AI as a New Kind of Labour

The reframing: from software you use to labour you manage

The defining rhetorical move of the 2024–25 enterprise cycle was to stop selling AI as software and start selling it as workers. Salesforce announced Agentforce 2.0 as “the digital labor platform for enterprises, enabling a limitless workforce through AI agents for any department,” and Benioff framed the company as “the leader in digital labor solutions,” letting “any company … build a limitless workforce.”¹ A year later, the Agentforce 360 launch generalised the move into a stated vision of the “Agentic Enterprise … where AI doesn’t replace people, it elevates them,” connecting “humans, agents, and data on one trusted platform.”⁷ The language is deliberate: a workforce, not a product; build and manage, not deploy and use.

The framing is not confined to one vendor. At Davos in January 2025, Benioff put it in generational terms — “Today’s cohort of CEOs will be the last to lead all-human workforces” — and described leaders as soon “managing not only human workers but also digital workers,” while insisting the technology is “supporting employees, not replacing them.”² Microsoft’s 2025 Work Trend Index introduced the Frontier Firm, an organisation “built around … intelligence on tap and human-agent teams,” and the agent boss: someone who “builds, delegates to and manages agents to amplify their impact.”¹⁰ Microsoft frames digital labour for these firms as being as critical to the business model as human labour.¹⁰ NVIDIA’s Jensen Huang, in his CES 2025 keynote, gave the management analogy its sharpest form: “In a lot of ways, the IT department of every company is going to be the HR department of AI agents in the future,” with agents that must be recruited, trained on company-specific vocabulary and policy, and supervised.³⁴

The investment thesis: “services-as-software”

Under the marketing sits a venture thesis that explains why the labour framing is more than branding. Foundation Capital’s Joanne Chen and Jaya Gupta named it a shift from software-as-a-service to service-as-software: where SaaS sells a tool and leaves the customer responsible for the outcome, the new model has the vendor assume outcome responsibility — “Instead of QuickBooks, you offer tax services … conducted by an AI accountant.”⁵ The size of the prize is the whole point. Foundation Capital frames a $4.6 trillion opportunity — roughly $2.3T of global salaries (sales and marketing, software engineering, security, HR) plus ~$2.3T of outsourced IT and business-process services — and is explicit that “the real prize isn’t the $200B SaaS market: it’s the $4.6T enterprises spend on salaries and services.”⁵⁶ The unit they keep returning to is the gap between a software vendor’s revenue and the wage bill it could displace: Salesforce earns ~$35B a year against the ~$1.1T companies spend annually on sales and marketing salaries.⁵

Foundation Capital’s year-one retrospective is useful precisely because it is less triumphant. The firms separating from the hype share three traits: forward-deployed engineers became “one of the most strategic assets,” because “integration is not a post-sale activity. It is the product surface”; the customer “expects to experience functionality, integration, and outcome before a contract is signed”; and pricing is moving from seats toward outcomes.⁶ The thesis, in other words, only pays off where the vendor can actually deliver an outcome — a caveat that matters when we reach the deployment data.

Software as capital that produces labour

The economic engine of the thesis is a reclassification: software stops being a tool that humans operate and becomes capital that produces labour directly. Rampell’s a16z essay states it as “software becomes labor.” The mechanism is that traditional enterprise software digitised offline processes — filing cabinets into Workday, tickets into Zendesk, ledgers into QuickBooks — but those systems “required human ‘users.’” The breakthrough is that “the ‘users’ of the digitized filing cabinet do not have to be humans.”¹² Once the user can be a machine, the relevant market is no longer the software budget but the labour budget — and the comparison is stark: ~$300B of enterprise software a year against a white-collar labour market of “many, many trillions of dollars a year.”¹²

This breaks the pricing model that built the incumbents. Workday, Intuit, Zendesk and Salesforce are per-seat businesses, and “a business will need fewer (if any!) seats as the system … takes actions on its own.”¹² Foundation Capital’s outcome-based-pricing argument is the same coin’s other face: charging for results “creates a much more scalable pricing model” and gives the vendor an incentive to expand usage rather than ration seats.⁵ Salesforce’s Agentforce skills are already priced per action — Sales Development and Sales Coaching at “$2 per conversation” — which is a per-unit-of-work price, not a per-seat one.¹ The labour-economics case is that capital scales in ways skilled labour cannot: “AI will always show up to work, can be trained instantly,” whereas nursing takes years and mortgage brokers could not be conjured when rates dropped in 2021.¹²

