The Line Is Gone

Why the Border Between Software and Services No Longer Holds

For forty years, software and services were two different businesses. Software sold a tool; services sold a person. Software was priced by seat; services by hour. Software ran at 80 percent gross margin and 6–15× revenue multiples; services ran at 30 percent and 1–4×. Every strategic question in enterprise technology routed through that binary.

In 2023, the middle started filling in. A new class of business began to emerge that is neither a software company with a services wrapper nor a services firm with a software tool. It is the category the dominant companies of the next decade will occupy — and most of the market has not yet priced that it exists.

The shape of this transition is not new

The transition to personal computers produced this shape. I lived through it. The PC arrived, the generation above me was certain the computer would take their jobs, and a real short-run displacement happened. Then the long run arrived — and before the PC, there was no software-engineering industry, no hardware industry as we now understand it, no Microsoft, no Google, no AWS, no social media. The wealth created on the other side of this shift was an order of magnitude larger than the wealth lost to the displacement. The same shape recurs across the railroads, air travel, the telephone, and the commercial internet. AI is running the pattern again.

The formalization of software and services were two ways of organizing a particular shape of economic output when the underlying labor was human. When the labor becomes an agentic workforce — a standing capacity of reasoning agents with deep domain workflow wrapped around them — the categorical distinction no longer bears relevant meaning. The primitives reorganize. The firms that will define the next decade need to operate with two distinct postures running together: AI-first strategy and AI-native architecture. I draw the distinction precisely in a later paragraph because the terms are often used interchangeably, and they shouldn’t be.

Software and services were two ways of organizing the same output when the underlying labor was human. The labor changed.

The line between software and services is blurry

The old mental model put software and services in opposition. Software was leverage; services were effort. The whole capital-markets architecture sat on top of that opposition. It collapsed because the binary shattered. When a software product is delivered by an agent that can read the primary source, apply the framework, and produce a defensible draft, the product is no longer a tool; it is a worker. When a services engagement is delivered by a system that does most of the work without a human in the loop, it is not really services anymore. Agents are both software and services at once. The buying categories no longer match what is being sold, and the mispricing in that gap is where the asymmetric returns live.

The services TAM is no longer locked behind an army of advisors

The reason services are one of the largest addressable markets in the economy is that the work inside it — tax, audit, accounting, FP&A, and compliance— has historically required human experts. The buyer who wanted the output had to buy the army that produced it. The firms producing the output were Big Four, top-tier advisory, premium boutiques — organizations whose competitive moat was the pyramid of trained humans. Agentic AI has unlocked the TAM fortress. Reasoning plus deep domain workflow now produces an output a senior partner can sign. The moat was not expertise; the moat was the army, and the army is replaceable. The expertise itself — centuries of accumulated domain knowledge in tax, audit, accounting, and the rest — does not disappear with the army, it becomes more potent. Agents carry the volume; senior practitioners supply reasoning at the points where the work is messy and the combined output is a white-glove service delivered at a fraction of the prior cost structure.

A services market measured at roughly $1 trillion in third-party advisory and outsourcing fees globally — and several trillion more in the in-house labor that performs the same work today — is suddenly in the addressable set. The broader global professional services economy within which these submarkets sit is estimated at roughly $6 trillion[1]. That is not a software opportunity. It is a services-market restructuring. The software opportunity is roughly one-third the size.

Buyers want outcomes — not subscriptions, not hours

Ask any CFO what they bought when they bought a SaaS subscription and the honest answer is: I bought a promise that I would build the outcome internally. The buyer’s effective cost is the recurring subscription, and the FTE payroll required to run it, not counting the implementation and configuration. The published subscription price is a fraction of the true all-in cost. The question surfacing in buying committees right now is the one nobody could ask five years ago: if the vendor can deliver the outcome directly, why pay for the SaaS and staff the SaaS? When that argument expires, the conversation shifts from seats to outcomes.

The services side has a symmetric problem. Billable hours create an adversarial contract — buyer minimizes, firm maximizes, both sides budget for scope creep. When the marginal cost of the next unit of output approaches zero, the constraint on pricing is no longer the hour. It is the outcome. Outcome pricing is also how the productivity gain from agents flows to the firm as margin rather than to the buyer as discount.

AI-first businesses need to rebuild their business model

AI-first and AI-native get used interchangeably — they shouldn’t be. AI-first describes a strategy that prioritizes AI in how the company operates and allocates resources, even if not every layer has been rebuilt on AI from scratch. AI-native describes something fundamentally built on AI as its core fabric — AI is not an addition but the foundation upon which everything else rests. Asking the team to automate their own workflows so they can move closer to customers is an AI-first strategy. Building the data foundation that an agentic workforce stands on is AI-native work. The two are complementary, not synonymous. The category-defining firms over the next decade will need both.

