AI Creates New Jobs and Shifts Value to Infrastructure

Brendan Foody of Mercor, in a revealing conversation with Harry Stebbings on The Twenty Minute VC, dismantles conventional wisdom surrounding AI's impact on the job market and the defensibility of software companies. The core thesis is that while AI will undoubtedly displace many current roles, it will simultaneously birth entirely new, massive job categories, particularly in the realm of agent training and management. This conversation uncovers the hidden consequence that the speed of AI-driven transition is far more disruptive than previous technological revolutions, demanding a radical rethinking of economic structures and corporate strategy. Anyone involved in building, investing in, or leading technology companies, especially those in the AI space, will gain a crucial advantage by understanding these non-obvious dynamics and preparing for the accelerated shifts in labor and value creation.

The Unseen Labor Market: Training Agents as the Next Economic Engine

The narrative surrounding AI often focuses on job displacement, a valid concern, but one that overlooks the generative power of technological advancement. Brendan Foody argues that the true economic story of AI will be the creation of entirely new job categories, with "training agents" emerging as a dominant force. This isn't just about automating existing tasks; it's about humans teaching AI to perform complex, nuanced work. The consequence of this shift is a fundamental redefinition of labor, where the ability to effectively curate, guide, and refine AI agents becomes a primary driver of economic value.

"One of the largest things that people underestimate, both in the context of AI labs as well as within the enterprise, is how significant of a job category it is going to be to train agents."

This highlights a critical downstream effect: as AI becomes more capable, the bottleneck shifts from raw model performance to the human expertise required to harness that performance. Companies that master this "training agents" paradigm will not only solve immediate operational challenges but will also build a durable competitive advantage. The conventional wisdom that AI leads to mass unemployment fails to account for this new layer of human-AI collaboration. This requires a long-term view, where investing in the infrastructure and talent to train these agents today pays off significantly in the future, creating a moat that is difficult for competitors to replicate. The immediate discomfort of re-skilling or building new teams for agent training yields a substantial, delayed payoff.

The Erosion of Application Layer Defensibility: Why Infrastructure Wins

Foody's assertion that the "model is the product" and the subsequent difficulty in building defensibility at the application layer is a stark warning to many startups. The argument is that foundation model providers, like OpenAI and Anthropic, can increasingly replicate the functionalities of downstream applications by improving their own models. This creates a dynamic where software layers built on top of these models are inherently less defensible because the underlying intelligence can be directly accessed and improved by the model providers themselves.

"building defensibility in the software layer on top of the models is going to be incredibly difficult."

The consequence of this is that value will increasingly accrue to the infrastructure layer--the providers of compute, data, and the foundational models themselves. Companies operating here can build moats through network effects and data advantages. For instance, Mercor's own business model, focused on data aggregation and expert networks, benefits from these compounding effects. This insight challenges the common startup strategy of building a niche application; instead, it suggests a focus on foundational elements or deeply entrenched services with strong network effects. The conventional wisdom of "build an app, then scale" is challenged by the reality that the "product" itself (the model) is evolving so rapidly that application layers can be quickly outmoded or replicated. The delayed payoff here comes from building a business that is resilient to the rapid evolution of AI capabilities, rather than one that is dependent on a specific software abstraction.

The Shifting Economics of AI: Compute Costs Outpacing Headcount

Perhaps one of the most counter-intuitive insights is Foody's revelation that Mercor is spending more on tokens for its internal agents than on employee headcount. This points to a profound shift in the economics of AI-driven businesses. While efficiency gains from AI are often framed as cost savings on human labor, the reality is that the consumption of AI services--compute and tokens--is escalating rapidly, creating a new, significant cost center.

"Right now, we're spending more on tokens for our internal agents than we are on employee headcount."

This has several downstream implications. Firstly, it means that companies must develop sophisticated strategies for managing and optimizing their compute spend, treating it with the same rigor as human capital. This leads to the need for "evals" to benchmark model performance and cost-effectiveness for specific workflows, effectively commoditizing the model layer through perfect competition. Secondly, it suggests that the long-term value will accrue to those who can provide cost-effective compute or highly optimized models. Companies that can manage this escalating cost structure effectively will gain a significant competitive advantage. The immediate discomfort of managing this new, substantial operational cost yields a long-term benefit: the ability to leverage AI at scale more efficiently than competitors who are slower to adapt their financial models.

The Primacy of the Forward-Deployed Motion: Beyond Go-to-Market

Foody argues that for AI-centric businesses, the "forward-deployed motion"--the post-sales effort of deeply integrating with and optimizing for a customer--is more critical for defensibility than traditional go-to-market strategies. This contrasts with the conventional view that sales and marketing prowess often dictates success. The reasoning is that as AI models become more capable and software layers become easier to replicate, the true differentiator lies in the bespoke services and deep customer understanding that enable AI to deliver maximum value.

"the forward-deployed motion being the post-sales, go-to-market being the pre-sales. Because ultimately, say you're just really good at sales, and then you provide a SaaS product, and you have a savvy customer who's spending a million dollars a year on the SaaS product, and they realize they could just telco to copy it, and they'll get the same exact thing. It feels very difficult to maintain your pricing power..."

The consequence of this is a strategic imperative for companies to invest heavily in customer success, implementation, and ongoing support that goes beyond basic service. This deep integration, which involves training agents on customer-specific tacit knowledge, creates a sticky relationship that is difficult for competitors to dislodge. The delayed payoff for this approach is a robust, defensible customer base that is less susceptible to commoditization. The immediate effort required to build these deep, service-oriented relationships, often perceived as less glamorous than sales, creates a lasting competitive moat.

Key Action Items

• Develop Agent Training Expertise: Invest in teams and processes specifically designed to train and manage AI agents. This is a foundational skill for future work. (Immediate to Ongoing)
• Focus on Infrastructure or Deeply Networked Services: Prioritize building businesses at the AI infrastructure layer or in application areas with strong, compounding network effects, rather than purely software-based applications. (Strategic Shift - Next 6-12 months)
• Optimize Compute Spend: Implement rigorous evaluation frameworks for AI models and providers to manage escalating token and compute costs effectively. Treat compute spend as a strategic financial imperative. (Immediate to Ongoing)
• Prioritize Forward-Deployed Motion: Shift focus from pre-sales to post-sales customer integration, training, and bespoke AI deployment to build defensible customer relationships. (Strategic Shift - Next 6-12 months)
• Embrace New Job Categories: Actively identify and recruit talent for emerging roles like AI trainers, agent managers, and AI ethicists, rather than solely focusing on traditional roles. (Hiring Strategy - Ongoing)
• Invest in Data Moats: Continue to build and leverage proprietary data sets and expert networks that enhance model performance and create unique value propositions. (Long-term Investment - Ongoing)
• Prepare for Economic Re-skilling: Begin planning for the societal impact of AI-driven job displacement and creation by exploring educational and training initiatives for new AI-centric roles. (Strategic Planning - Next 12-18 months)