How does a startup out-recruit Google? Palantir cofounder Stephen Cohen explains.

In a conversation at Stanford with Peter Thiel and Max Levchin, my friend Stephen Cohen (cofounder of Palantir Technologies) says:

We tend to massively underestimate the compounding returns of intelligence. As humans, we need to solve big problems. If you graduate Stanford at 22 and Google recruits you, you’ll work a 9-to-5. It’s probably more like an 11-to-3 in terms of hard work. They’ll pay well. It’s relaxing. But what they are actually doing is paying you to accept a much lower intellectual growth rate. When you recognize that intelligence is compounding, the cost of that missing long-term compounding is enormous. They’re not giving you the best opportunity of your life. Then a scary thing can happen: You might realize one day that you’ve lost your competitive edge. You won’t be the best anymore. You won’t be able to fall in love with new stuff. Things are cushy where you are. You get complacent and stall. So, run your prospective engineering hires through that narrative. Then show them the alternative: working at your startup.

My friend and fellow Y Combinator partner Aaron Iba gives incredible lectures when we visit top CS schools on behalf of YC. He says the sense you get from college recruiters is that you get to write new code. Most great hackers love to do this. But that's not what you get from a software engineering position at the tech giants. You get to maintain old code. The job should really be called software technician. 

So that's why starting a startup or working at a startup is so much more rewarding. You get to keep your edge. 



26 responses
Corollary: Never be the best at anything at any place, at anytime. If you're the best, you're probably learning from the best and thus not growing. Always do something that makes you uncomfortable.
"But that's not what you get from a software engineering" Çok haklısın.
Every programmer should shoot for working both for a startup AND an internet giant early in their career. The downside of startup first is that you are typically going to come out with *relatively* bad habits and a limited view of databases management and scalability.
It is more than just new code..

It is the combination of new code and the act of learning the process of selling that idea represented by code to a group of need both to get that 'edge'.

There is nothing wrong in dealing with old code. I have learnt a lot by reading "old" code written by super smart people. Speaking of Google, I have heard they have some amazing pieces of technology in place (protocol buffers, map reduce) which are very good resources for learning. There is no point in writing *new code* without having seen existing approaches, unless you want to re-invent the wheel.
cyrux004 -- All of those are Google-proprietary, and only tangentially help you when you need to do something on your own.

stationstops -- The appropriate way to scale the tech of a startup happens to be the most efficient. Never prematurely optimize, because if you do, you'll waste time on doing things that don't matter. Startups die generally because they don't create something people want, not because they fail to manage their databases.

sometimes having the resources enables you to do things that were not possible otherwise. everyone should be doing R&D.
An established company should actually provide an avenue to rewrite code or build a better mousetrap. Otherwise they begin to ossify. If done correctly, the old school can learn from the new school and vice versa.
I'll just add that I think the "write new code" vs. "maintain old code" distinction can be *too* appealing. I've been at a few midsize start-ups that are great at starting new projects and horrible at shepherding them into the happy stage that follows, where well-done work can require little maintenance but continue delivering value. Instead, projects accumulate "warts" and become less appealing to work on, and are eventually replaced by complete re-writes (if the company is lucky enough to last that long).
17 visitors upvoted this post.