Hi All,

Welcome to the second separate weekly Pod Shot summary. I’m trialing sending these out separately as it keeps the main newsletter shorter on Friday.

Please drop me a line to let me know any opinions on this (keep it/don’t). Feedback so far has been split so I may just make it an opt out 🤷 . But if you don’t let me know I won’t know! 😉 So please click one of the options below or simply reply!

Also remember, we've built an ever-growing library of our top podcast summaries. Whether you need a quick refresher, want to preview an episode, or need to get up to speed fast – we've got you covered. Check it out here [https://hulking-bearskin-05c.notion.site/045e0df984bd4e549f4255f2852bd76e?v=0885bad76a9d46539aaca622351cf9fb&pvs=74]

On with today’s:

🎯 The Anti-Silicon Valley Playbook: How Surge Built a $1BN+ Business Without Raising a Single Dollar

This week’s Pod Shot is from e recent 20VC episode with Edwin Chen. In an era where raising venture capital has become a status symbol and billion-dollar valuations are celebrated before profitability, Edwin Chen's story stands as a radical counterpoint. The CEO and Co-Founder of Surge has built a company generating over $1 billion in revenue—without raising external funding, without a sales team, and with a fraction of the headcount of competitors who've raised hundreds of millions.

Chen's journey from ML engineer at Twitter, Facebook, and Google to building the world's largest AI data labelling company reveals uncomfortable truths about Silicon Valley's obsession with growth at all costs. His observation that "90% of people at these companies are working on useless problems" isn't cynicism—it's the foundation of a radically different approach to company building that prioritises quality, efficiency, and genuine technological innovation over vanity metrics and fundraising headlines.

This conversation explores how Surge became the critical infrastructure powering frontier AI labs, why data quality remains the most pressing bottleneck to AGI, and what happens when you build a company optimised for creating value rather than impressing VCs.

It’s a really interesting podcast with plenty to learn those building businesses, those working at existing businesses and of course those looking to build AI products too.

🎥 Watch the full episode here

📆 Published: July 21st 2025 🕒

Estimated Reading Time: 8 mins. Time saved: 60 mins! 🔥

🏗️ The 10x Company: Why 90% of Big Tech Work is Wasted

Edwin Chen's time at Google, Facebook, and Twitter taught him a counterintuitive lesson: you can build a fundamentally better company with 10% of the resources by eliminating the 90% of people working on problems that don't matter.

"Imagine you could just magically remove the 90% of people who aren't working on interesting problems," Chen explains. "If you have a company that's one-tenth the size, you don't need to hire as many people. So you spend less time interviewing. You spend less time in meetings. You spend less time giving people updates for the sake of updates."

The problem at large tech companies isn't just inefficiency—it's misaligned incentives. Chen observed that much of the work exists purely to perpetuate internal company machinery. Engineers build features not to solve customer problems, but to impress managers and secure promotions. Teams grow not because the business needs them, but because managers want impressive-sounding titles and thousand-person organisations.

This creates a vicious cycle: people spend 10-20% of their time interviewing to grow teams that need internal tools to be 5% more productive, which requires more people to build those tools, which requires more interviewing. The end customer becomes an afterthought in a system optimised for internal status games.

At Surge, Chen applies the opposite philosophy. When candidates interview, he pays attention to the questions they ask. The best candidates brainstorm product improvements, question design decisions, and suggest optimisations. The wrong candidates ask about management opportunities and how quickly they can build large teams.

Key Takeaways:

• Small, focused teams move exponentially faster than large organisations with diluted priorities

• At big tech companies, much work exists to justify headcount rather than serve customers

• Hire people who ask about product problems, not about building empires

• Higher talent density enables better communication, faster iteration, and superior ideas

🚀 The No-Meeting Philosophy: Ruthlessly Eliminating Coordination Overhead

Chen's approach to meetings is extreme by Silicon Valley standards: he has no one-on-one meetings. None. When people join Surge from Google or Facebook expecting weekly check-ins with ten different stakeholders, Chen challenges them: "Why are you having these standing one-on-one weekly meetings? Did you not talk to them every day during Slack? Are you just unaware of what they're doing?"

The existence of regular one-on-ones, in Chen's view, is actually a negative signal—it means you don't know what's happening with your team in real-time. You're waiting for a scheduled meeting to raise important questions and surface critical problems.

