Discipline is Overrated
Delip Rao,
Aug 3, 2021

I advise a few founders and one general product heuristic seems to surface over and over:

 

Any aspect of your work or life requiring “discipline” is usually a sign of a tooling gap or an automation gap. Many of these gaps are product/startup opportunities waiting to be taken.

 To bring the conversation closer to home, consider a typical deep learning researcher as a target customer. The best researchers, besides creativity and knowledge, have something else — they are disciplined. But what exactly do we mean by “discipline”? The word discipline often leads to images of overwhelm and can lead to total paralysis in some.

 

We lionize those who demonstrate discipline. I think this is partly because of some weird masochistic epistemology that goes like this: truth (value) is bitter (found in hard work); therefore, working hard (with discipline) should be valuable. If you think a little about it, all of that is a value judgment. If we set the value judgments aside, we can begin to break down what being “disciplined” means. One way to look at discipline is doing the following two things:

 

  1. Showing up every day
  2. Doing something each day that’s aligned with the research goal.

It’s interesting to note that item #1 — showing up every day — gets easier IF item #2 gets easier each day.

Certain physical and mental health issues — like depression, for example — may make showing up every day hard regardless of progress in #2. That’s a complex and orthogonal problem, and I am intentionally putting it aside for this post. Be assured; I am not ignoring it.

 

To engineer productivity, it seems our biggest imperative should be in making #2 easier. A lot of developer/researcher productivity tools fall short of attempting that.

 

For example, AI-based code completion tools are great, but they don’t deal with the drudgery of experimentation while a typical researcher spends a relatively short amount of time coding. Drag-and-drop modeling tools look cool, but writing model code is the least painful (and probably most fun) part of a data science role. It seems a lot of the researcher productivity products aim to solve problems that are not making us more disciplined with less effort.

Dealing with messy data, working with and managing annotators, doing effective error analysis, writing and managing configuration files, dependency hells (yes, even with Docker), versioning everything so it can be reproduced, being able to recall a result or an experiment done a year ago in a new context, answering “data questions” from stakeholders on-demand, etc. are all examples of Sisyphean problems that suck willpower. There are products or product features that somewhat address each of these problems. Still, I’ve yet to experience a solid solution for any of these problems without involving a hodge-podge mashup and tears.

 

Aren’t you just restating “solve your own pain”?

No. Solving your pain is a useful product heuristic, but it’s different from “Discipline is Overrated”. Solving your pain is an obvious path to product building. It’s focusing on the negative (pain) and fixing the negative. Still, if we shift the frame a bit and focus on an apparent positive word like “discipline” in a value-free manner, we can then see underlying inefficiencies as new product opportunities. However, most of us are stuck with being in awe with discipline because, let’s face it, that’s been drilled into us as a good thing since childhood. In doing so, we fail to see high-value product opportunities right in front of us!

Most work is a combination of creativity and discipline. The goal of automation should be to move people from the zone of discipline to the zone of creativity. I hope the next generation of productivity products answer the question, “Is this feature making my user lazy?” in the affirmative. Because the lazier we get to be, the more creativity flourishes. Discipline truly is overrated.