Thời đại “nhóm nhỏ” của Silicon Valley đã đến, khi các startup dùng AI để tối ưu hóa hiệu suất
Giai đoạn “blitzscaling” (2012–2023) từng đề cao định giá doanh nghiệp và số vốn huy động, nhưng giờ đây, chỉ số “doanh thu trên mỗi nhân viên” mới là thước đo danh giá.
Sam Altman từng chia sẻ một nhóm CEO đang cá cược xem khi nào xuất hiện công ty tỷ đô do một người điều hành – điều này giờ đã khả thi nhờ AI.
Khái niệm “botscaling” mô tả các startup tăng trưởng mạnh mà không cần mở rộng nhân sự nhờ sử dụng AI trong lập trình, marketing, nghiên cứu khách hàng, và thậm chí là lên ý tưởng kinh doanh.
Jeffrey Bussgang và Clarence Wooten đã viết sách hướng dẫn xây dựng startup chỉ với một người và AI, không cần đồng sáng lập con người.
Nhiều founder như Roo Harrigan (Portrait), Myles Lazerow (Addie), và Chris Mannion (Meander) đang dùng AI để xây dựng sản phẩm, thương hiệu, chiến lược bán hàng và chăm sóc khách hàng.
Bussgang thử nghiệm việc dùng ChatGPT như đồng sáng lập ảo trong lớp MBA tại Harvard. Kết quả nghiên cứu cho thấy một cá nhân dùng AI có thể sáng tạo tốt ngang một nhóm người.
Các AI được gọi bằng nhiều tên: “cộng sự kỹ thuật số”, “bạn đồng hành”, “cố vấn”, hay “bóng ma năng suất” – nhưng vẫn bị giới hạn bởi tính không đáng tin và xu hướng nịnh nọt.
AI dần trở thành công cụ “vô hạn thực tập sinh” nhưng nếu dùng đúng cách, có thể là "đồng sáng lập" đầy tiềm năng. Ví dụ, startup như Pillar Square Health do sinh viên y Harvard sáng lập đã dùng AI hoàn toàn thay cho đồng sáng lập con người.
Một số founder “đào tạo” AI bằng cách đưa vào thế giới quan, tài liệu học yêu thích để AI hoạt động giống như chính họ – chuẩn bị cho tương lai công ty không cần người điều hành.
📌 AI đang thay đổi cốt lõi khởi nghiệp: Thay vì mở rộng đội ngũ, các startup tập trung tối ưu hóa hiệu suất qua AI. Doanh thu hàng triệu USD chỉ với vài người – thậm chí một người – là xu hướng mới. Botscaling trở thành chiến lược phổ biến, khi AI đóng vai trò từ lập trình viên đến nhà đồng sáng lập.
In the era of startup “blitzscaling,” which lasted roughly from Facebook’s IPO in 2012 until WeWork’s bankruptcy filing in 2023, market capitalization and total capital raised were prized metrics. The ultimate milestone was reaching “unicorn status” — a $1 billion valuation that was often accompanied by rapid hiring.
These days, bragging rights are going to entrepreneurs who keep headcount the lowest.
If blitzscaling was powered by smartphones and the cloud, today’s faster, leaner growth comes courtesy of AI assistants, advisers, coders and marketers. Last year, OpenAI’s Sam Altman said his tech-CEO group chat had a “betting pool for the first year there is a one-person billion-dollar company.” The idea “would’ve been unimaginable without AI,” Altman said. “And now [it] will happen.”
Jeffrey Bussgang, a partner at Flybridge Capital Partners and a lecturer at Harvard Business School, says that the new holy-grail metric isn’t a billion-dollar valuation but revenue per employee. In talks, Bussgang cites a website called the “Tiny Teams Hall of Fame” that lists companies with just a handful of employees claiming tens or even hundreds of million in annual revenue. He refers to it as “scaling without growing.” Goodbye, blitzscaling; hello, botscaling.
But if AI can effectively fill nearly any role at a company, what is the human in charge really bringing to the table?
The Botscaling Era
The mantra of Bussgang’s recent book, The Experimentation Machine: Finding Product-Market Fit in the Age of AI (Damn Gravity Media, March 2025) is essentially: Do what you do best, and let AI do the rest. Bussgang wants to fund startups that use artificial intelligence for coding, for customer research, for marketing, for pitching — even for coming up with their initial idea. He also writes about entrepreneurs who declined to partner with a human co-founder, opting to rely on AI instead.
Bussgang isn’t the only one providing a playbook for AI-native startups. In “Cofounder.AI,” a website and accompanying book (BookBaby/Self-published, March 2025), Clarence Wooten, an entrepreneur in residence at Google’s [X] moonshot factory, urges founders to consider “building a billion-dollar startup… by yourself.” Me, My Customer, and AI (Lasega Books, August 2025), by entrepreneurs Henrik Werdelin and Nicholas Thorne, argues that AI will democratize entrepreneurship — albeit with slightly more focus on small businesses than on tech startups.
These are not think-y big-idea books, and they engage only sparingly with the wider societal impact of creating hyper-productive, low-employment companies. All three are how-to guides aimed squarely at current or aspiring entrepreneurs. And together they provide a useful window into what work looks like when you go all in on AI, a philosophy around which companies are already being built.
Take software. Ten years ago, the first task for a tech entrepreneur who couldn’t code was to find someone “technical” to build a prototype. But large language models (LLMs) are particularly good at coding; even their mistakes are less of a problem for a prototype that isn’t intended to be perfect.
“All of a sudden we could write a 10-paragraph essay on what we wanted to build then plug it into these demo tools and say: ‘Go, show me what you’ve got,’” says Roo Harrigan, co-founder and head of engineering at Portrait, a network for creative professionals. “Which was very strange as an engineer.”
