A Chinese version of this essay is available at the end of the post.

Seven Years in Startups

It’s early March.

Over the past couple of months, I’ve often found myself returning to a familiar mental state in my dreams — a mode where it feels like I’m trying to finish as much work as possible before the end of time.

For the past 20 months, I was at CurveBio, building a computational platform from scratch, assembling a team, and serving as Head of Computational Science / Senior Director of Machine Learning. It was intense, but also one of the fastest growth periods of my career.

I genuinely enjoyed my time there.

But long stretches of overwork eventually started triggering warning signals from my body. Headaches became more frequent. Each one felt like a reminder to slow down.

During my exit interview, Nathan — Curve’s co-founder and CSO — said something that stuck with me: I should reflect on whether the intensity comes from the environment, or from my own habits.

So I’ve been trying to write some of those reflections down.

This is the first of two essays. This one is about working in innovation-driven startups. The next one will be about a career decision I arrived at after a lot of thinking over the past six months.

Startups Are Survival Environments

When I think about startups, two images often come to mind.

• A glass frog in the Amazon rainforest.
• A newborn zebra on the African savanna.

In both environments, survival is immediate and unavoidable.

Working at an early-stage startup often feels like that.

Most startups operate with limited runway:

• 18 months of runway is considered well financed
• With 12 months left, fundraising usually begins
• With 6 months left, companies often race toward milestones while preparing contingency plans

This pressure shapes almost every decision a company makes — business model, technical direction, team structure, hiring, and execution speed.

To attract investors, startups often pursue areas where few people have succeeded before. These are places with enormous upside if they work — but also huge uncertainty, and a long trail of forgotten attempts.

So progress has to be fast. Fail fast becomes a necessity.

When plans change — and they always do — bursts of intense work become normal. In today’s hypercompetitive environment, especially with the explosion of AI startups, it’s not unusual to see job descriptions openly mention expectations like “996” or beyond.

Of course, financial pressure isn’t unique to startups. Big companies more and more frequently lay people off. Academic labs compete for grants.

But startups have something different: a much shorter window to survive, and a much higher bar for differentiation.

They have to become something unique before they run out of time.

Founders often intentionally project strong optimism — investors look for that kind of signals. Experienced teams, however, usually approach opportunities with more caution, carefully examining the assumptions behind them.

In my view: Science determines whether the destination of success even exists. Execution determines whether we can reach it.

Interestingly, many founders later admit that their successful business model looked nothing like their original idea.

The pattern is familiar: commit first, learn quickly, adapt along the way.

It’s a mix of courage, risk tolerance, and sometimes sheer stubbornness.

Why Startups Are Incredible Learning Environments

For employees, startups can be extraordinary growth environments.

You see how companies actually work. You participate in real strategic decisions. You watch business models evolve. You observe competition firsthand. And your work often directly shapes the company’s future. You also build deep bonds with colleagues who go through intense periods together.

If you happen to join the 1% of startups that succeed, there may also be rare financial rewards and major career acceleration. Those opportunities are far less common in large organizations or academia.

Another interesting feature of startups is focus.

Early on, companies may explore many possibilities. But once founders choose a direction, the organization often aligns tightly around it.

Paradoxically, this can mean less internal politics than large companies. With fewer layers and fewer competing agendas, things often move faster.

Startups also tend to hire generalists — or people who can become generalists. Especially before Series B or C, everyone ends up wearing multiple hats.

Over time, you become what people jokingly call a “hexagon warrior” — someone capable across many areas.

You grow alongside the company. Often very quickly.

The Tension Between Theory and Speed

One thing I’ve gradually realized about myself is that I have a strong theory-first instinct.

Before building something, I often want to understand the underlying abstraction — the mathematical or conceptual structure of the problem.

Once that abstraction is clear, it becomes easier to build tools that solve not only the current problem, but future ones as well. In other words, I have a bias toward general solutions.

This probably comes from my early fascination with physics. Physics tries to look past surface phenomena and explain the world with elegant, compact equations.

This mindset has helped me find creative solutions to difficult problems.

But it also has a cost.

Developing deeper abstractions takes time — often far more time than people realize. And when outcomes are uncertain, that effort can create pressure, both internally and externally.

In startup environments, this sometimes clashes with the dominant philosophy:

move fast, fail fast, iterate quickly.

When I started leading teams, I began seeing the problem from another perspective.

In resource-constrained organizations, leadership needs visible signals of progress. Even rough prototypes can help guide decisions: whether to invest more resources, pivot, or stop.

This is why MVP-style development is often favored.

If everyone spends months building perfect systems or theoretical frameworks, decision-makers may have no clear signal about progress.

Balancing “start running quickly” with “build the right foundation” became one of the hardest leadership questions I faced.

There is no universal answer.

But I learned something simple that helps: pause briefly before acting on instinct, and think about the system as a whole.