This is the part founders and CEOs most consistently get wrong. The business model itself must be redesigned. Four shifts, at minimum. Pricing moves from seats and hours to outcomes. Sales move from features to guarantees — the conversation is no longer about what the product does but what the firm will deliver. Hiring inverts: entry-level compresses because agents handle preparation; principal-level expands because judgment, accountability, and escalation become the scarce resource. Contracts shift from tool provider to accountable counterparty. The firm carries the risk of the outcome.

Tech-enabled services achieve SaaS economics — and the re-rating follows

The thesis I want to defend is not that AI-Native services firms have already arrived at SaaS economics. It is that they are on the curve, the curve is bending, and each iteration of the underlying agentic stack pulls the unit economics meaningfully closer.

The data we have so far supports the trajectory. ICONIQ’s State of AI report, surveying ~300 builders, puts AI product gross margins at 41% in 2024, 45% in 2025, with builders projecting 52% in 2026^[2]. Best-in-class is already further along: Bessemer’s “Shooting Star” cohort is averaging ~60% gross margins, with their leanest software-disguised-as-services companies running revenue per FTE above $1 million^[3]. Compare that to the traditional services baseline visible in public 10-Ks: Accenture at ~32% gross margin and ~$84K revenue per employee, the offshore IT/BPO majors at 27–35% and $50–90K, the Big Four at ~$130–150K^[4]. The gap between best-in-class AI-Native services and traditional services is real and wide. The gap between average AI-Native and traditional is narrower but moving in the right direction at roughly four to seven percentage points of gross margin per year.

The mechanism is the agentic workforce, and here the honest picture is more nuanced than the headline numbers suggest. Peer-reviewed research clusters realized productivity gains in expert services workflows at 12–40% while many firms anecdotally quote 70%+ gains. The truth lies somewhere in the middle when accounting for hallucinations, learning curves and token utilization. However, it is still transformative against a 28–36% gross margin base. It just is not yet the single-step jump to software economics.

The valuation re-rating follows. Services have always traded at 1–4× revenue; software at 6–15×. Two firms in the same vertical, delivering the same output will be worth very different multiples a year from now. The re-rating runs in three phases: private-market repricing now, strategic M&A over the next 18–24 months as legacy advisors and incumbents buy delivery-model capability, and the public-market repricing in 2027–28 once the first AI-native services IPO sets the comparable.

The re-rating is not a thesis waiting to be proven. It is a repricing already happening in private markets, not yet priced in public ones.

The target markets are right in front of us

The services TAM is hiding in plain sight, sitting on the line item that runs the modern economy. The US Professional, Scientific and Technical Services sector — NAICS 54, the official taxonomy for legal, accounting, consulting, engineering, design, R&D, and the rest of the professional-work economy — is a $3.2 trillion industry domestically[5]. The global picture roughly doubles that: published estimates put the global professional services economy at approximately $6 trillion[6], with the rest of the world contributing roughly $3 trillion on top of the US share. That is the universe inside which the agentic restructuring will play out.

For a FinTech operator anchored in international tax, indirect tax, and incentives, the cluster that matters sits inside the IBISWorld parent category that contains audit, tax preparation, bookkeeping, and adjacent advisory work — is $644 billion globally[7]. That universe shares a common substrate of expert-driven financial workflow: tax($40B)[8], audit ($250B)[9] , FP&A ($15B)[10] with the remainder distributed across bookkeeping, payroll, and other accounting services. These are the functions where the old model collides hardest with agents. The Big Four army. The billable hour. The partner who hoards the domain expertise.

A note from inside the transition

It is easy to write this thesis in the abstract; it is different to having lived inside the transition. I have spent the last several years moving across four vantage points that cover most of the angles in this essay.

At Pioneer Square Labs and AI Fund — two of the pre-eminent startup studios — I have had a front-row seat on the pace of innovation and formation in the agentic era. Teams assembled from a thesis, primitives tested in weeks rather than years, the patterns that survive contact with customers separating quickly from the ones that do not. The studios are where the velocity is most visible. They are also where the line between software and services has all but disappeared. The founders building today aren’t choosing one or the other. The market that still organizes around that binary is organizing around the wrong question.

At LumaTax — which I founded in 2016 and merged with Taxually in 2023 — I was building directly inside one of the services markets this essay names. Tax compliance, expert-dependent, dominated by a handful of incumbents for decades. The seams in the legacy delivery model were visible from the founder’s seat, and the early version of the restructuring this essay describes had begun to surface.

At Exactera, where I took the CEO seat with the mandate to build the next generation of AI-powered services, the playbook this essay describes is the one we are running. We made a concrete decision: a firm organized around an agentic workforce delivering against defined buyer outcomes, with software as the primary unit of production. The decision required both postures running together — the architectural and strategic. The AI-Native rebuild of the data foundation, workflow primitives, and encoded institutional knowledge the agents stand on, paired with the AI-first discipline of prioritizing that rebuild over the legacy investment producing this quarter’s revenue. Either one alone is a stuck firm.