It’s not just coding. Portrait also used AI to research the subscription market, and it maintains an ever-lengthening AI prompt describing its ideal customer, which the founders use to solicit advice and discuss strategy. Myles Lazerow, CEO of matchmaking app Addie, said he relies on AI for branding and marketing. He shared his pitch deck as well as competitors’ with OpenAI, and asked for ideas to improve it. Chris Mannion, who founded a workforce-planning startup called Meander, uses AI to tailor customer outreach — it suggests which product features a potential user will find most valuable, based on their public profile and background.
As I talked to entrepreneurs about how they use AI, I came to think that the approach in these books mostly made sense: New ventures are almost by definition short on resources, and need to try out lots of different approaches to see what works. Instead of hiring a marketing assistant only to realize a few months later that what you really need is a head of sales, an LLM can write marketing copy one day and outline sales strategy the next.
The AI Co-Founder
Entrepreneurs have long balanced out their skills by finding a co-founder who complements them. Visionary Steve Jobs had hardware guru Steve Wozniak; designers Brian Chesky and Joe Gebbia co-founded Airbnb with their developer roommate Nathan Blecharczyk. But as AI improves, it can provide a growing number of complementary skillsets — even the business concept itself.
In his MBA class at Harvard, Bussgang has students prompt ChatGPT “to act as a co-founder” and develop startup ideas. And in a recent study, researchers at Harvard and the University of Pennsylvania assigned some employees at Procter & Gamble to come up with new product ideas using AI while others did so without it.
“The first big ‘aha’ was that an individual with AI was as good as a team without AI,” said Karim Lakhani, a professor at Harvard Business School and one of the study’s authors. Teams using AI did better still, as judged by a panel of professionals. The researchers also found that AI helped the employees come up with ideas outside their areas of expertise.
“I would never consider bringing on a co-founder at this point,” said Haan Razak, a medical and MBA student at Harvard, and a former student of Bussgang’s, who is launching a health-care startup called Pillar Square Health. “Why would I give away 50% of the company so early on? I can get to a [minimum-viable product] on a seed round without a technical co-founder.” He also noted that strife between co-founders frequently brings down startups: “I’m not willing to let my company be contingent on [that relationship].”
Mannion is also staying solo. “I took [Bussgang’s] advice and built a co-founder in ChatGPT,” he said. “There’s certainly no replacement [for a human co-founder], but [AI] has given me an opportunity to move forward with less doubt and less analysis paralysis.”
Of the three books, Wooten’s leans furthest into the idea of AI co-founders — it’s right there in the title. But even he acknowledges it’s “not a total replacement.” “Whether you opt for a human co-founder, an AI assistant, or some hybrid approach, the key is to build a support system that amplifies your strengths and shores up your weaknesses,” he writes.
That tracked with the entrepreneurs I spoke to about pushing the limits of AI management. The bots were cheap, always on, and willing to take a crack at anything. The trick was knowing when to trust them, and how to put them to use.
The Jagged Frontier
Throughout my reading and conversations, I kept asking: Is AI a tool or a teammate? Did it require constant human direction, or was it closer to a substitute for a person?
The answer seemed to be both and neither. Lakhani’s paper dubbed it a “cybernetic teammate.” Razak bucketed his use into “tool,” “teacher” and “thought partner.” Wooten describes it as a “copilot” and as “a digital species.” Lazerow said he used it as a “mentor.” Harrigan called it a “sidekick.” But my favorite descriptor came from Jean Ellen Cowgill, Portrait’s CEO, who referred to the LLMs as “ghosts around you who are helping you do work.”
One thing nearly everyone mentioned is that AI models can’t always be trusted. “They lie confidently,” said Max Parsons, Lazerow’s co-founder and CTO at Addie. Even more common, they can lapse into sycophancy. “It kind of has a tendency to want to agree with you,” Razak said. In April, OpenAI acknowledged that a recent model of ChatGPT had this issue and rolled back an update to fix it.
In a separate study, Lakhani and colleagues write that “the advantages of AI, while substantial, are… unclear to users. It performs well at some jobs, and fails in other circumstances in ways that are difficult to predict in advance.” They dubbed this inscrutable set of aptitudes “the jagged frontier.”
One answer, then, for what’s left for a human founder to do as AIs take on more and more work: They obsessively figure out what AI is and isn’t good at, and delegate appropriately.
On a recent podcast, economist Tyler Cowen asked Anthropic co-founder (and former Bloomberg News reporter) Jack Clark whether AI’s coding ability meant “the age of the nerds” was over.
“I think it’s actually going to be the era of the manager nerds now,” Clark replied. “I think being able to manage fleets of AI agents and orchestrate them is going to make people incredibly powerful.”
When Mannion wanted help deciding between three versions of his business, for example, he headed off any sycophancy by prompting his AI assistant to predict the chances of success for each path — a tactic borrowed from a research paper that found AI assistants improved forecasters’ accuracy.
As AI improves, the frontier should become smoother. The technology may well improve across the board, and different models might excel at different things. In coding, for instance, Harrigan told me she now uses “two full-time assistants, Claude and ChatGPT. And I have ChatGPT check Claude’s work all the time.”
Am I Special?
If AI can fill almost any role, and if it can increasingly act as a check on another AI’s work, what’s left for the human outside of delegation and bot management? In Me, My Customer, and AI, Werdelin and Thorne suggest the founder’s job is to know their customer better than anyone else. And all three of the botscaling bibles argue that human judgment remains essential.
But as I tried out some of the prompts in Bussgang’s book, I had doubts about the importance of my own judgment. First, I used ChatGPT to come up with startup ideas. Next, I created two “CustomGPTs” — persistent AI personas — to act as my co-founder and a prospective customer. To help me simulate a customer interview, the former came up with questions I could ask the latter.