But once the applause fades, the operators and investors are left asking a harder question: What did we learn, and what actually scales?

Three questions I keep coming back to:

1. The companies raised a combined ~$1.5B to reach their current scale. Still burning real cash. Growing 40–50% YoY. Rewarded by a shrug 4x multiple. Is the employer channel broken?

2. What does this say about the healthcare investing landscape?

3. How should we be designing the next generation (“Gen3”) differently?

Let’s break it down.

1. Employer sales is getting saturated

In the early innings, selling to employers was a cheat code. Livongo, Omada, Hinge… these companies launched when their categories were new. They could educate the buyer, ride a wave of novelty, and scale fast.

That dynamic has changed. Employers are now overwhelmed with vendor pitches. Benefits teams focus on one trending category at a time - smoking cessation, heart health, diabetes, mental health, fertility, GLP-1s, etc. And even within those categories, there’s room for maybe one breakout per cycle. This isn’t SaaS, where you can spin up dozens of large companies serving overlapping use cases. In employer healthcare, the category tends to crown one leader per cycle.

What used to be differentiators (e.g. risk-sharing, cost guarantees, integrations) are now just table stakes. And with consultants and brokers influencing procurement, even the strongest products can get buried under politics and channel inertia.

The bar is higher. The cycle is longer. And the juice is harder to squeeze.

2. On market structure and the elusive platform thesis

Many companies raised capital on the vision of becoming multi-condition platforms. In theory, it made sense: one vendor, many needs, consolidated spend. But the buyer behavior hasn’t caught up

If multi-condition breadth were a true wedge, we’d see nonlinear growth from those who went all-in on it. But companies like Dario have hovered around $30M for years. Even Omada, despite expanding across multiple conditions, reports that only 30% of clients use more than one.

Why? I suspect buyers still behave like they’re shopping for conditions, not platforms. Procurement stays siloed. And benefit consultants have little incentive to push unification. Add in payer segmentation (Medicare, Medicaid, ACA, fully insured, etc) and suddenly your “platform” has to support 16 permutations of care models, workflows, GTM, and ops overhead.

What should’ve been leverage becomes drag. Instead of compounding, complexity dilutes velocity.

3. What does Gen3 look like?

Gen3 healthcare companies won’t all look the same, but there’s a pattern emerging.

There will be two dominant types:

• Verticals that go deep on one population or problem space

• Horizontals that power the ecosystem: infra, rails, orchestration layers

The holy grail is building systems and GTMs that flex across variation without bloating the org chart or the codebase.

More capital efficient
The market already paid the tuition. Gen3 needs to get to outcomes faster, with tighter GTM loops and smarter distribution. Augmenting human labor with AI in the care model will be a major lever. And while that once seemed like a stretch, patient acceptance is shifting, especially in behavioral health and lower-acuity use cases.

Development is also getting cheaper and faster (if you saw Anthropic Dev Day yesterday, you no longer think that this is pie in the sky). That doesn't just make care delivery more efficient, it makes the cost to build and iterate on care models themselves meaningfully lower.

Population-centric > condition-centric
Gen2 companies typically started with a single condition and expanded outward. That often led to complexity without coherence.

Gen3 should flip the model: start with a defined population and design end-to-end for their needs. That also means targeting buyers who are buying differently. Often through different channels, contracting paths, and trust networks. This requires not just a new care model, but a new GTM stack and often a cleaner, purpose-built tech foundation.

You can see glimpses of this in Gen2 plays like Iora and Oak Street (Medicare), Cityblock (Medicaid), and One Medical (Commercial), though most remain capital-intensive. The constraint isn’t ambition, it’s channel fit.

Redefining the scale curve
Many care delivery companies will remain important, but top out below venture-scale. I think of them as profitable niche leaders: PE-backable, resilient, and strategically valuable, but not blitzscaling.

And the journey to scale is jagged. I’ve seen structural cliffs around ~$20M, ~$50M, and ~$100M revenue bands. Between those bands, teams hit margin compression, channel friction, or operating complexity. You have to know what curve you’re on… and how much risk you’re willing to take to get to the next one.

The next venture-scale plays
Most of them won’t look like digital clinics. They’ll be the pipes and logic layers:

• AI-native documentation

• Programmatic billing and RCM

• Infrastructure for payments, eligibility, and risk

• Underwriting, routing, compliance

AI is already accelerating this category; not just scribes, but infra-level primitives. As these grow, interoperability becomes the new scalability. If these platforms don’t integrate cleanly, we’ll recreate the same back-office fragmentation we just spent a decade fixing in care delivery.

And we haven’t even touched biotech or drug discovery, those may see even more traction.

One last thing often missed in these breakdowns:

What did $1.5B in venture dollars actually deliver for patients?

Slightly faster onboarding? Better UX? Maybe. But friction, confusion, and fatigue still define the average experience. Gen2 companies did normalize virtual care and prove that outcomes can happen outside of visit-based models. That matters.

But Gen3 has to finish the job: designing systems that don’t just scale, but feel human again.

Would love to hear where others are seeing leverage, and how you're designing for Gen3.

(Thanks to Stephanie Tilenius and Sarah Cohan for reading the drafts)

As we enter Q4, I imagine your inbox is filling up with pitches touting the latest AI enhancements. And if you’ve seen a few, you’ve likely noticed a pattern—most AI pitches fall into three familiar buckets:

1. Enhanced search

2. Content summarization

3. Assisted writing or image generation

These features are useful, sure, but here’s the paradox: they’re often surface-level enhancements, quick fixes aimed at juicing ACV by 20-30%. It’s a strategy, but is it truly innovation?

Klarna’s Signal to the Market

Klarna leaving Salesforce sends a strong message. This isn’t just about one company switching CRMs—it's a signal that the market is primed for change.

