2025 September Digest

The most interesting interview I watched recently was with Edwin Chen, CEO and Co-Founder of Surge — a company that has achieved impressive revenue without taking external funding. In the conversation, Edwin shares his philosophy on company building and his perspective on the future of artificial intelligence.

I first crossed paths with Edwin early in my career at Twitter. At the time, he was a prolific blogger writing about data science, and I remember thinking Twitter must be an incredible place if it had people of his caliber. We’ve had a few conversations over the years, and I’m genuinely excited to see how well he’s doing today.

This interview gave me a lot to think about — from the right reasons to start a company to how someone can tackle big problems by leveraging their own expertise and background. It was refreshing to watch someone I know, even peripherally, share their insights so thoughtfully.

The Unfunded Path to Scale: Edwin Chen's Surge Story

Surge's primary differentiator and core principle is an unwavering commitment to data quality, which Chen identifies as the most significant bottleneck to AI progress today. He sharply contrasts Surge's technology-driven approach with competitors, which he dismisses as "body shops" that lack the sophisticated algorithms necessary to measure quality, identify top-tier human contributors, and prevent adversarial behavior. This focus on quality has allowed Surge to become an indispensable partner to leading AI labs, who rely on its data to train and evaluate frontier models.

Chen is also a vocal critic of Silicon Valley's fundraising culture, which he describes as a "status game" divorced from genuine product-building. He advocates for founders to pursue ideas they fundamentally believe in, build a Minimum Viable Product (MVP) themselves, and remain profitable from the outset to maintain complete control. This ethos is reflected in his personal motivation; he is driven not by financial milestones but by the mission to help achieve Artificial General Intelligence (AGI). He states he would not sell Surge for even $100 billion, viewing an acquisition as an "admission of failure" when the company is already successful and he is doing exactly what he wants to do.

The Founding and Growth of Surge

Surge's creation and bootstrapped trajectory are a direct embodiment of Edwin's principles. The company was born from a specific, deeply felt problem and scaled through product-market fit rather than venture capital.

Origin Story: While working on ads and search systems at Twitter, Chen experienced firsthand the impossibility of acquiring high-quality data. A simple request for 10,000 labeled tweets for a sentiment classifier took months to fulfill with a two-person team hired from Craigslist, and the resulting data was "completely junk." He ended up labeling the tweets himself in a week, highlighting a critical bottleneck in ML development.

MVP and Launch (2020): A strong believer in MVPs, Chen built the first version of Surge himself in a couple of weeks after the launch of GPT-3. Leveraging his deep domain expertise, he posted about it on his blog and immediately found significant demand.

Bootstrapped Growth: Surge was profitable from its first month. This financial independence allowed the company to grow on its own terms, without the need for a sales team or the pressure to chase logos to impress investors. Chen states, "there was nothing that raising would help us with."

Customer Selection: By not employing a sales team, Surge attracted customers who already understood the value of high-quality data and shared the company's vision. Chen emphasizes the importance of early customers shaping the product in line with its core principles.

Inflection Point: While growth was strong from day one, it hit a major inflection point with the launch of ChatGPT, as the industry recognized the critical importance of high-quality human data for Reinforcement Learning from Human Feedback (RLHF).

Core Principles & Competitive Differentiation

Surge's market position is built on a foundation of non-negotiable principles, chief among them being an obsessive focus on data quality, which is enabled by a proprietary technology stack.

• The Prime Directive: The company's most ingrained principle is that "quality is the most important thing... it's more important than anything else." Employees are empowered to let deadlines slip or say "no" to projects if the quality bar cannot be met. Chen states, "I think we've never let quality slip."
• Adversarial Challenge: Chen notes that achieving quality is a difficult, adversarial problem. He states that even coders from MIT will "try to cheat you" by selling accounts or using LLMs to generate data. This necessitates sophisticated systems to detect both high-quality and low-quality work.
• The Competitors: Chen categorizes most other companies in the space as "body shops" or "body shops masquerading as technology companies." He defines this as entities that lack the technology to measure or improve the quality of the data they produce.
• The Surge Difference: Surge's technology is its core differentiator. The company builds complex algorithms for: Measuring the quality of data being produced. A/B testing methods to improve worker quality and efficiency. Identifying and routing tasks to the top 1-2% of specialists from a pool of millions of workers. Managing tens of thousands of unique projects running concurrently. Detecting and removing low-quality or fraudulent work.