The Hidden Complexity of Small Teams

Startup teams are small by necessity. That creates both opportunity and risk.On the positive side, small teams allow individuals to grow rapidly. Everyone takes ownership of meaningful work.

But small teams also mean high individual load. Many team members — myself included — experienced periods of sustained overwork. Even with careful planning, the tension between limited resources and ambitious goals never fully disappears.

Fortunately, in the domain, modern AI tools have dramatically accelerated many coding and implementation tasks, which helps reduce some of that burden.

Another challenge is the alignment between individual trajectories and company needs.

Even when hiring people who strongly believe in the mission, individuals still have their own career goals, life circumstances, and financial priorities.

Over time, those trajectories don’t always align perfectly with the company’s stage.

As a manager, I tried to create as much flexibility and autonomy as possible.

But turnover is natural in any organization. And in small teams, the departure of a single person can have outsized impact.

This taught me the importance of:

• encouraging collaboration early
• sharing knowledge across the team
• maintaining traceability in projects

Even while pushing forward at startup speed.

Looking Back

Seven years in startups has been both fascinating and difficult.

If I could rewind time, I might make some different choices.

But I also have a quiet sense that someday I might start something again — perhaps a company, perhaps a research institute, or something closer to a childhood dream: building a science museum in my hometown.

These years did not bring financial independence.

But the growth, perspective, and experience they brought will stay with me for a long time.

---

三月春光。2026年马年的拜年还没来得及说完，元宵节却已经快到了。

过去的一两个月里，我常常在梦里回到一种熟悉的状态——像是在离世之前拼命想把所有事情 做完一样的工作节奏。过去20个月，我在 CurveBio 作为计算带头人，从零到一搭建平台、 组建团队。那是一段强度很高，但也让我成长极快的经历。

老实说，我其实很喜欢自己在 Curve 的这段工作经历。但长时间的超负荷工作也让身体越 来越频繁地发出警报。每一次头痛都像是在提醒我应该停一停。正如 Nathan 在 exit interview 时对我说的，我或许需要认真想一想，这究竟是环境带来的，还是我自己的习 惯。

最近离开之后，我一直想把这些经历和一些思考整理下来。如果不出意外，大概会写两篇。 这一篇主要谈这些年在创新型初创公司的感受；另一篇则会写我过去半年思考后做出的一个 职业生涯决定。

Curve 其实并不是我第一次在创新型初创公司工作。在此之前，我还经历过 Freenome，以 及研究生时期合作的 NanoMedica。对比我在 GSK.AI 的经历，以及身边许多在大型公司或 学术机构工作的朋友，我越来越清楚地感受到，初创企业确实有一些非常独特的特征。

初创企业：生存是第一问题

如果要用一个画面来形容初创企业的生存环境，我常常会想到两个场景：

亚马逊雨林里的 glass frog，和非洲草原上刚出生的 小斑马。对于很多早期员工来说，在 创新型初创公司工作往往就像置身这样的环境——生存是一个你无法回避的问题。

大多数初创企业的 runway 都非常有限：

• 一年半的 runway 已经算是 well financed
• 剩一年时往往就要开始准备下一轮融资
• 如果只剩半年，公司就不得不一边冲刺完成 milestone，一边准备 contingency plan

这种生存压力会深刻影响公司的几乎每一个决策：商业模式、业务方向、团队结构、股权安 排、项目执行方式。

为了获得投资人的支持，公司往往不得不选择那些鲜有人涉足、也鲜有人成功的方向。那里 既可能通向巨大的机会，也充满摸着石头过河的不确定，以及无数被遗忘的先驱。

为了避免资金耗尽，很多事情必须快速推进，fail fast。当计划赶不上变化时，加班加点 的突击几乎成为常态。在这个竞争激烈、AI 进展井喷的时代，硅谷的一些初创公司甚至在 招聘启事里直接写明需要适应“996”，甚至更多。

当然，资金压力并不是初创企业独有的问题。企业如此，学术界的 PI 也是如此。即使曾经 被认为无限优渥的互联网大厂员工，在如今的裁员潮中，也不得不面对类似“鱿鱼游戏”的现 实。

但我的感受是，创新型初创企业往往拥有更短的生存窗口期，同时又必须足够与众不同才能 活下来。

创始人和团队需要持续关注需求和竞争者。当找到一个鲜有人尝试的方向时，尽管充满未 知，但也往往蕴含着令人兴奋的机会。

创始人通常会表现得比团队更加乐观和确定，因为他们的信心本身就是投资人首先关注的信 号。但一个有经验的团队往往会更谨慎地审视每一个机会背后的假设。

在我看来，对于创新型初创企业而言：

科学上的正确往往决定成功这一目的地是否存在，而执行力则决定我们能否抵达那里。

我也曾听一些创业者讲述，他们后来真正成功的商业模式与最初的想法相去甚远。他们往往 带着一种孤注一掷的决心，先置身其中，再不断试错。这种模式既令人感叹，也似乎与他们 所处的人生阶段、风险偏好以及承受能力有关。当然，其中也少不了创业者的勇气和企业家 的精神。