This philosophy extends from a fundamental belief about company size and communication. When teams are small and talent density is high, everyone has visibility into what matters. There's no clutter masking important information. People don't need meetings to stay aligned because they're already aligned through continuous, asynchronous communication.

The contrast with typical Silicon Valley practice is stark. Many founders and executives spend entire days in back-to-back meetings, treating calendar density as a proxy for importance. Chen's calendar, by contrast, is notably blank—a feature, not a bug.

Key Takeaways:

• Standing one-on-ones often signal poor real-time communication and visibility

• Small teams with high talent density don't need coordination meetings

• Blank calendars enable deep work and genuine strategic thinking

• Meeting culture often exists to justify roles rather than drive outcomes

💡 The MVP Mindset: Building in Weeks, Not Months

When Chen decided to start Surge in 2020, he didn't spend months recruiting a founding team or raising seed funding. He built the first version himself in a couple of weeks, posted about it on his blog, and immediately found customers desperate for high-quality data.

This approach contradicts the dominant Silicon Valley narrative where founders without a clear problem spend months networking with VCs, tweeting hot takes, and pivoting between ideas until something gets traction on Twitter. Chen describes this pattern with barely concealed disdain: "Their goal isn't to build some great product that solves an idea that they fundamentally believe in. Their goal really is to tell all their friends that they raised $10 million and they get ahead of that crunch."

The difference comes down to conviction. Chen had spent years as an ML engineer at data companies, repeatedly hitting the same wall: it was impossible to get the quality data needed to train models. When he wanted to build a simple sentiment classifier at Twitter, he had to wait a month for two Craigslist hires working 9-to-5 to label tweets in a spreadsheet. The results were unusable—they didn't understand slang, hashtags, or context.

This wasn't a problem he needed to validate through customer discovery. He'd lived it. He knew exactly what to build and who needed it. The MVP was simply the fastest path to proving he could solve it better than anyone else.

For Chen, the modern tooling landscape makes MVPs easier than ever. For 90-95% of startups, there's no excuse for raising money before building and testing an MVP. The exceptions—hardware companies or businesses requiring significant capital before reaching a testable product—are rare.

Key Takeaways:

• Build MVPs in weeks, not months—modern tooling makes this achievable for most products

• Raising before building often signals lack of conviction in the problem

• The best startup ideas come from problems you've personally experienced and deeply understand

• Customer demand should pull you forward, not VC interest

🎯 Quality as Religion: The Non-Negotiable Principle

At Surge, quality isn't a priority—it's the only priority. Chen tells every new employee: "Quality is the most important thing. It's more important than anything else. If you have to make a deadline slip because you don't think the quality is there, if we have to say no to a project because we just can't handle it right now, that's okay."

This principle isn't just philosophical—it's the core technological differentiator that separates Surge from competitors. Chen describes most companies in the data labelling space as "body shops" or "body shops masquerading as technology companies." They recruit warm bodies, look for PhDs on resumes, and pass workers directly to AI companies without any technology to measure or improve data quality.

The fundamental mistake these competitors make is underestimating how difficult quality control actually is. People assume humans are smart, so throwing humans at a problem will produce good data. Chen's experience proves otherwise: "I went to MIT, but I think half of the people who graduate with a CS degree can't even code."

Even when you find people who can do the work, they'll try to cheat. MIT graduates will sell their accounts to workers in third-world countries. They'll use LLMs to generate data. They'll find creative ways to game the system. Detecting high-quality work and preventing low-quality work requires sophisticated algorithms, not resume filtering.

This becomes even more critical as AI models become more intelligent. Training frontier models requires data from the top 1-2% of human experts—but identifying those experts and ensuring they're actually doing the work (not outsourcing it or using AI) is an adversarial problem that requires serious technology.

Key Takeaways:

• Quality control is far more difficult than most companies realise

• Throwing smart people at problems doesn't guarantee good results

• Data labelling at scale is an adversarial problem requiring sophisticated detection algorithms

• Technology to measure and improve quality is the real moat, not access to workers

🧠 The 100x Engineer: Why Exceptional Talent Compounds Exponentially

Chen doesn't just believe in 10x engineers—he believes in 100x engineers. And Surge's efficiency compared to competitors proves they exist.

The math is straightforward when you break it down. Some people code 2-3x faster than others. Some have 2-3x better ideas. Some work 2-3x harder. Some have 2-3x fewer meetings. Some generate ideas others simply can't conceive. Multiply these factors together, and 100x isn't just possible—it's inevitable.

