A services firm that ships software, prices against outcomes, and reuses agents across customers is not the same business as a traditional consulting firm with AI bolted on. Its unit economics, its scaling curve, and its exit math all differ, and the three differences together are the strategic case for the model.
Three components that have to work together
The economic structure has three components that have to work together. Services pricing, software economics, and end-to-end automation of the discovery-to-deployment arc. The customer relationship is worth multiples of a SaaS seat (because the customer is paying for the work to get done, not for access to a tool, and the work has services-tier pricing power). Marginal engagements do not require marginal teams (because the agent infrastructure built for previous customers handles most of what each new customer needs, and the human team scales sub-linearly with customer count). The discovery and integration work that consumed most of the cost in the consulting model is automated (because the agent does the discovery as a side-effect of being installed, and the same agent handles the integration as part of the production deployment). Each of these three components is necessary; none of them is sufficient on its own. A consulting firm that adopts outcome pricing without the agent reuse is still a consulting firm; it just has a different P&L. A SaaS firm that adopts services pricing without the embedded delivery is still a SaaS firm; it just charges more per seat. The combination is what is novel.
Why agent reuse changes the scaling curve
The scaling curve is the property that makes the model venture-backable rather than just operationally interesting. A traditional consulting firm scales linearly with headcount: revenue equals consultants times utilization times bill rate, and the next dollar of revenue requires the next consultant. A services-as-software firm scales the way software does once the agent reuse compounds. The first ten customers cost most of the engineering work to build the underlying agent infrastructure. The next hundred cost a fraction of that effort, because the infrastructure already exists and each new customer's needs are a configuration of capabilities the infrastructure already supports. The next thousand cost a fraction of that. The marginal cost of delivery falls as the customer count grows. That falling marginal cost is what makes software a venture asset class; its absence is what makes traditional services a private-equity asset class.
The labor TAM versus the software TAM
The exit math, the third component of the strategic case, is the one that public-market and growth-stage investors are still working through. Sequoia's framing is six dollars of services spending for every one dollar of software spending; the labor TAM that AI now addresses is six times the size of the software TAM that SaaS addressed in the previous era. General Catalyst puts annual US services spending above $6 trillion, against a software market it sizes at $370 billion. Foundation Capital's headline number is the $4.6 trillion services TAM of its April 2024 piece, split evenly between salaries and outsourced services. The multiple depends on what gets counted, but every version of the comparison puts the addressable market for services-as-software companies roughly an order of magnitude above the market SaaS addressed, and the exit math for the category starts from that larger base.
Addressing the AI roll-up counter-thesis
The counter-thesis deserves a direct answer, because the most sophisticated objection to the venture-scale framing is the one Nathan Benaich and Nikola Mrkšić made in Fortune in June 2025. Their piece argued that AI roll-up investors think services firms can trade like software companies, and they are wrong. The data behind the piece is real: Concentrix has deployed gen-AI at more than a thousand customers, yet its EBITDA margin still hovers around ten percent and its EV/EBITDA multiple remains stuck in the low single digits, and the piece groups Genpact and Infosys with it among outsourcers trading between five and twenty-three times EV/EBITDA. The conclusion is the one to reckon with: AI roll-ups may still deliver returns, but not the kind VCs are underwriting; at best, tech-enabled PE.
The distinction to draw in response is the one between services-as-software and the AI roll-up, and it cuts through the middle of the new cohort rather than around it. Concentrix is AI bolted onto a traditional services business, so its pricing power is bounded by what the customer used to pay for the non-AI version of the same service; the customer can always go back to a non-AI provider in the same category. Much of the new cohort is acquisition-led rather than built from scratch: Crescendo acquired the outsourcer PartnerHero and its 3,000 human support professionals, Eudia acquired the 300-person legal services provider Johnson Hana, Crete plans $500 million to buy US accounting firms and upgrade them with AI, and Long Lake has raised $670 million to buy homeowners-association managers. Those four are bets that an acquired services business can be rebuilt to software economics faster than the Concentrix precedent suggests. The cleaner test of the thesis is the firms with no legacy book at all: Crosby, the Sequoia-backed law firm whose lawyers work with its internally developed AI, and Manifest OS, which raised $60 million in April 2026 to build and operate its own AI-native law practices. The customer of a firm built this way is comparing the offering to the in-house team they do not have, so the pricing reference shifts from cost-of-service to cost-of-labor-displaced, which is the order-of-magnitude larger number. The exits will trade between the two depending on which model the firm actually built, and the firms that built the second kind will trade like the second kind even if they have services as a delivery mechanism.
Why the market window for services-as-software is open now
The market window for building this kind of firm in this category is open now and will not stay open indefinitely. Customer demand is real and growing, because the buyer is the operating business that all four existing delivery models leave behind. The technology is finally adequate, because foundation models can do the unstructured-input judgment work that was the structural bottleneck for fifteen years. The talent is reorganizing around the forward-deployed engineer, the role this delivery model runs on. And the capital is committed, most visibly into the adjacent roll-up version of the thesis: General Catalyst says it has committed billions of dollars to its AI-enabled roll-up strategy, and Thrive Holdings, the holding company spun off by Thrive Capital, is committing $1 billion to an AI-powered accounting roll-up. The next several years will determine which firms reach scale first, after which the category will consolidate around the winners.
Horizontal versus vertical services-as-software bets
A reasonable counter at this point is that the horizontal services-as-software bet (across verticals) is harder than the vertical bet (one industry at a time), and the vertical firms will win because the data and integrations compound faster within a single vertical. There is something to this, and the vertical bets do compound faster inside their chosen industry, but the horizontal bet wins on a different dimension. The agent infrastructure that does observation, redesign, and deployment is the same regardless of the customer's industry, so it reuses across customers in different verticals in a way that the vertical firms' integrations cannot. The vertical firm accumulates a vertical's worth of data while the horizontal firm accumulates the discovery-to-deployment infrastructure itself, and both can be venture-scale without being in direct competition, because they capture different parts of the same market. Flowscope is one of the firms building the horizontal version of the bet.
The bet is straightforward. The largest unaddressed market in enterprise AI sits in services work that the four existing delivery models do not serve, the technology to address it now exists, and the services-as-software firms that will serve it are being built now. The question is which of them reach scale first, and flowscope is building one of the contenders.