AI Agents Now Run Companies

Sequoia Capital's recent agent deployment chart presents a startling, often underestimated, reality: while software engineering approaches 50% agent deployment, most other sectors remain mired in single-digit adoption. This stark visual underscores a profound, uneven transformation underway, delineating industries aggressively leveraging autonomous AI from those lagging significantly. The data highlights a critical, often overlooked, paradigm shift in how companies approach operational efficiency and innovation.

Howie Liu, co-founder and CEO of Airtable, argues this 50% figure dramatically understates the true operational reality within frontier tech teams. He asserts that the most advanced organizations have moved far beyond what mainstream data captures, having crossed a critical threshold in the last four to five months. Here, frontier agents function as true autonomous coworkers, not just chat assistants, fundamentally changing how companies get built and scaled. This shift signifies a move from mere automation to genuinely intelligent, goal-oriented execution.

The crucial distinction lies far beyond a single developer using a Copilot for code suggestions or simple task assistance. Instead, leading teams now deploy and manage dozens of autonomous agents in parallel. These sophisticated systems, like Claude Code instances, execute complex, multi-hour tasks autonomously, shipping clean pull requests, performing market research, or generating comprehensive reports. This represents a fundamental shift from tool-based assistance to outcome-driven digital labor, where AI agents deliver tangible results without constant human supervision, effectively acting as "digital labor" and moving from "software as a tool" to "software as a service-provider."

This rapid, concentrated advancement by a select group creates a massive adoption gap between early movers and the broader market. While many industries struggle to implement three-year-old AI capabilities, frontier companies are already building and scaling businesses with entire fleets of always-on agents, often aiming for $100 billion valuations with under five human employees. This disparity represents a colossal, trillion-dollar agent opportunity, which Howie Liu frames as addressing the entire white-collar GDP across the Western hemisphere, poised to redefine economic value and corporate structures by enabling unprecedented productivity and scale.

The era of AI Copilot, where algorithms assist human operators, rapidly gives way to Autopilot systems. Copilot models augment human capabilities, offering suggestions for code completion, drafting text, or summarizing data. Autopilot, however, signifies fully autonomous AI, capable of perceiving environments, making decisions, and executing end-to-end tasks without constant human supervision. This fundamental shift moves beyond mere augmentation, redefining the very nature of digital work and unlocking unprecedented efficiency.

Howie Liu, co-founder and CEO of Airtable, pinpointed the AI transition's inflection point just four to five months ago. Frontier models, notably Claude 3 Opus, crossed a critical threshold, enabling agents to operate as truly autonomous coworkers, not mere chat assistants. This marked a profound change in their ability to perform complex, multi-step workflows independently, delivering complete outputs rather than just intermediate steps.

This autonomy dramatically expands the Total Addressable Market (TAM) for AI, moving far beyond initial estimates. While early AI discussions centered primarily on software, Howie Liu contends the true TAM is the entire white-collar GDP across the Western hemisphere. Sequoia Capital previously flagged a "trillion-dollar agent opportunity," but Liu's perspective suggests a significantly larger economic re-evaluation, encompassing all knowledge work.

The 'Autopilot' era was definitively unlocked when agents demonstrated the ability to ship clean, production-ready code autonomously. These sophisticated systems now tackle multi-hour tasks, delivering complete solutions from initial problem definition to deployment-ready outputs, often as "clean PRs." This capability transforms company building, enabling lean teams to achieve outputs previously requiring vast human workforces. The promise of $100 billion companies with under five employees, powered by fleets of always-on agents mapped to human job roles, now appears increasingly tangible.

New economics of talent fundamentally alter how businesses operate. AI agents introduce a token-based work model, replacing traditional hourly wages or salaries with micro-transactions. This paradigm shift means hiring "digital labor" for specific tasks costs pennies, not thousands, executing tasks that would demand days from a human analyst. For instance, a complex workflow on HyperAgent, like generating a hyperlocal real estate market report, costs around $35 in tokens.

Businesses now reframe expenditures from mere costs to strategic investments. Consider the immense opportunity cost of human executive time. What might take a team days to research and synthesize, an agent completes in hours for a fraction of the price. The value delivered far outweighs the token spend, anchoring decisions on output rather than input.

Greg Isenberg’s "Making $$ with AI Agents" video with Howie Liu, co-founder and CEO of Airtable, illustrates this clearly. Generating a comprehensive board memo or a detailed market report, a task typically consuming days of executive effort, costs approximately $150 in tokens. This isn't just cheaper; it's a profound acceleration of strategic output, freeing up valuable human capital for higher-level decision-making and innovation.

