Every platform shift creates its own delivery model.
Much like how calculators drove the marginal cost of arithmetic to nearly zero, we now enter an era where the marginal cost of reasoning approaches zero. Intelligence, which was once uniquely human, is becoming abundant, delivered not only faster but directly through agents.
For decades, software was built to help humans do their work faster, cheaper, and better. Now agents are increasingly capable and, at times, more than capable of doing the work themselves — reshaping the labor market in ways software alone could not. It is both astonishing and unsettling to imagine a world where agents deliver reasoning, and ultimately “intelligence,” which is commoditized over time.
This transition creates both opportunity and tension. Tools are rapidly gaining capability, but adoption lags. Enterprises measure success in outcomes, not consumption, which makes the delivery model itself the critical unlock.
This shift opens a market far larger than software has ever addressed. For every dollar spent on software, there are hundreds spent on wages and headcount. Enterprise software is a $900B market, yet labor spend runs into the tens of trillions. Capturing even a fraction of that through AI agents represents a TAM an order of magnitude greater than all of enterprise software combined.
At Madrona, we believe that the biggest opportunity for AI is targeting labor directly. Meeting the market where it is and operationalizing outcomes is the foundation for our thesis: “Service as Software.”
The Awkward Transition Phase for Widespread AI Adoption
Historically, the mention of “services” entailed low-margin grunt work that overcompensated for the lack of product usability. Especially in the eyes of venture investors, services meant low gross margins, messy delivery, and businesses that didn’t scale well or trade at attractive multiples. But with advancements in LLMs surpassing the rate of market adoption, there is a lot to be unlearned.
In other words, we are in an awkward transition phase as agents are gaining intelligence, but humans still require clear direction and white-glove implementation. Where the capability of the tools in many verticals exceeds the user’s ability, and where product is still priced by consumption, but customer success is measured in outcomes. In fact, if the business model for AI trends toward outcome-based pricing, as many have predicted, the need for successful product delivery and customer success has never been higher.
The collision of an immature market and the desire for mainstream, widespread AI adoption that the market craves — since it underpins the entire thesis behind the massive CapEx dollars that has been invested in AI infrastructure (1.2% of GDP and growing) — has cultivated the perfect condition for services to be delivered in a software-like workflow to customers, delivering real outcomes as the atomic unit of value.
We believe “service” is more about deliverability and success than it is about the service itself. It drives adoption at a level that optimizes usability of the tool and ultimately bridges the gap between prototype and production, between theoretical claims and tangible outcomes. It’s not a surprise that “forward deployed engineers” is one of the fastest-growing roles for many software companies today! These teams aren’t doing custom consulting; they’re accelerating the productization of delivery itself. In doing so, they collapse the distance between what’s technically possible and what’s commercially adoptable.
As an interesting comparison, in the years leading up to its IPO, 30-40% of Workday’s revenue came from professional services. Similarly, yet at $800M of ARR, over 25% of Qualtrics’ revenue came from professional services. And Informatics General, one of the first enterprise software businesses in the 1970s, saw professional services make up nearly 40% of revenue – proof that this pattern has been with us for half a century. In fact, the cost of implementation of software ranges anywhere from 1-3x the cost of the software license itself. This model has proven to be the scaffolding that holds up software until the flywheel can spin on its own.
In today’s day and age, many AI products will need to leverage the “Service as Software” model to meet the market where it is. Over time, we expect ample opportunity to productize elements of the service workflow as the market matures, which simultaneously creates trust, operating leverage, and durable value for the customer. In many ways, while the hurdle for software creation has never been lower in terms of capabilities, the need for an outcomes-driven model for AI software has never been higher.
Next Generation of Applied AI
We think of the distribution of value in the market as an hourglass. In this super-cycle, massive value has already been accrued at the infrastructure layer, but we continue to believe that there will be an even bigger value unlock in the application layer. And we are in the early innings of value capture. As with the “O-ring” model in economics, where the overall output is limited by the least effective component, the application of context, data, and workflow will prove to be the enduring moat.
That said, applied AI will look different in the years to come. We can’t circumvent this awkward phase where tools are highly capable, but users and adoption are not. Our collective embrace of AI tools and the potential they represent must be complemented by our acknowledgement of the need for value delivery to drive product adoption, customer success, and, ultimately, outcomes as the atomic unit of value in AI. This is why we believe “Service as Software” is more relevant today than ever before.
We are on a journey of incessantly going after the largest pools of labor spend, unveiling what we believe are the best companies within each category. We have already backed a handful of incredible founders with this vision, leveraging the best of what AI models and tools have to offer and marrying them with tangible business outcomes that customers care about, fueling their incredible growth.
- GrowthX is replacing agencies by combining expert talent with AI-powered growth marketing workflows. Their model isn’t to sell a tool. It’s to deliver a result: scalable acquisition, continuously optimized.
- Trek Health transforms pricing data into mission-critical insights for healthcare payers and providers. The insight is the product, but service is how it gets delivered, adopted, and trusted.
- Fyxer AI is building a reasoning layer for knowledge workers, automating the workflows traditionally handled by assistants or chiefs of staff. But instead of offering a blank-slate tool, Fyxer focuses on doing the work, right away, and learning from every interaction.
- Gradial provides content supply-chain agents that streamline your content operations and accelerate your digital experiences. They focus first on execution and then build toward a sustained, self-reinforcing strategy flywheel
And we are doubling down on our thesis and looking for the next generation of founders who are riding the AI innovation wave and disrupting the labor markets across every vertical, one “forward deployed engineer” at a time! Give us a shout at [email protected] and [email protected] to share your feedback and vision for the future of applied AI.