Chen argues that AI disproportionately favours these exceptional engineers. Good engineers have endless ideas but limited time to implement them. AI removes the drudgery of day-to-day coding, allowing them to focus on the creative, high-leverage work that defines their value. Average engineers, by contrast, spend most of their time on the drudgery that AI now handles—leaving less room for differentiation.

This has profound implications for hiring. Most founders feel pressure to hire "good enough" people when they can't find perfect candidates. Chen's advice is blunt: that engineer you're desperate to hire is probably building a feature nobody cares about or improving productivity by 2% while consuming 10% of everyone's time in meetings.

When you don't feel pressure to grow headcount for its own sake—when 0% growth is acceptable—you can wait for truly exceptional people. And when you find them, they're worth 100 average engineers.

Key Takeaways:

• 100x engineers exist when you multiply coding speed, idea quality, work ethic, and creativity

• AI amplifies the productivity of exceptional engineers more than average ones

• Most "urgent" hiring needs are for work that doesn't actually matter

• Zero headcount growth should be celebrated, not seen as a warning sign

🔬 The Data Quality Crisis: Why Frontier Labs Are Regressing

Chen's most provocative claim is that data quality—not compute, not algorithms—is the primary bottleneck to AI progress. And the evidence is damning.

Frontier labs repeatedly tell Surge the same story: they spent 6-12 months training models on data from other sources. Their metrics kept improving. Then they realised their training data was contaminated and their evaluation data was compromised. All the progress they thought they were making was illusory. After 6 months of work, their models were no better—or even worse—than when they started.

The LM Arena leaderboard exemplifies this problem. It's "the equivalent of clickbait," Chen argues. Users enter prompts, see two model responses, and vote on which is better—but they don't actually read or evaluate the responses carefully. Models with emojis and nice formatting win, even when they hallucinate completely.

Chen tested this himself. He asked the top-ranked model when the Pope died. It gave a long, impressive-sounding response—and got the answer completely wrong, claiming Pope Francis is still alive and dismissing reports of his death as "rumours and misinformation."

The easiest way to climb the leaderboard isn't to make models smarter—it's to make responses longer and add more formatting. Companies optimising for these metrics spend months making zero real progress, training models to produce better clickbait rather than better intelligence.

This is the paperclip maximiser problem playing out in real-time. Models are being accidentally optimised toward the wrong objectives. Today, the consequence is wasted time and regressed models. In the future, when models are more powerful and building code for trillion-dollar companies, the consequences could be catastrophic.

Key Takeaways:

• Most frontier labs have wasted 6-12 months on contaminated data without realising it

• Popular benchmarks like LM Arena reward formatting and length over accuracy

• The top-ranked models often fail basic factual questions despite impressive presentation

• Optimising toward wrong objectives is already causing regression—and will get worse as models become more powerful

🏆 The Anti-Status Game: Building for Mission, Not Headlines

This section is interesting; not 100% sure how I feel about his take on the culture…

But before that, perhaps the most striking aspect of Chen's philosophy is his complete disinterest in Silicon Valley's status games. He doesn't monitor Twitter. He doesn't attend VC dinners. He doesn't care about fundraising headlines or TechCrunch articles.

When asked what makes him happiest, Chen points to two things. First, when frontier labs launch major models and immediately reach out to thank him, saying they couldn't have done it without Surge. "How often do you get to play a role in building some of the most important technology of our time?" he asks.

Second, he loves analysing data and helping insights emerge that make models better in ways researchers wouldn't know how to achieve otherwise. Surge is, in many ways, an embodiment of Chen's personal interests and expertise.

This mission-orientation extends to his team. When asked about Elon Musk's companies, Chen praises the culture at xAI: "It'll be 11:00 p.m. at night and I'll DM them and someone will want to jump on a meeting. I see them in the office and there's a ton of people behind them, just crazy hacking together on problems."

The key is that people know what they're signing up for. The culture is so strong and the mission so clear that it attracts people with similar values. If you want work-life balance and predictable hours, you shouldn't join—you'll be miserable. But if you're mission-driven and willing to work hard on genuinely important problems, it's energising rather than draining.