There is a serious academic articulation of this reclassification. Farach models AI as “agent capital” — a distinct production input that “reduces the friction of managing workers, expanding spans of control, compressing hierarchies.”¹⁶ Acemoglu, the leading sceptic, also frames AI in capital-vs-labour terms but predicts the gains accrue to capital while labour’s share erodes, with only “modest” aggregate productivity effects (below).¹⁴ The reclassification is widely accepted; what it produces is contested.

Uncoupling output capacity from headcount

The operational consequence vendors emphasise is the breaking of the link between how much a firm can produce and how many people it employs. Microsoft’s framing of the capacity problem is the demand-side case: 80% of the global workforce reports too little time or energy for the job, employees are interrupted “every 2 minutes,” and 53% of leaders say productivity simply must increase — so 45% of leaders name “expanding team capacity with digital labor” a top priority.⁹¹⁰ The supply-side promise is the vendor outcome data: Salesforce reports its own support agents resolving 83% of customer queries without human intervention, and customer deployments resolving large shares of work autonomously.¹⁷

Two structural signals support genuine uncoupling, both from Microsoft’s data. AI-native startups grew headcount 20.6% year-over-year against 10.6% for Big Tech — but the more telling number is forward-looking: 33% of leaders are considering headcount reductions, even as 78% consider hiring for new AI-specific roles (95% at Frontier Firms).⁹¹⁰ Capacity is being decoupled from bodies in both directions at once — fewer of some roles, net-new categories of others. But the honest empirical question is whether output-without-headcount is actually showing up in employment, and here the corpus is careful.

Org structure becomes a design choice, not an inherited constraint

If output capacity uncouples from headcount, two pillars of organisation design — the span of control (how many reports one manager can handle) and the number of layers between the front line and the top — stop being fixed by human cognitive limits and become parameters a firm can set. This is where the academic literature is most useful, because it complicates the glib “AI flattens the org” claim in both directions.

Farach’s model is the clearest statement of the design-choice view: because “agent capital reduces the friction of managing workers,” it directly expands spans of control and compresses hierarchies — and, importantly, the same technology forks into different outcomes depending on parameters. Under positive “task-creation elasticity,” AI “expand[s] the frontier of feasible work” rather than only cutting jobs; and an “elite-complementarity” parameter determines whether AI is general infrastructure (broad-based gains) or a tool that “amplifies top managers selectively” (superstar concentration). Structure becomes a chosen regime, and that parameter “becomes a policy lever.”¹⁶ The org chart is no longer downstream of human limits; it is a decision with distributional consequences.

Xu, Hou, Chen and Xie sharpen the nuance and puncture two simplifications. First, the much-discussed decline in junior employment “reflects deployment choices favoring automation over augmentation, not an inevitable consequence of GenAI itself” — directly echoing the Stanford finding that effects concentrate where AI automates.¹⁵¹³ Second, span of control does not simply widen: “across all four deployment architectures, the span of control initially contracts before eventually expanding” as the technology improves — because early, fallible AI needs more skilled oversight before it needs less.¹⁵ The design space they model is itself a 2×2 — automation vs augmentation, worker- vs expert-level deployment — which is exactly the set of choices a firm now has to make deliberately.

Orchestration and the emerging “agent manager”

The moment a firm runs more than one agent, it inherits a coordination problem that looks exactly like management. IBM’s reference definition is explicit that the dominant solution is a literal hierarchy: in hierarchical orchestration, “a manager agent decomposes a complex goal into sub-tasks and delegates each to a specialist agent. The manager monitors progress, handles failures, and reassembles results into a final output.”²² The other patterns — sequential (each agent passes output to the next) and concurrent (agents work independent sub-goals, merged by an orchestrator) — are the same primitives an operations manager already knows.²² Deloitte describes the enterprise versions as supervisor-agent models, adaptive networks and hybrids, and predicts firms will move along an “autonomy spectrum” — human in-the-loop, on-the-loop, out-of-the-loop — beginning the shift toward human-on-the-loop orchestration in 2026.²¹