What we deliver is the outcome a senior partner at a top-tier firm used to produce, assembled by the agentic workforce with our practitioners supplying reasoning where complexity genuinely demands it. Centuries of domain expertise live inside the workflows the agents execute, and our people apply judgement where the stakes do not tolerate ambiguity. We call this “Verified by Design.”

The most important operating move was asking the team to automate their own work. The framing mattered. This was not a cost-reduction program. The directive was: identify the manual work eating your time, deploy the automation that absorbs it, and redirect the recovered capacity toward customer-facing and high-value work — the work that compounds. The goal is customer obsession, delivered by putting our most capable people closer to our customers. Across the workstreams we measured, the result was up to 85 percent reduction in manual workflows inside the affected functions. That is a capacity-redeployment number, not an efficiency number. The firm did more, not less. The thesis in this essay is not an external observation about other firms. It is what I have watched happen inside my own.

What this means over the next eighteen months

The category boundary that organized enterprise technology for forty years has dissolved. The services TAM has opened up, the buyer wants outcomes, the billable hour is losing its grip, AI-first firms can achieve software economics while delivering services output, the valuation re-rating is in motion, and the target markets are hiding in plain sight. The window for founders and investors who want to be early is 12 to 18 months. After that the categories lock in. From here, the winners look like asymmetric bets hiding under legacy labels. From 2028 they will look like inevitability.

---

[1] Global professional services market, $6.07 trillion in 2024, encompassing legal, accounting, consulting, design, research, and other professional and technical services. Source: Research and Markets / The Business Research Company, Professional Services Global Market Opportunities and Strategies to 2034 (projecting growth to $7.89 trillion by 2029). The “in-house labor” estimate refers to enterprise spending on internal staff performing the same professional services work (finance, legal, HR, compliance, advisory, etc.) — a category that typically exceeds third-party services fees by a substantial multiple, as most enterprises retain the bulk of these functions internally rather than purchasing them from outside firms.

[2] AI product gross margins of 41% (2024), 45% (2025), and 52% (projected 2026), based on a survey of approximately 300 AI builders. Source: ICONIQ Capital, State of AI 2025; summary at SaaStr, “AI Gross Margins Are Up: 5 Key Takeaways from ICONIQ’s State of AI Report”

[3] “Shooting Star” cohort of high-growth, capital-efficient AI-native companies averaging ~60% gross margins, with the leanest companies (the “Supernova” subset) running average revenue per FTE above $1.1 million. Source: Bessemer Venture Partners, State of the Cloud / Top AI Startups Benchmarks.

[4] Traditional services baseline reconstructed from public filings: Accenture FY25 gross margin 31.9% and revenue per employee ~$84K; Cognizant ~34.5% and ~$59K; Capgemini 27.4% and ~$70K; Infosys 30.5% and ~$57K; TCS ~39% and ~$50K; Genpact 35.5% and ~$34K; IBM Consulting 27.0%; EPAM 30.7%; Globant 35.7%. Big Four global revenue per partner-track professional approximately $130–150K. Sources: Accenture FY25 Earnings Release; Infosys Form 20-F FY2025; Capgemini FY 2024 Results; Genpact FY24 Results; IBM 10-K FY2024

[5] US Professional, Scientific and Technical Services sector (NAICS 54), $3.2 trillion in 2025–2026, growing at 2.7% CAGR (2020–2025). Source: IBISWorld, Professional, Scientific and Technical Services in the US.

[6] Global professional services market, $6.07 trillion in 2024, encompassing legal, accounting, consulting, design, research, and other professional and technical services. Source: Research and Markets / The Business Research Company, Professional Services Global Market Opportunities and Strategies to 2034 (projecting growth to $7.89 trillion by 2029). The “in-house labor” estimate refers to enterprise spending on internal staff performing the same professional services work (finance, legal, HR, compliance, advisory, etc.) — a category that typically exceeds third-party services fees by a substantial multiple, as most enterprises retain the bulk of these functions internally rather than purchasing them from outside firms.

[7] Global Accounting Services market (IBISWorld code L6713-GL), $643.8 billion in 2025–2026, including auditing, tax services, bookkeeping, and other accounting services. Source: IBISWorld, Global Accounting Services Industry Analysis.

[8] Tax advisory services, $36.4–42.5 billion globally (2024). Sources: Straits Research, Tax Advisory Services Market; Market Research Future, Tax Advisory Services Market.

[9] Auditing services, $236–292 billion globally (2024). Sources: Research and Markets / Global Industry Analysts, Auditing Services — Global Strategic Business Report; Market Data Forecast, Auditing Services Market.

[10] Outsourced FP&A as a sub-segment of the global Finance & Accounting BPO market ($60–67 billion, 2024). Sources: Grand View Research, Finance and Accounting BPO Market; Business Research Insights, Finance and Accounting Outsourcing Market.