This token-based economy democratizes access to world-class strategic analysis. Solopreneurs and lean startups, once constrained by budget, can now leverage sophisticated AI agents for tasks previously reserved for large enterprises. A micro-budget gains access to capabilities like competitive analysis, market research, and even initial app builds. Howie Liu envisions a future of $100 billion companies operating with fewer than five employees, powered by fleets of always-on agents. Explore more about these capabilities at Hyperagent.

Howie Liu, co-founder and CEO of Airtable, introduces HyperAgent as his definitive solution to the complex challenge of widespread AI agent deployment. Liu positions HyperAgent as a foundational platform, designed to democratize access to powerful autonomous AI while simultaneously enabling enterprise-grade scalability. This initiative aims to bridge the gap between nascent agent capabilities and real-world business integration.

Liu famously likens HyperAgent to the "Macintosh to OpenClaw's Linux," articulating a design philosophy centered on accessibility and power. It offers a low floor, making it incredibly easy for anyone to begin deploying agents for practical tasks, yet maintains a high ceiling, providing the robust capabilities necessary for sophisticated enterprise operations and fleet management. This dual focus ensures both rapid adoption and deep utility.

HyperAgent’s core capabilities empower agents to execute a vast array of tasks autonomously. These include: - Comprehensive web browsing and information gathering - Advanced data analysis and synthesis - Dynamic content generation, spanning images, video, and even interactive applications - Seamless utilization of other tools and services via API connectors, integrating with existing tech stacks. One compelling demonstration showed an an agent generating a hyperlocal real estate market report, complete with market research, Reddit validation, competitive analysis, a V1 app build, a marketing site, and ad creative—all for just $35 in tokens.

Crucially, HyperAgent agents possess a self-learning ability, accumulating knowledge and refining their performance without requiring explicit retraining. Each session enhances an agent's understanding and skill set, making it progressively more valuable over time. This persistent learning creates an arbitrage of knowledge, where agents become domain experts through continuous operation.

This continuous improvement, combined with features like LLM-as-judge evaluations and fleet-wide observability, transforms HyperAgent into a command center for managing a scalable workforce of digital employees. It moves beyond individual task execution, enabling organizations to deploy and oversee entire fleets of autonomous agents, each growing in expertise and contributing to significant operational efficiency. HyperAgent thus provides a complete operating system for the agent economy.

Howie Liu, co-founder and CEO of Airtable, offered a live demonstration that vividly illustrated the profound shift from Copilot to Autopilot. Appearing on Greg Isenberg’s "Making $$ with AI Agents" video, Liu introduced HyperAgent, showcasing its capability to tackle a complex, real-world business challenge from inception to a deployable product. This wasn't a theoretical exercise; it was a practical display of an AI building a business.

The challenge: Generate a comprehensive, hyperlocal real estate market report. This seemingly straightforward task demands extensive research, validation, and product development. HyperAgent, operating as a fully autonomous agent, accepted the prompt and immediately began executing a multi-stage workflow, demonstrating intelligence far beyond a simple developer tool.

In a sequence of self-directed actions, HyperAgent autonomously performed: - In-depth market research to identify key trends and data points. - Validation of market insights through analysis of Reddit discussions and community sentiment. - Comprehensive competitive analysis to understand existing solutions and identify gaps. - Building a functional V1 application, demonstrating its ability to translate research into a tangible product. - Creating an entire marketing site, complete with ad creative and messaging.

The cost for this entire, end-to-end business creation workflow was a staggering $35 in tokens. This figure starkly contrasts with the tens, if not hundreds, of thousands of dollars typically required for human teams to achieve a similar outcome, underscoring the revolutionary economics of token-based work. The demonstration obliterated traditional notions of labor cost and project timelines.

This live show-and-tell fundamentally reframed the agent's role. HyperAgent wasn't merely assisting a human developer; it *was* the founder. It acted as the strategist, autonomously identifying market needs and competitive landscapes. It functioned as the executor, building both the product and its marketing infrastructure. The agent seamlessly integrated these traditionally disparate roles into a single, cohesive, and incredibly efficient operation.

Liu’s demonstration solidified the argument that frontier agents have crossed a critical threshold. They are no longer just tools for optimizing existing workflows; they are autonomous entities capable of conceiving, developing, and launching entirely new ventures. This shift demands a re-evaluation of business structures, entrepreneurial pathways, and the very definition of a "company."