Company Building Philosophy and Efficiency

The central theme of Edwin's story is a rejection of conventional Silicon Valley norms in favor of a first-principles approach to company building. Edwin's philosophy is rooted in extreme efficiency, a cautionary lesson learned from his time at large tech companies like Google and Twitter.

• Useless Problems: Chen asserts that at his former employers (Google, Facebook, Twitter), "90% of the people there were working on useless problems." This leads to a belief that a company can be built with 10% of the resources and people while moving 10 times faster and building a 10 times better product.
• Internal Company Machinery: He argues that much of the work in large companies is divorced from end-customer value. Instead, it serves to perpetuate the bureaucracy, such as improving internal tools for marginal productivity gains or building features solely to impress management for promotions.
• Misaligned Incentives: A primary goal for many managers in large organizations becomes growing their org size to enhance their status (e.g., "tell their friends they're a VP of a thousand person org"), leading to hiring for the sake of hiring rather than for genuine need.

Principles of Lean Operations

Here are some of Edwin's principles for running a lean operation with a high caliber team.

• Small, Focused Teams: By removing the "90% of people who aren't working on interesting problems," a company can reduce time spent on interviewing and meetings, giving everyone a clearer view of what is important.
• Hiring "Doers": Chen distinguishes between "doers" and managers who prioritize headcount. He identifies "doers" during interviews by the questions they ask; they focus on the product and brainstorm improvements, while others inquire about future management opportunities and their ability to hire a team.
• "Ruthless" Meeting Policy: Chen has a minimalist approach to meetings, personally holding no standing one-on-one meetings. He considers the need for a standing weekly one-on-one a "negative sign," as it implies a lack of daily, organic communication. The company culture is to "ruthlessly internally about killing meetings."

Views on Silicon Valley, Fundraising, and Motivation

Edwin's personal and corporate ethos stands in stark contrast to the prevailing culture in Silicon Valley, particularly regarding fundraising, risk, and the definition of success.

• A "Status Game": Edwin views the venture capital fundraising circuit as "a status game for most people." He argues that the goal for many is not to solve a problem but to "tell all their friends that they raised $10 million."
• Idea First, Money Later: He believes the first instinct for a founder should be to "find some big idea that they fundamentally believe in that could change the world" and double down on it for years. Constant pivoting based on what might attract VC funding is a sign of not taking real risks.
• MVP is Non-Negotiable: With modern tooling, Chen argues that "for 90 to 95% of startups," there is no excuse for raising money before building an MVP and testing for traction.

The Future of AI and Data

Edwin offers a clear, data-centric perspective on the trajectory of AI development, emphasizing the bottlenecks, the changing nature of data, and the future competitive landscape. When asked to rank the bottlenecks to AI progress, Chen's order is unequivocal:

• Data Quality
• Compute
• Algorithm

He argues that throwing more compute at a problem with poor-quality data is not only ineffective but actively harmful. He cites the popular LM Arena leaderboard as an example where models are rewarded for superficial traits like using emojis and writing longer responses, leading developers to believe they are making progress when they are actually "training their models to produce better clickbait." This can result in labs wasting six months to a year making "zero progress" or even regressing.

Edwin believes synthetic data is useful for "academic, homework-style" problems but is "terrible at real-world use cases." He reports that clients have found a few thousand high-quality human data points to be more valuable than 10 million synthetic ones. Humans are required as an "external value system" to catch fundamental model errors that humans would never make.

Edwin believes there will be "multiple frontier AGIs" because different companies will make different trade-offs in focus, personality, and boundaries. He cites the current differences between OpenAI, Anthropic, and Grok as evidence of this trend. He does not believe the biggest model providers have all been founded yet, suggesting that if we are only 1-5% of the way to the full potential of AGI, there is immense room for "serendipitous... creative breakthroughs."