YC's Garry Tan Advocates for Workflow Redesign in the AI Era

In 2026, as generative AI penetrates deeper into practical applications, Silicon Valley’s entrepreneurial ecosystem finds itself at a crossroads of fundamental restructuring. Garry Tan, the head of Y Combinator (YC), argues that the real competitive edge in the AI era does not stem from technical distinctions like model weighting or fine-tuning, but rather from redesigning workflows. In a lecture at YC’s startup camp, Tan presented compelling data to illustrate how the traditional business logic of “scaling up manpower to generate value” is on the verge of collapsing.

The Reality of a 400-Fold Efficiency Leap

Garry Tan began his lecture with a personal anecdote. In 2013, during his tenure as a YC partner, he could only write about 14 lines of useful logic code per day on average, excluding comments and blank lines. Fast forward to 2026, and his productivity has surged approximately 400-fold, despite spending nearly the same amount of time. Even under the assumption that half of the code written by AI Agents is scaffolding frameworks and the data is self-promotional, a minimum 8-fold increase and a median 80-fold improvement have been achieved.

What Tan emphasizes is that this disparity is not rooted in the model itself. “The difference between someone achieving 2x efficiency and another achieving 100x efficiency lies in the fact that they are using the same Claude model, the same weights, the same context window, and the same API,” Tan explains. In other words, the foundation of leverage does not lie in model weights but in the design of workflow serialization (the sequential chaining of processes).

YC’s internal data corroborates this claim. In the Winter 2025 batch, 95% of the codebase for a quarter of the startups was generated by AI. This batch turned out to be the fastest-growing and most profitable in YC’s history, with 94 companies achieving $100 million in revenue within a single seed round. Tan notes, “While we cannot prove that AI-generated code directly caused this growth, the fastest-growing founders we funded treated AI not merely as a code-completion tool but as a productivity workforce.”

Coding Organizational Structure:

A New Set of Laws

The crux of Tan’s lecture lies in mapping insights gained from building AI Agents onto real-world organizational structures. According to him, in the AI era, “one skill manual is equivalent to one employee.” A skill manual encapsulates a single capability, handles specific tasks, and is clearly documented, enabling any entity to execute it. A parser table serves as the organization chart, deciding who handles incoming tasks. Archival rules correspond to internal processes, and trigger evaluations align with performance reviews.

“All the functional modules of an organization that once required a thousand employees are now reduced to a few code files and descriptive documents,” Tan remarks. This transformation doesn’t merely enhance efficiency; it rewrites the fundamental laws of organizational physics.

One concrete example is the AI application-building tool project “Emergence” from YC’s Summer 2024 batch. The company reached nine-figure ARR (Annual Recurring Revenue) within just eight months of its official launch. When ARR surpassed $15 million, the team consisted of only 15 members. By Winter 2024, retail revenue had reached $60 million, with the team growing to around 40 people. “This extraordinarily high per capita productivity has never existed in the software, oil, or railroad industries,” Tan evaluates.

Expanding AI Utilization Beyond Engineers

Tan is particularly intrigued by how this transformation extends beyond engineering. During YC’s digital transformation, media personnel, event staff, and finance teams also participated. Employees who had never opened a terminal in their lives are now constructing their own skill manuals and periodic tasks.

Tan shares the example of a YC finance staff member who used the company’s open-source system and enterprise brain to consolidate around 100 complex Excel workbooks into a single application. Although not a programmer, this individual has since become a manager overseeing AI Agents. According to Tan, every YC employee occupies a similar role.

“This isn’t about engineers becoming 400 times more efficient; it’s about creating a company with 400 times higher operational efficiency,” Tan concludes. In AI-native companies, the sole purpose of hiring engineers is to solve tasks that AI Agents cannot yet handle and to maintain those skills.

The Practical Significance of Workflow Redesign

Tan’s argument seeks to rectify misconceptions about AI adoption. He warns, “If you treat AI as merely a code-completion tool and see prompts as one-off utilities, you’ll be outcompeted in the next phase of the race.”

The core message he described as “worth engraving on the inside of your eyelids” is this: “Leverage does not come from model weights but from how workflows are serialized.” Instead of chasing performance differences in AI models, redesigning the collaborative processes between humans and AI is the key to true competitive advantage.

Editorial Opinion

Garry Tan’s lecture is commendable for clearly demonstrating the superiority of “workflow design” in AI adoption. While much focus tends to be placed on technical elements like model selection and prompt engineering, the overarching concept of redesigning organizational workflows to be AI-native stands as the sustainable source of competitive advantage—a perspective that offers valuable insights for Japan’s DX strategies.

In the short term, businesses need to shift away from simply using AI for “code completion” or “partial task automation” and move towards the idea of coding organizational structures themselves. In the medium to long term, Tan’s model of “achieving nine-figure ARR with a 15-person team” has the potential to fundamentally challenge the advantages of large-scale organizations. Traditional business models reliant on economies of scale may struggle to adapt to an era where small AI-native organizations generate high added value. Particularly, the value of “meticulously crafted services” that Japanese companies excel in may require redefinition.

That said, it remains uncertain whether the model Tan proposes can be universally applied to all industries and business models.

References

• “YC掌门人最新分享：AI时代，永远不要做一次性工作”, by 划重点KeyPoints — 钛媒体, 2026-07-20T10:42:40.000Z (ARR)
• Source URL: https://www.tmtpost.com/8071908.html