Skills represent the foundational primitive for frontier agents, transforming general AI models into specialized domain experts. These sophisticated capabilities allow an agent to move beyond basic instruction, embodying a specific knowledge base or operational methodology. This crucial shift empowers agents to tackle complex tasks with expert precision and domain-specific nuance.

Howie Liu vividly demonstrated this power live on HyperAgent, meticulously constructing a custom "Greg Isenberg contrarian AI" skill in real time. The process involved HyperAgent autonomously researching Isenberg's distinctive voice, analytical framework, and typical rhetorical patterns across his public discourse. It then distilled these elements into a reusable, executable profile, capable of generating new content in his signature style.

This newly minted skill allowed the agent to draft contrarian tweets and develop unique perspectives, showcasing its learned style and consistent application. This live demonstration underscored how easily a general-purpose AI can assimilate and replicate a highly specific, nuanced human persona, turning subjective style into an objective, replicable asset.

A Skill functions as an adaptable, reusable playbook, enabling an agent to consistently replicate a specific voice, style, or analytical approach. This critical abstraction ensures uniformity and quality across an agent's outputs, moving beyond one-off prompts to systematic, predictable execution. It establishes a repeatable, high-fidelity operational standard for any task.

Businesses can leverage these skills to codify their unique brand voice, proprietary processes, or specialized analytical frameworks into scalable, digital assets. This mechanism allows companies to embed their institutional knowledge directly into their AI workforce, ensuring consistent brand representation and operational excellence at scale, even across a fleet of agents. For companies looking to integrate AI agents into their existing infrastructure or develop custom solutions, Airtable offers advanced capabilities. Explore further at Airtable: Build Enterprise-ready AI Workflows, Apps & Agents.

Deploying a fleet of agents equipped with defined skills provides a robust observability layer, crucial for managing and scaling an AI-driven operation. This capability means a company’s collective intelligence and operational nuances become programmatic, drastically reducing human oversight while amplifying efficiency. It enables a small human team to manage vast digital workforces, each operating with expert-level consistency.

Companies now move beyond isolated agents to managing an entire fleet of 20, 50, or even 100+ digital employees. This scale demands a new operational paradigm, enabling the vision Howie Liu articulated: $100B companies operating with fewer than five human employees, powered by always-on agent fleets. This represents a fundamental shift in how businesses are constructed and managed.

Overseeing these burgeoning digital workforces requires a central command center. Such a hub provides fleet-wide observability, offering comprehensive management capabilities crucial for seamless operations. HyperAgent delivers this essential system, enabling oversight, quality evaluation, and performance tracking across every deployed agent.

Maintaining consistent, high-quality output from a diverse agent fleet demands robust quality control. This is where rubrics and 'LLM-as-judge' mechanisms prove critical. Rubrics function as precise evaluation frameworks, where a separate LLM scores agent outputs, transforming generally intelligent models into domain experts through specific playbooks.

Rubrics allow businesses to score agent performance on dimensions vital for operational success. This includes metrics like creativity, factual accuracy, or adherence to a specific brand voice. Such granular, automated quality assessment provides the observability layer necessary to confidently run and scale a business on a fleet of agents, unlocking unprecedented efficiency and consistency.

Visionary leaders now present a provocative future: building a $100 billion company with fewer than five human employees. This isn't a distant fantasy but an emerging reality, driven by the capabilities of autonomous AI agents. The era of exponential human headcount growth to match market scale is rapidly drawing to a close.

This revolutionary organizational structure features a lean team of human strategists overseeing an expansive fleet of always-on AI agents. Humans define high-level objectives, establish rubrics, and provide strategic feedback, while the agents handle the relentless execution. This model completely redefines the very essence of a "company."

Agent roles map directly to traditional human job functions, but with unparalleled efficiency and persistence. Imagine a digital workforce comprised of: - AI Market Researchers: Continuously identifying untapped niches, analyzing real-time trends, and validating product ideas. - AI Sales Reps: Prospecting, qualifying leads, personalizing outreach, and closing deals around the clock. - AI Coders: Developing, debugging, deploying, and maintaining complex applications with incredible speed. - AI Marketers: Crafting targeted campaigns, managing social media, optimizing ad spend, and generating compelling content.

