a16z
Kavak’s AI chief openly says he is investing in tokens instead of knowledge workers, that agents outperform humans, and that an AI can even run a city. Then comes the escape hatch: six-week Jedi training, perpetual upskilling, brave founders, and the promise that creative destruction will be better for everyone—the employment circuit is terminated, but apparently with a startup pitch deck and a lightsaber.
The interview describes Kavak’s transformation from a vertically integrated used-car marketplace into an AI-native company organized around long-running agents. Ale Massa says each customer can receive a dedicated agent with persistent memory, access to Kavak’s APIs, and a long-term objective of maximizing customer lifetime value. He reports that agents handle approximately 96% of customer interactions and 95% of transactions, with between 100,000 and 200,000 agents instantiated daily. Massa explains that Kavak rebuilt its APIs, feedback loops, evaluation systems, and organizational structure around these agents. He claims improvements including higher conversion, tripled NPS, faster loan approvals, better vehicle inspections, and a 1.5x increase in profits in a trial where an AI agent managed operations in Cuernavaca. Physical workers, including mechanics, remain involved but use AI sidekicks. The company has also created a six-week Jedi Academy to train employees across roles to build and work with agents. The final section presents AI-native startups as the likely source of creative destruction, comparing the shift to electricity, factory redesign, and computing. Massa argues that incumbent firms will struggle because they cannot fully rebuild around AI, while new founders can design organizations from scratch and use increasingly capable models to create new products and services.
Relative to the Discontinuity Thesis, this video is unusually explicit about the mechanism of labour substitution. It treats agents as superior to human workers across sales, finance, customer interaction, management, and eventually executive work. The claims that agents handle 96% of interactions, that tokens are replacing knowledge workers, and that an AI-managed city can outperform prior operations all acknowledge a structural break at the firm level rather than merely describing temporary disruption. The video also demonstrates the cost and scale advantages that make substitution economically compelling. Its argument breaks down when it moves from firm-level displacement to society-wide consequences. It never seriously addresses aggregate wage loss, ownership of the productivity gains, or how consumption continues when fewer people receive labour income. Instead, it offers retraining, human-agent collaboration, entrepreneurial opportunity, and creative destruction as an assumed route to a better future. The speaker recognises that many jobs will be performed better by AI, but treats the resulting exclusion as a skills and organisational-design problem. Under DT logic, this is a strong diagnosis followed by heavy solution cope: the termination is visible, but no credible successor income or social system is supplied.
This is not crude denial; it is the polished version that is more useful to capital. The speaker hands over nearly the entire employment function to agents, admits they outperform humans, and then tells the remaining workforce to attend the Jedi Academy, upgrade their skills every few months, become useful to the machines, or leave. That is reskilling fantasy wearing a lab coat. Once the agent is better than the best human, training everyone to build agents does not restore the jobs those agents eliminated. The beneficiaries are the owners of the company, the infrastructure, the data, and the token economy: they get lower labour costs, infinite patience, persistent surveillance of customer behaviour, and scalable extraction of lifetime value. The video never asks who gets purchasing power after wages disappear. Its electricity analogy, creative-destruction story, and founder-celebration simply smuggle in the assumption that new firms and better products will make everyone richer. That is not an answer to structural exclusion; it is a venture-capital fairy tale about why the people being automated should applaud the efficiency gains.