If you're running a software company - or a services firm - the question of what business you're actually in has gotten significantly harder to answer in the past 18 months. Not in a philosophical sense. In a very practical, board-meeting, compensation-plan sense.

The traditional distinction between a product company and a services company used to be clean: one sells access to a tool; the other sells the output of human expertise applied to a problem. Different gross margins, different hiring profiles, different sales motions, different valuation multiples. Investors in both public and private markets strongly favour the former category.

A lot of commentary in recent months has focused on the death of per-seat pricing: Kyle Poyar wrote about it; Martin Casado at a16z explored it in a podcast with Metronome CEO Scott Woody. Most of that conversation treats pricing as the headline story. We think it's a symptom. The underlying shift is larger, and its consequences extend well beyond what appears on a software company’s invoice.

In our view, successful AI companies of tomorrow will look a lot less like traditional software products and a lot more like services firms. And the services firms that survive will look a lot more like software companies. AI is blurring the line between the two, and the businesses that recognise this early will have a structural advantage at every renewal and engagement conversation.

To make the case, it helps to look at both business models through a simple lens: what happens to the bottom line, and what happens to the top line, as AI adoption gains ground.

Software businesses were built on a financial profile that investors loved: gross margins of 70–80%, low cost of goods sold once the core product was built, and revenue that scaled without proportional headcount growth. Services businesses - consulting firms, implementation partners, managed service providers - ran at 25–40% gross margins, because delivery required people, and people are expensive and unpredictable.

That gap is closing from both directions.

A16z started the conversation in February 2024, identifying three reasons why AI companies look structurally different from the SaaS businesses that preceded them: heavy cloud infrastructure costs that depress gross margins; persistent scaling challenges from edge cases that require ongoing human involvement; and weaker defensive moats as AI models commoditise. Their expectation was that AI company gross margins would land in the 50–60% range - well below the 60–80% benchmark for comparable SaaS businesses. TechCrunch weighed in saying “If a16z is correct about AI startups having slimmer gross margins than SaaS companies, they should — all other things held equal — be worth less per dollar of revenue generated; or in simpler terms, they should trade at a revenue multiple discount to SaaS companies, leaving the latter category of technology company still atop the valuation hierarchy.”

The market has largely absorbed this recalibration. Investors who once required SaaS-style margins to justify SaaS-style multiples have, somewhat pragmatically, concluded that high infrastructure costs and people-intensive deployments don't necessarily stand in the way of building a great business. Gross margins don’t really matter and we should rather be focused on “gross profit per token”. 

Now consider what is happening on the services side. Management consulting has historically achieved gross margins of 40–55%, with that cost almost entirely attributable to labour. As firms augment or replace portions of their delivery capacity with AI agents, that cost structure changes. The labour line shrinks. The infrastructure line appears. The resulting gross margin profile starts to look, in an uncanny way, like what a16z described for AI companies.

Software is coming down. Services are coming up. The convergence isn't coincidental. The same underlying technology reshaping both models simultaneously.

The revenue side of the question is where the shift becomes most visible and most consequential.

We wrote about Bret Taylor’s interview with John Collison a few weeks back. Sierra’s pricing model was a large part of that conversation. Taylor’s highly successful AI company charges the customer a pre-negotiated rate only if the AI agent successfully resolves a customer's case without any human intervention. If the case has to be escalated to a human representative, Sierra provides the service for free. Historically, there was a stark divide between software creation, implementation, and usage. When an enterprise rollout failed, software vendors, IT implementers, and clients would all blame each other. Outcome based pricing transforms the value equation and aligns all stakeholders. Taylor goes even further, saying that by charging for outcomes rather than usage, Sierra is incentivised to drive down its own token utilization and compute costs behind the scenes. Because reducing token usage becomes Sierra's problem rather than the client's problem, Sierra is driven to constantly make its underlying product more efficient. 

That is not software pricing. That is the structure of an outsourced services contract.

Intercom has been on a parallel journey, though starting from a very different position. Eoghan McCabe wrote about it recently, describing how the company doubled down on its own AI-powered chatbot called Fin, what it took operationally to get there and how the company had to transform itself to stay relevant.

Both Sierra and Intercom operate in customer service, where success is clearly defined and easily measured. That makes them the natural early movers. The transition to outcome-based pricing will be harder and slower in functions where value is less cleanly quantifiable but the direction feels inevitable.

There is also a middle ground worth noting, away from pure outcomes-based pricing. This week, I spoke with the founder of an early-stage AI product company - serving top-tier consulting firms and investment funds - who described flipping his business model in the space of a few months. He started out charging a monthly per-seat fee in the thousands of dollars. He now charges a flat $50 per seat, with clients bearing their own token consumption costs directly, plus a separate implementation fee.

What his new model reveals is something important: customers are implicitly paying for three separate things, even if the invoice doesn't break them out that way. The first is compute: the underlying LLM and infrastructure. The second is workflow design: the consulting-equivalent work of understanding the unique process and configuring the agent. The third is ongoing execution: the automation layer that runs the workflow reliably, even when the employee running the workflow is having a bad day. These have always been traditionally bundled inside a per-seat fee. Unbundling them makes the services-like nature of the offering visible.

Scott Woody, CEO of Metronome, put it clearly on the a16z podcast: the modern CFO has two dominant line items: headcount (the biggest), and AWS (usage-based, variable). Finance leaders have spent 15 years learning to manage variable cost structures. And from that perspective, usage and outcome based pricing is the natural next step.  

Pull these threads together and a picture emerges. AI is making the top line and bottom line of software and services businesses look more alike than different. Customers are increasingly demanding that what they pay be tied to the value they receive. From that, it is not unreasonable to suggest that the dominant B2B vendor of the future is a hybrid: a software-service-agentic company that charges for outcomes, scales through automation, and earns its margin by encoding process knowledge into agents in addition to hiring people and selling access.

We are already seeing the early shape of this in customer service with Sierra and Intercom. Other functions will follow.

Which brings the harder question: who gets there first? Software companies or services firms?

There are strong arguments on both sides.

Software companies have distribution, tech talent, and a track record of building products that scale. Established players, in particular, have the customer relationships and the technical infrastructure to move quickly. The bias against building point solutions for one client, however, is real. Most software companies are deeply uncomfortable with the bespoke, process-specific work that building a truly capable agent often requires, and equally scared of disruption their still very profitable core business.

Services firms, on the other hand, have other advantages: deep, tested, documented knowledge of how complex business processes actually work in practice, not in theory. An agent that manages credit underwriting, runs a portion of a due diligence process, or handles compliance review requires a deep understanding of each client’s unique process. The entity that understands that process at a level of detail sufficient to encode it reliably into an agent has a structural advantage that distribution alone can't replicate.

I suspect we will see, within the next two to three years, a cluster of former consulting firms that have repositioned as AI product companies, and a cluster of software companies that have quietly become managed services providers without fully acknowledging it. Are incumbents better positioned to capitalise on this opportunity than AI-native newcomers? Time will tell, and I’ll refrain from offering an opinion here as a matter of principle it's bad for your soul to spend too much of your time “talking your book”.

— Daria