This model fundamentally transforms corporate scalability. Traditional growth demanded exponential increases in human resources, office space, and HR overhead. Now, a handful of humans can orchestrate a vast digital workforce, instantly scaling operations globally with near-zero marginal cost per additional agent. The arbitrage of persistence, as Howie Liu, co-founder and CEO of Airtable, highlights, means these agents execute daily, achieving results human teams often abandon.

Overhead plummets, shifting from salaries, benefits, and physical infrastructure to token costs and platform subscriptions. This allows for unprecedented agility and resource allocation, channeling capital directly into output rather than operational bureaucracy. The nature of a company evolves from managing people to intelligently orchestrating autonomous processes and outcomes.

Howie Liu’s HyperAgent, for example, provides the command center for this new paradigm, enabling humans to manage a fleet of 20, 50, or even 100+ agents. These frontier agents, empowered by "Skills" that turn general models into domain experts, are the building blocks of this digital workforce. This isn't just automation; it's the dawn of a new corporate operating system.

As AI agents take on increasingly complex tasks, the path to success in this new economy shifts dramatically from technical prowess. Mastery of coding, once paramount, now yields to a more fundamental human attribute: persistence. Building a five-person, $100 billion company hinges less on writing lines of code and more on the iterative process of working *with* intelligent systems.

Howie Liu, co-founder and CEO of Airtable, articulates this crucial insight as "the arbitrage of persistence." He observes that 99% of individuals abandon an endeavor after a single failed attempt. This human tendency to quit creates an enormous opportunity for those willing to push through initial setbacks and refine their approach.

Becoming a top 1% agent operator demands daily practice, not just sporadic engagement. Consistent effort over 30, 60, or even 90 days transforms a novice into an expert. This dedication allows operators to develop an intuitive understanding of agent behavior and performance, turning general models into specialized, reliable digital employees.

Winners in the agent economy will not necessarily be the most skilled programmers. Instead, they will be the ones who meticulously master prompt engineering, design effective feedback loops, and craft precise rubric design for evaluating agent outputs.

This consistent, iterative effort, far more than raw coding ability, unlocks the full potential of frontier agents. For further reading on this evolving landscape, consult analyses like AI 50: AI Agents Move Beyond Chat - Sequoia Capital. This new paradigm redefines the essential skills for entrepreneurial success.

The agent economy isn't a distant future; it's here now, demanding a fundamental shift in how you approach work. Begin by identifying workflows, not just individual tasks. Consider any repetitive process or system that currently consumes human hours, from market research to data synthesis or content generation.

Break down these larger problems into discrete, rule-based components. Can an agent autonomously gather information, analyze trends, or draft initial outputs? This requires reframing your perspective from manual execution to designing an automated process that an AI can manage end-to-end. Think of agents as digital employees ready for specific, defined roles.

Platforms like Howie Liu's HyperAgent offer a direct entry point into this new paradigm. HyperAgent provides the command center to deploy and manage a fleet of agents, turning general models into domain experts through custom Skills. It's designed for "builders" eager to create autonomous workflows.

For those ready to dive in, a limited-time bonus offers the first 1,000 builders $1,000 in credits to begin experimenting. This opportunity, mentioned by Greg Isenberg, significantly lowers the barrier to entry, allowing you to deploy agents without immediate substantial token costs. Claim this quickly, as supplies are limited.

Don't overthink your first project. Choose a small, definable task with clear success metrics. Build your initial agent, deploy it, and meticulously evaluate its output. The real power comes from embracing an iterative process of feedback and refinement, continually improving your agent's performance and expanding its capabilities.

This persistence, as highlighted earlier, unlocks the true arbitrage of the agent economy. Start today, build your first autonomous coworker, and contribute to the accelerating shift towards a world where a five-person team can build a $100 billion company.

What is the 'agent economy'?

The agent economy is a macroeconomic shift where autonomous AI systems, or 'agents', execute end-to-end business workflows, functioning as digital labor rather than just software tools.

What is HyperAgent?

HyperAgent is a platform introduced by Airtable's CEO, Howie Liu, designed for building, deploying, and managing fleets of autonomous AI agents to perform complex, real-world business tasks.

How do AI agents differ from chatbots or copilots?

While chatbots and copilots assist humans with specific tasks, AI agents are autonomous. They can perceive their environment, make decisions, and execute complex, multi-step workflows without constant human supervision.

What are 'frontier agents'?

Frontier agents refer to the latest generation of highly capable AI agents that have crossed a performance threshold, allowing them to function as true autonomous coworkers on complex, multi-hour tasks.