When hiring for a technical position, I like to give candidates a coding exercise to test their mettle. In this post, I talk about traits to look for in coding exercises and use those to discuss why coding exercises are useful.

Obviously, much of this advice would apply to a similar professional exercise like a backlog grooming exercise for a prospective product owner, but I’m hiring for developers at the moment, so that’s the focus of this post.

When to give the exercise

Normally, as the hiring manager, I’ll talk to candidates about the position and answer any questions they might have about the position. In the course of that discussion about their background, the candidate and I jointly decide if the position is worth pursuing for both of us.

I like to administer the coding exercise before the first team interview, which I like to conduct in-person where possible. This gives the team and the candidate something concrete to discuss during the interview. Otherwise, I find that these discussions devolve into a resume review, which is less valuable than “talking shop”. With the results of a coding exercise in front of the team, they conduct a code review with the candidate, which naturally leads to deeper discussions, and also often elicits better probing questions from the candidate about the position and the environment.

Traits of Good Coding Exercises

Clearly explain the end goal

I think it’s crucial to let candidates know the philosophy behind the exercise, and especially what aspects of the exercise they can ignore. In general, I ask candidates to develop a few things well as opposed to many things poorly.

Time-boxed

In general, candidates will be completing this coding exercise on their own time. To be respectful of that fact, coding exercises should be time-boxed to a reasonable amount of time. During my MBA classes, most exams that required freeform responses lasted no more than two hours, and I think that’s an appropriate duration for a coding exercise also.

Having a time limit allows evaluators to compare candidates as well. While speed is not the only criteria, it helps me compare candidates to people on my current team and previous teams. If I’m trying to decide if a person is a junior, intermediate or senior developer, it’s useful to know what they can accomplish given a new problem in a given amount of time.

The downside to this approach is that time pressure can cause some aberrant behaviors, outside the normal stress of a job interview.

Some scaffolding

This time pressure often means that I will provide some scaffolding. This allows me to better target the exercise to whatever aspects of the solution I’m looking to examine, as opposed to seeing if the candidate can create a new project correctly. That’s why some exercises I give are refactoring exercises instead of new feature writing, so that the candidate has time to show me their design chops without having to create all the necessary infrastructure.

It also allows me to illustrate how the candidate will fit into the team. For example, if I’m hiring for a .NET developer on a team that has already has a MS-SQL database developer, I will often work with said database developer to provide a database schema with several useful stored procedures so that they get a head-start and so they know they don’t need to worry so much about databases in this position.

On the other hand, if the candidate will be expected to work on SQL stored procedures, I’ll design the exercise to include a SQL component to signal that. This exercise is as much about the candidate deciding if they want to work with us as much as it is for me. In my experience, if they don’t like SQL stored procedures, they will either submit an alternative or bow out of the candidacy.

Require minimal extra setup by the candidate

A good coding exercise will focus on whatever technologies are needed for the position, and will leverage a standard coding environment whenever possible. For example, if I am interested in finding out the candidate’s approach to coding in Java, I avoid obscure libraries that would distract candidates.

Where possible, I use common industry tools like Git to distribute and evaluate the exercise, because those are in heavy use by my team and it allows me to ensure that a candidate can use them. When using Git, I collect answers as patches rather than as a fork so that candidates can’t look at each others’ work.

A word on the SQL example above: unless I expect that all candidates will have a working SQL environment, I avoid technologies that require a lot of setup. I want developer candidates to focus on showing their value, not their environment setup expertise.

Include some interesting decisions, appropriate to the level of the position

To me, a test like FizzBuzz is like a parlor trick. Many programmers know this exercise, and the solutions don’t vary much. Thus, they are like a Boolean test of programming prowess: can you program? (true/false). I liken a coding exercise to an essay question rather than multiple choice, so I design meatier problems to solve than can be handled by one method.

I also detail the outcome I expect to see, and these are the only clarification questions I’ll answer from candidates before submitting their answer. Unless I’m interviewing for a junior position where a test that that might be appropriate or where I expect to be giving detailed instructions, I like to give instructions with the same level of ambiguity that the candidate can expect when they join the team. This means that intermediate candidates can expect more prescriptive objectives and more scaffolding than senior or architect candidates (more on scaffolding later).

My coding exercises often include more work than can be done in the time allowed (more on that later), to force candidates to make priority decisions. With more senior positions, I even leave architectural or framework decisions to the candidates, expecting that they have relevant experience to bring to bear. These exercises often include some red herrings as well, to see if candidates can deduce what portions of a system are important to an issue and those that can safely be ignored.

In many cases, I suggest that senior candidate introduce their own features, not only to see how they explain their feature but also to provide them a chance to exercise their imagination. To me, coding has elements of art as well as science, so a coding exercise would not be effective if there was only one acceptable solution.

Offsite, in comfort

One deficiency of whiteboard exercises where candidates work a simple coding problem on a whiteboard is that they favor people who are socially extroverted and quick to adapt to new circumstances. In my experience, people who excel at whiteboard exercises and are a good social fit make for good team members.

However, whiteboard tests tend to provide false negatives for introverted candidates, who may not be comfortable with the “thinking out-loud” component, as well as for junior or intermediate candidates that are not confident in their abilities. People tend to be concerned about syntax errors that an IDE would immediate point out or even correct. I observe that many candidates are self-conscious and don’t produce their best efforts.

To mitigate this, I like to allow candidates to work on coding exercises in one of two environments:

  • the candidate’s chosen environment, using their own equipment and setup
  • as a pair programming exercise, in-person using their equipment when possible

While pair programming is my personal preference, it does require synchronous timing between the evaluator and candidate. And, I’ll admit, often the issue is my schedule! Thus, I design my exercises for asynchronous use as a remote exercise and administer in-person when I can.

Conclusion

I hope you found this post interesting. If you’d like to see examples of coding exercises I’ve given, have a look at my coding-exercises repo. These two are in use for the positions I’m hiring for. If you’d like more information on those positions, hit me up.