill’s core thesis – that we’re building digital labor, not better software – has become somewhat LinkedIn-commonplace at this point (thank you, AI slop!). But seeing it through his analytical lens brings a clarity that the not so hot takes haven’t quite captured.
The SaaS Playbook Doesn’t Apply Here
Since ChatGPT launched, founders have pitched AI as “the next SaaS wave,” pattern-matching to what we know because that’s how humans make sense of disruption.
Yet Bill so effectively articulates that chatbots were never the destination. Instead, they were always the on-ramp.
When the iPhone launched, everyone obsessed over checking email without a keyboard. We gave up our beloved Blackberry Messenger (RIP) for blue bubbles and autocorrect fails. But the real revolution came when developers realized they could build native apps that did things no mobile website could. Instagram couldn’t exist in a browser. Neither could Uber.
Agentic AI is that inflection point. These aren’t tools that make humans faster. They’re systems that perceive environments, make decisions, and execute with real-world consequences. They’re not brittle procurement software that requires three people to approve a $200 purchase order. They’re intelligent systems that monitor inventory levels, negotiate with suppliers in real-time, and automatically reorder materials when production schedules shift – effectively executing decisions that would’ve taken your overburdened procurement team days to coordinate.
The Economics That Actually Matter
Many of us investors have burned considerable calories hypothesizing how AI reshapes end markets and company economics. Bill’s labor lens reframes this debate exceptionally well.
When you’re selling software, you’re constrained by seat counts. However, when you’re selling digital labor, your TAM becomes a function of the global labor market. Accordingly, agents exponentially expand what’s economically feasible and addressable. Tasks too expensive or complex to staff become viable. And shadow TAMs emerge in ways that make traditional market sizing look provincial. Today, this remains one of the most underappreciated dynamics in AI – the ability to unlock demand that simply didn’t exist when the work was too expensive to do at all.
Token deflation drives this flywheel. As Bill notes, the cost to use AI falls roughly 10x every 12 months – far faster than Moore’s Law ever moved. Cheaper tokens expand what’s automatable, making it viable to run agents on tasks that would’ve been cost-prohibitive six months ago. Better models drive down costs, which increases usage, which funds better models. Yes, the implementation gap is real (enterprises have the fun job of wading through data swamps), but government involvement and AI-assisted implementation platforms are creating forcing functions that compress timelines in ways previous cycles never saw.
Where Moats Actually Form
A few years into the AI wave, it’s become increasingly apparent that LLMs themselves probably won’t be the moat in most markets. Standards like MCP are commoditizing connectivity faster than expected. So where does defensibility emerge? Bill makes a compelling case for context capture, which aligns with my own recent musings.
Ultimately, agents are context junkies. They need refined data at precisely the right moments. The winners will be those who build context capture into their DNA from day one – not as an afterthought, but as their core competitive advantage. For enterprises, this likely means becoming the system of record over time, not just another application sitting on top of one. For consumer plays, it means accumulating behavioral patterns and environmental awareness that compound over time, creating switching costs that feel less like restrictive lock-in and more like “this thing just gets me.”
This is why the opportunity in vertical AI remains so compelling, despite what feels like increasing crowdedness in some of the buzzier end markets. While general agents make great demos, real value comes from deeply understanding specific workflows and building orchestration layers that know when precision matters and when creativity should take the lead.
The companies solving specific end market problems – think Finch or Legora in legal, Abridge or Valerie Health in healthcare, Liberate or Trunk Tools across legacy industries – aren’t trying to be everything to everyone. At Redpoint, we’ve watched their proprietary context moats compound as they expand into end-to-end platforms that are more expansive than even our initial expectations at investment.
And looking ahead, we see ample opportunity across industries that haven’t yet been touched by this wave – real estate, manufacturing, and countless others where the implementation gap is only now starting to close.
The Main Event Has Arrived
Bob Dylan (my Top Spotify artist 2 years in a row) once sang about the times changing. The verse feels prescient these days:
“The slow one now will later be fast, as the present now will later be past.”
We’re in that liminal moment where the old order hasn’t quite collapsed (despite the stock market’s recent reactions) but the new one is undeniably arriving.
If chatbots were the beta test, agents are the full release. And as Bill argues convincingly, unlike previous software waves where we augmented human capabilities, this time we’re fundamentally reshaping what labor means, who performs it, and how value gets created.
That’s the inflection point worth paying attention to.
Building in this space? Find me on LinkedIn or subscribe for more on how AI is reprogramming reality.