Salesforce has been piling on AI features for years. On paper, it seems they’re doing everything right. So why would a major player like Klarna walk away?

My guess? It’s because slapping AI onto an existing product isn’t enough anymore (and it's becoming ever easier to build a vanilla enterprise software).

The Future of Invisible AI

In the near future:

• AI-powered search will simply be... search

• AI-assisted writing will just be... writing

• Tech-enabled services will become simply... service

It reminds me of the early days of cloud computing. Every product boasted about being “cloud-based.” Now, we don’t even think about it—it’s just assumed. AI is heading down the same path.

Disruption Hiding in Plain Sight

For founders and product leaders, there’s an important takeaway here. Every time an incumbent adds another AI feature, they’re shining a light on a vulnerability. The message is clear: “Our core product isn’t enough, so we’ve bolted AI onto it.”

For nimble startups, this is a golden opportunity. It’s like a neon sign flashing, “Disrupt me!”

What This Means for SaaS Builders

If you’re building a product today, here are some key lessons:

1. Rethink workflows from the ground up: Don’t just sprinkle AI on top. Ask yourself, “If we were designing this today, with AI from the start, what would it look like?”

2. Focus on outcomes, not features: Users don’t care about AI—they care about results. Can they do their jobs faster, easier, or better?

3. Build data flywheels: The winners in this space will be companies whose products get smarter with every interaction

4. Simplicity is key: The best AI is often invisible—it just makes the experience feel better

The Clean Slate Advantage

Here’s a paradox: in today’s AI-driven SaaS world, starting from scratch can be an advantage. Without the weight of legacy systems or outdated assumptions, you can build AI at the core—not as an add-on.

This might explain some of the excitement around new players in the market. They’re not retrofitting—they’re rethinking entire workflows with AI as the foundation (and with a lower cost structure).

The Road Ahead

Klarna parting ways with Salesforce and the rise of AI-first startups are signals of a broader trend. Soon, the line between “AI-powered” and “regular” software will blur, maybe even disappear.

The real winners won’t be the ones with the most AI features. They’ll be the companies solving real problems with AI in ways that feel natural and intuitive.

The real winners will be the ones who look back and say, "Your upsell was my opportunity."

So the next time an AI upsell lands in your inbox, don’t dismiss it too quickly. There might be a million-dollar opportunity hidden in there.

This book:

1. Demystifies the black box of AI.

2. Lays out AI's strengths and limitations.

3. Connects AI to its broader societal impact.

4. Reads like a novel while packed with technical concepts.

Let's dive into some key takeaways.

1. From Narrow Classification to Holistic Perception

Dr. Li chronicles the pivotal shift from rigid classifiers to neural networks. The catalyst? Data, along with cheaper computing power. Early AI models like Support Vector Machines (SVMs) did well with small datasets but quickly hit a ceiling. Neural networks, on the other hand, have much more headroom but need large datasets.

ImageNet, a massive image database created by Dr. Li's lab, gave neural networks a turbo boost. As these models processed more examples, they developed a more nuanced, human-like understanding of visual information. It was a leap toward generalizable intelligence, mirroring how human brains learn from diverse experiences.

Key Takeaway for Operators: The success of neural networks underscores the importance of data as a competitive advantage. Players that can acquire, curate, and leverage unique datasets may find themselves with a significant edge in the AI race.

2. AI as a Transformative Force

This assertion is a wake-up call for both technologists and domain experts. Technologists need to approach collaboration with humility, while domain experts should approach AI with curiosity.

The true game-changers won't just be AI companies; they'll be the ones that deeply understand AI and use it to reimagine entire industries.

Action Item: Foster cross-disciplinary collaboration in your organization. Create opportunities for AI experts and domain specialists to co-create, not just co-exist.

3. Human-Centric Development: Starting with "Who"

Dr. Li emphasizes centering AI development around human needs and experiences. This shifts the focus from technology-first to people-first, ensuring that AI solutions address real-world problems effectively.

She also shared this reflection with Dr. Terry Platchek:

The next wave of AI services should be our trusted sidekicks, empowering us to be our best human selves.

It's a reminder that our goal should be to amplify what makes us human, not replace it. The next wave of AI-powered services should be our trusted sidekicks, empowering us to be our best human selves. In healthcare, that means helping professionals operate at the top of their license, with AI as a critical ally.

4. Reimagine, Don't Just Increment

In the current AI gold rush, there's a temptation to tack on AI features and call it innovation. The real opportunity, though, lies in reimagining user experiences from the ground up.

Challenge to Incumbents: Don't just slap an AI feature on your product and hike up the prices (those juicy upsells!). With AI making building easier and capital eager to fund new ideas, we're entering a new era. The real winners will be the visionaries who reimagine entire experiences from the ground up, not just those making small, incremental improvements. The future winners will be those who look back and say "Your AI myopia was my opportunity”.

Looking Ahead: Cross-Pollination Unlock

Dr. Li's work is shining a light on something huge: the collision of AI smarts and real-world expertise. It's like we've stumbled onto a goldmine of potential, where tech geeks and industry pros can team up and create magic.

Just imagine for a sec: what if the experts in every field approached AI with the same wide-eyed excitement as the researchers building it? That's the kind of mashup that could blow the doors off innovation.

So there I am, tapping away at my laptop in some hip SF bar. On my left, there's this guy who's all in on AI. On my right, someone's going on about how AI is just another toy for the rich to get richer. Classic SF, right?

But here's the thing: Dr. Li's journey is showing us that the real breakthroughs are gonna come from the folks who can bridge that gap. The ones who get both the tech and the human side of things.

So here's the homework for us: how are you gonna take these ideas and run with them? Are you gonna let AI be just another tool, or are you gonna use it to completely reimagine what you do?