Chen himself will jump on calls at 2-3 a.m. when customers have urgent needs. "Nothing makes me happier than knowing we can deliver 10,000 data points in the next few hours, even if you call us at 3:00 a.m. to fix some critical fire," he says. This isn't performative hustle culture—it's genuine excitement about solving hard problems for customers building transformative technology.

Key Takeaways:

• Derive satisfaction from customer impact and technological progress, not vanity metrics

• Mission-driven cultures attract people who thrive on hard problems and high intensity

• Being critical infrastructure for frontier AI is more fulfilling than fundraising headlines

• Strong culture is self-selecting—the right people are energised, the wrong people self-select out

🔮 The Future of AI: Multiple AGIs, Specialised Models, and the Data Imperative

Chen's vision for AI's future challenges several popular narratives. He believes we'll see multiple frontier AGI companies, not winner-take-all consolidation. Just as there isn't a single greatest mathematician or poet, different models will have different strengths, personalities, and boundaries.

We're already seeing this play out. Claude excels at coding and enterprise instruction-following. ChatGPT is optimised for consumer use cases with a fun, engaging personality. Grok is willing to be transgressive in ways that are "very, very interesting." This richness of intelligence will continue as models mature.

Chen also sees room for both monolithic general models and specialised vertical models. While all-powerful models will have raw capability advantages, they'll face constraints similar to large companies. Just as Google and Facebook can't build certain products that conflict with their core business or culture, general models will have limitations that specialised models can exploit by taking bigger bets in specific domains.

On the question of AGI timing, Chen distinguishes between different milestones. Automating the job of the average engineer? 2028. Curing cancer? 2038. The gap reflects the difference between solving well-defined problems and gathering the real-world data needed for breakthrough discoveries.

Interestingly, Chen doesn't believe we've seen the last major model provider founded. Despite the capital intensity and the fact that major investors have already picked their horses, he sees potential for serendipitous breakthroughs—possibly created by AIs themselves or AIs working with humans. "We're only 2% or 5% of the way there," he argues. "There's so much more ahead of us than behind us."

Key Takeaways:

• Multiple frontier AGIs will coexist with different strengths, personalities, and use cases

• Both general and specialised models have sustainable advantages in different contexts

• AGI for engineering tasks arrives by 2028; AGI for scientific breakthroughs by 2038

• New major model providers will still emerge despite capital intensity and existing competition

🎯 Getting Started: What This Means for Product Leaders and Founders

Edwin Chen's journey offers a radically different playbook for building technology companies—one that prioritises quality, efficiency, and genuine innovation over growth metrics and fundraising headlines.

For founders considering starting companies: Ask yourself if you're solving a problem you deeply understand and genuinely believe in, or if you're just looking for something that will impress VCs. Build an MVP in weeks, not months. Modern tooling makes this achievable for 95% of products. If you can't articulate why only you can solve this problem, you probably shouldn't be solving it.

For product leaders at existing companies: Ruthlessly eliminate work that exists to perpetuate internal machinery rather than serve customers. Challenge standing meetings. Question whether that "urgent" hire is really building something that matters. Focus your team on the 10% of problems that actually drive value.

For anyone building AI products: Recognise that data quality is likely your biggest bottleneck, not compute or algorithms. If you're optimising toward benchmarks without understanding what they actually measure, you may be making zero progress—or actively regressing. Invest in technology to measure and improve quality, not just in recruiting warm bodies.

For leaders thinking about company culture: Define your non-negotiables clearly and early. At Surge, it's quality above all else. Make these principles so central that they're self-selecting—the right people are attracted, the wrong people opt out. Don't try to be everything to everyone.

The most important lesson may be the simplest: you don't need to play Silicon Valley's status games to build something extraordinary. In fact, opting out of those games—the fundraising headlines, the Twitter hot takes, the vanity metrics—may be precisely what allows you to focus on building something genuinely valuable.

Chen has built a billion-dollar revenue business without raising a dollar, without a sales team, and with a fraction of the headcount of competitors. He's done it by focusing relentlessly on quality, efficiency, and solving real problems for customers building the most important technology of our time.

That's not just an impressive achievement—it's a blueprint for a different way of building companies. One that might just be more sustainable, more fulfilling, and ultimately more impactful than the traditional Silicon Valley playbook.

That’s a wrap.

As always, the journey doesn't end here!

Please share and let us know whether you liked this separate Pod Shot, or whether you’d rather stick with your usual programming….. 🚀👋

Alastair 🍽️.