The human role this creates is the through-line of the whole vendor literature. Microsoft’s agent boss — “someone who builds, delegates to and manages agents” — is the named version, and it explicitly applies “at any organizational level,” with new job categories already cited: AI Trainer, Data Specialist, Security Specialist, Agent Specialist, ROI Analyst.⁹¹⁰ Huang’s “IT as the HR of AI agents” is the same role located at the infrastructure layer — recruiting, onboarding and supervising a digital workforce.³⁴ And Microsoft’s Agent 365 is that HR function turned into a product: a “control plane for AI agents” with a registry providing a “comprehensive inventory of all agents,” unique agent IDs operating under “principle of least privilege,” role-based dashboards, and “unified observability across your entire agent fleet.”¹¹ The vocabulary — fleet, registry, least-privilege, observability — is borrowed equally from HR and from IT, which is precisely Huang’s point.

The scale that makes orchestration a first-order problem rather than a curiosity comes from the market projections: Deloitte sizes the autonomous-agent market at $8.5B in 2026 rising to $35B by 2030 (or $45B “with improved orchestration”), and Microsoft cites an IDC projection of 1.3 billion agents by 2028.²¹¹¹ A firm managing thousands of agents will need the same things a firm managing thousands of people needs — identity, access control, performance monitoring, and a manager — which is why “agent sprawl” and the registry that contains it are now explicit product categories.¹¹

The reality check: hype, cancellation risk, and “agent washing”

The corpus would be dishonest if it presented the labour framing without the gap between rhetoric and deployment, which is wide. On adoption, the surveys are consistent that production use is the exception: BCG finds only 13% of organisations have deployed agents integrated into workflows, with 56% experimenting and 31% having done nothing; only 33% of workers even claim a clear understanding of what an agent is.²⁰ McKinsey, on a different base, finds 88% of organisations use AI somewhere but only 23% are scaling agents in at least one function, and “nearly two-thirds … have not yet begun scaling AI across the enterprise.”¹⁹

Gartner supplies the sharpest counterpoint: “More than 40% of agentic AI projects will be canceled by the end of 2027,” on a poll of 3,400+ organisations, because “most agentic AI projects right now are early-stage experiments or proof of concepts that are mostly driven by hype.”⁸ The firm coined “agent washing” for the rebranding of chatbots and RPA as agents, and estimates that “of the thousands of vendors claiming agentic solutions … only around 130 offer real agentic features.”⁸ Deloitte independently echoes the >40% cancellation figure.²¹ The labour framing, in other words, is running well ahead of the deployable technology.

The macroeconomic sceptic completes the picture. Acemoglu’s task-based model predicts AI delivers “no more than a 0.71% increase in total factor productivity over 10 years,” revised “below 0.55%” once hard-to-learn tasks are counted — a direct deflation of multi-trillion-dollar transformation narratives — and warns the capital–labour income gap may widen even so.¹⁴ Set against the vendor framing, this is the strongest reminder that “digital labour” is a thesis about the future, not a measured fact about the present.

What the evidence actually supports

Reconciled honestly, the corpus supports a narrower and more interesting claim than the marketing. The reframing is real and near-universal across vendors and investors: agentic AI is being built, priced and governed as labour, with a coherent economic logic — software as capital that produces work, sized against the wage bill rather than the software budget.¹⁵¹² The organisational implication is also real and is the part most likely to outlast the hype cycle: once managing workers gets cheaper, span of control and layer count become design variables, and a management role for agents — orchestrator, agent boss, “IT as HR” — emerges by necessity rather than fashion.^15161022

But the substitution claim — that AI is replacing human labour at scale now — is not what the evidence shows. Production deployment is a minority,²⁰¹⁹ a plurality of projects is expected to fail,⁸ the employment effect so far is concentrated in entry-level hiring rather than broad displacement,¹³¹⁸ and the single cleanest behavioural signal points the other way: on Claude.ai, augmentation has just overtaken automation, 52% to 45%, reversing the August-2025 position — having moved 41% → 42% → 49% → 45% on the automation side over the year.¹⁷ For now, the modal interaction is a person working with an agent, not an agent working instead of a person.

References