Garry Tan: The traditional business rule that 'revenue growth must be accompanied by large-scale hiring' is failing

Y Combinator President and CEO Garry Tan stated that AI startups Emergent and Retell are achieving rapid recurring revenue growth with very few employees, indicating that the traditional business rule of 'revenue growth must be accompanied by large-scale hiring' is failing.

Reportedly, Emergent achieved a nine-figure annualized revenue in about 8 months with a team of around 15 people; based on a lower limit of $100 million annualized revenue, its annualized revenue per employee is at least about $6.67 million. This figure has not been corroborated by Emergent's official financial reports or audit documents and should be viewed as operational data cited by Garry Tan rather than verified revenue disclosures.

Retell is described as having an annualized recurring revenue of about $60 million with a team of around 40 people, corresponding to an annualized revenue per employee of about $1.5 million. Public information shows that Retell previously disclosed about $40 million in annualized revenue and a team of about 25 people, with subsequent market estimates placing its annualized revenue at about $60 million; differences in figures can arise from varying time points, statistical criteria, and employee scopes.

Retell's product positioning is as a real-time AI voice agent infrastructure for enterprises, which customers can use for call centers, sales, customer service, and phone automation. Its core business logic is to replace some high-frequency, standardized phone tasks traditionally handled by agents with models, speech recognition, speech synthesis, workflow orchestration, and real-time call infrastructure.

Garry Tan proposed the organizational view that 'one Markdown file equals one employee': companies write repeatable processes, knowledge bases, task specifications, and failure correction rules into callable documents, which are then executed by AI agents. After each manual correction of agent errors, the process can be rewritten as new rules, allowing organizational knowledge to expand without increasing management layers, training cycles, and repetitive human resources.

In market mechanisms, buyers are enterprise clients looking to reduce costs in customer service, sales, R&D, operations, and outsourcing; sellers are AI companies providing model invocation, voice agents, code generation, and automation orchestration capabilities. Funding will shift from traditional BPO, low-end SaaS implementations, repetitive operational roles, and substantial middle management coordination costs to GPU cloud resources, foundational model suppliers, agent platforms, data governance, and high-value sales deployment teams; beneficiaries will be AI vendors that master customer workflows and low marginal cost delivery capabilities, while service companies relying on 'headcount expansion' to maintain revenue will face pressure.

Source: Public information

The growth model of traditional SaaS companies typically requires simultaneous expansion of revenue, customer success, sales, support, engineering, and management teams. Salesforce initially expanded enterprise software revenue through a direct sales system, while ServiceNow relied on sales, implementation partners, and customer success systems to promote large enterprise deployments; high revenue per employee is not uncommon, but maintaining a team of dozens at the scale of tens of millions in ARR is extremely rare. Garry Tan emphasizes that it is not that employees suddenly become more efficient, but rather that some tasks previously performed by junior engineers, QA, operations coordinators, or outsourced agents are absorbed by models, prompts, workflows, and automated evaluation cycles. In terms of capital pathways, Emergent and Retell have rewritten the ratio of financing to labor expenditure as 'capital prioritizing the purchase of machine capabilities rather than prioritizing the purchase of organizational scale.' If the former reaches nine-figure annualized revenue, its main resource consumption will no longer necessarily be hundreds of software engineers, but may focus on inference computing power, model APIs, agent orchestration, automated testing, and customer deployment; Retell encapsulates real-time voice interaction capabilities as a platform, transforming individual customer needs for agents, phone systems, and process engineers into continuous subscriptions and usage-based billing. After revenue growth, costs will rise with model invocation and cloud infrastructure, but typically will not rise in proportion to traditional service industry headcounts. Historically, this is similar to how Amazon Web Services transformed enterprise server procurement into on-demand cloud resources, and Twilio transformed communication infrastructure into APIs: customers no longer need to build complete teams for each capability but obtain standardized services based on usage. The difference is that AI agents are beginning to cover previously unstandardizable knowledge work processes. Retell enters from real-time voice conversations, facing not just the pure communication API market, but the labor costs of call centers; if Emergent enters with agent development or automation to build products, it faces the engineering hours required for software delivery. They are in the expansion phase of AI's transition from 'auxiliary tools' to 'measurable delivery units.' This represents a form of technological substitution, but the substitute is not a specific position, but rather the internal coordination, handover, and repetitive decision-making costs within the organization. In the past, companies increased employees to expand output, and with more employees, they needed management layers, knowledge transfer, meetings, and process control; AI agents crystallize processes into reusable instructions, tool calls, and evaluation data, allowing new output to rely more on computing power and software replication. As a result, revenue per employee may surge, but the true moat lies not in low headcount, but in whether there is exclusive customer data, continuously improving agent performance, and the deployment capability to bear responsibility for results in high-risk workflows. ABAB News · Cognitive Laws

Headcount determines cost, process determines leverage
Automation replaces positions, organizational memory replaces management
Revenue growth relies on replication, profit growth relies on not expanding headcount.

ABAB AI Insight