在初创公司工作的成长

对于员工来说，这样的环境带来的全面成长几乎是可以预见的。

你有机会参与公司的重大决策，了解行业竞争，探索商业模式，寻找新的机会。你的工作可 以真实地影响公司的未来，你也会收获一群一起日夜奋斗的同事和朋友。

如果碰上那 1% 的幸运，公司发展顺利，也许还会带来难得的财务回报¹ 以及职业平台的跃迁。这些机会，在大型公司和学术界往往并不常见。

虽然在商业模式探索期大家往往对各种可能保持开放态度，但在具体决策和执行层面，创新 型企业通常会非常强调专注。当创始团队认定某一方向为公司的定位之后，公司上下往往会 高度聚焦。这种专注反而减少了很多大型组织中常见的资源争夺和沟通成本，这也是不少从 大公司出来的人会怀念的一点。

在这种极简结构下，初创企业（尤其是 B 轮或 C 轮之前）通常更倾向寻找多面手，或者有 潜力成长为多面手的人。于是，每个员工往往身兼数职，在生存压力下逐渐成为“六边形战 士”。

最终，你会和公司一起成长，而且往往成长得更快。

但与此同时，一个问题也很自然地出现：在有限的时间里，是选择更聚焦，还是选择更全 面？

理论优先 vs Fail Fast

我常常意识到自己多少带有一些完美主义或理想主义的倾向。

更准确地说，在工作中我通常是一个 理论优先的决策者和执行者。我习惯先花时间思考问 题的本质是什么，背后的数学或物理抽象是什么，然后再去构建解决当前问题以及未来可能 出现问题的工具。我似乎对“通用解”有一种天然的偏好。这或许与我从小喜欢物理学有关—— 物理学总是试图透过现象看本质，用优雅而简练的公式解释万物。

在实践中，这种思维方式确实曾帮助我找到一些独到且最终有效的解决方案。但与此同时， 它也意味着投入大量常人难以想象、也不为人所见的时间和精力。当结果尚不确定时，也不 可避免地伴随着来自他人和自我的怀疑与压力。这种习惯与初创环境里强调的 fail fast 有时并不完全契合。面面俱到的倾向，也让我在很多时候陷入超负荷工作的状态。

在开始带领团队之后，我逐渐学会从另一个角度思考。

在资源有限、多人协作的环境中，领导层往往需要看到一个可以感知进展的原型，哪怕它很 粗糙、很局限。领导者需要依赖这些信号来判断：是继续加大资源，还是调整方向。

因此，MVP（Minimum Viable Product）式的工作方式往往更容易被接受，也让项目进展更 可控。

如果每一个执行者都把时间投入到抽象理论或工具工程上，决策者反而很难判断项目距离里 程碑还有多远。

如何在 “先跑起来” 与 “先选对路、铺好路” 之间取得平衡，成为我在 Curve 同时作为核 心执行者和技术团队负责人时经常思考的问题。

每一个具体情境都会有不同的答案。但我逐渐学到的一点是：在惯性行动之前，先暂停一 下，做一次全局性的思考。

小团队的另一面初创企业的团队构建也是一个很有意思的问题。

由于资源有限，团队通常规模不大，但成员往往都很精干，同时也是多面手。这既带来了成 长机会，也带来了一些管理上的挑战。

首先是工作负荷。包括我自己在内，团队成员都经历过非常高强度的工作节奏。尽管我和我 的经理一直在尝试优化短期和长期的工作计划，以缓解人手不足与企业生存之间的矛盾，但 压力依然真实存在。

幸运的是，在这个AI工具迅速发展的时代，许多 coding 和 implementation 的工作被显著 加速，一定程度上缓解了负担。

第二个挑战是个人发展与公司需求之间的张力。即使在招聘时我们努力寻找志同道合的人， 每个成员仍然有自己的职业规划、生活方式和经济回报的考虑。这些需求并不总能与公司某 个阶段的状态完全契合。久而久之，就不可避免地会出现机会成本与沉没成本之间的权衡。

作为管理者，我曾尽可能创造更灵活的工作方式和更大的自主空间。但不可否认，人员流动 依然会自然发生。而在初创团队中，每个人往往都独当一面，因此成员离开带来的影响也更 明显。这也让我逐渐意识到，从一开始就鼓励协作、共享知识，并在推进项目进度的同时保 持工作的可追溯性，是非常重要的。

回头看这七年

七年多的初创经历很有趣，也很不容易。

如果时光倒流，也许我会做出一些不同的选择。但我始终隐约觉得，在未来的某一天，我也 许仍然会再次走向创业——也许是公司，也许是研究所，又或者像儿时的梦想那样，在家乡建 一座科学馆。

这些年的经历没有带来财富自由，但它带来的成长与视野，大概会一直伴随着我。