Designing an Engineering Performance Management System from Scratch

In this case, the measure is your position on the career ladder, not e.g. the success/failure of your product or company as a whole (unless that is required to advance). I feel that because of this, performance management at big companies leads to unnecessary product churn. Think about how many messaging apps Google has released, or how every product seems to get a top-to-bottom redesign every year or two... having been on the inside, I think this is definitely a symptom of employees optimizing for climbing the career ladder. Promotion Driven Development is a real phenomenon and is in the long-term detrimental to everyone. I don't know an easy solution to it, since bigger and bigger projects are required to "prove" that you should be promoted. Maybe in a few years people will have experienced this enough to think of a better way.

For me, performance reviews are the single most demoralizing aspect of working at a large company. They are doubly demoralizing because I can see so many things wrong with them, but also can't imagine any system that would work better.

So what's wrong with them?

1. Over the long term, they incentivize people to work on things that have easily measurable short term costs and benefits. More difficult to measure costs and benefits get ignored. They also encourage sunk-cost fallacy. In almost any company, you're better off, at least in the short term, if you ship something that turns out to be a huge waste of resources later on, than if you make the decision "this isn't turning out the way we wanted, we should just cancel this project".

2. They pretend to be a relatively objective system, when in fact your performance rating at most companies is strongly (though not entirely) dependent on how your manager views you. Attempts to lessen the impact of point 1 above will generally increase the impact of this point. Research suggests that most of what passes for rational justifications are made up after the fact to support gut-level emotional decisions.

3. They replace intrinsic motivation with extrinsic motivation, which at least some research has shown can lead to much less durable motivation over the long run.

At one point I was very enamored of the way Valve does things, but now it seems that that sort of management approach produces super toxic cultures in the long run.

Ultimately it seems like the only way to get away from this stuff is to work for yourself or in a very small team with trusted partners.

Then you need to re-evaluate your definition of "Staff Engineer" at your company. Those titles are all relatively defined internally in the first place, just like junior/senior/staff/senior staff.

There is no industry-wide benchmark for those levels, so an internal curve is the only system that makes sense.

In the imaginary case where everyone in the company is a super high performer and everyone is equally as good, then it's not crazy to do away with those titles entirely as long as that's the case. But obviously that scenario won't last long as the company grows, and that's why titles/levels are gradually introduced.

Now it's more a performance review of the company and myself, and attempts to encourage my team to think about how we could improve the company. "What do you hate about your job? What do you like about your job? How could we apply some of the good points to other areas of the company?" Stuff like that.

If you're underperforming, I should already know (if I'm worthy of my position). And I definitely shouldn't be waiting a year to talk to you about it.

To me, the right kind of review (and done every six months) is something that takes an employee only an hour and the same for the manager. Its only purpose is to set and review goals specific to the function and success of the team(s) the employee is/was on. The scores are more like pass/fail or exceeds/meets/doesn't meet expectations. It gives a regular structured time for employee growth review and can help with correcting communication and other soft skills.