During PASS Summit, I wrote a post about the broadening data platform. I talked about the term Data Professional, and how I feel how it describes the changes going on the in SQL space. Here’s the problem: It’s a terrible guide. It’s a great description, it’s a wonderful attitude, it’s an ambitious goal; but it’s a terrible guide.
Being a data professional means being a jack of all trades. It means being a renaissance man (or woman). It’s means a career plan that looks like this:
Here’s my summary of Eugene’s argument:
- The concept of “data platform” is too broad to be meaningful, because
- nobody can gain expertise in the entire data platform. Therefore,
- focus on a particular slice of the platform.
Before I go on to make a counter-argument, let me start by saying that I agree almost whole-heartedly with this summary. But that never stopped me from arguing before…
So here’s my counter-argument: the concept of “data platform” is quite broad and nobody will master it all. Within that group, there are core skills, position-specific core skills, secondary skills, and tertiary skills. I recommend one of two paths:
- The Polyglot: be great at the core and position-specific core skills, good at secondary skills, and aware of tertiary skills. Be in the top 25% of your field at one to two skills, ideally one core and one secondary.
- The Specialist: be outstanding (in the top 1-3% of your field) at one to two skills.
With that in mind, let’s flesh this argument out.
Gauss and Mathematics
Carl Friedrich Gauss was the last true polyglot mathematician. He was able to make significant contributions to pretty much every branch of mathematics at the time, something no mathematician has been able to do since. The reason for this is not that Gauss was that much smarter than any other person since him, but rather that Gauss himself helped expand the world of mathematics considerably, so a 21st century Gauss would need to know about everything Gauss did plus what his contemporaries did plus what their chronological successors did. This spider-webbed growth of knowledge makes it impossible for one person to repeat what Gauss did.
Even though nobody can be a “true” polyglot mathematician—in the sense of knowing everything about mathematics at the present time—anymore, it doesn’t mean that “mathematics” is so broad a term as to be meaningless as a career plan. Instead, it means that we all have to specialize to some increasingly greater extent relative to the entire body of knowledge.
What’s The Right Level Of Specialization?
One of my colleagues, Brian Carrig, was working the SQL Clinic at SQL Saturday Raleigh. When he lost his original “I can help you with…” badge, he created his own.
This…might not be the right level of specialization. Brian’s a sharp enough guy that he knows more than the average practitioner on a wide range of topics, but I don’t think the choices are to be Brian or to be replaced by robo-developers; there are few enough people who can reach Brian’s level of skill that if these were the only choices, it’d be a dystopian nightmare for IT practitioners (and I’m not just saying that because I want Brian to provision me some more SQL Server instances).
So there has to be a middle ground between “know everything” and “exit the industry.” I agree with Eugene that we have to specialize, and here’s what I see, at least based off the current landscape.
Sub-Data Platform Job Categories
To save this from being a 5000-word essay, let’s pick four very broad categories for data platform jobs. These share some overlap and there are certainly people who don’t fit in any of these roles, so this is not a complete taxonomy. It should serve as a guide for us, however.
The four broad categories I have in mind are as follows: database developer, database administrator, Business Intelligence specialist, and data analyst. Database developers focus on writing and tuning queries and tend to specialize in performance tuning. Database administrators focus on backup and recovery, dealing with database corruption, and availability; they tend to specialize in process automation. Business Intelligence specialists build warehouses and migrate data from different systems into warehouses; this is a broad enough term that it’s hard to say what they specialize in, but pick one piece of the puzzle (cubes, warehouse modeling, ETL) and you’ll find people who specialize there. Finally, data analysts apply concepts of statistical analysis to business problems and come up with explanations or predictions of behavior.
Choosing Your Skills
I see four major categories of skill, but the specific details of what fits into each category will differ based on the role. Again, this is not intended to be a taxonomy but rather a conceptual description. We have the following concepts: core skills, position-specific core skills, secondary skills, and tertiary skills.
Core skills are skills which are common to all data platform professionals. These are relatively uncommon but tend to be fundamental to all positions. Think of things such as an understanding of SQL and relatively basic query tuning (which includes figuring out when to tune a query and what information is available on your platform for tuning queries). But really, when we think of core skills, we’re thinking of position-specific core skills.
As an example of a position-specific core skill, administrators need to know how to back up and restore the databases under their care. How you do this will differ based on the product, but if you administer a database without knowing how to recover it, you’re running a major risk and have a major skill gap. So basically, position-specific core skills are the things that you train juniors to do and expect mid-levels to know already.
Secondary and tertiary skills are even more nebulous, but I see them as skills which are ever-more-distant from the position-specific core skills. For a database administrator, the ability to write .NET code is a secondary skill: relatively few employers or practitioners think of a database administrator as someone who needs to write C# or F# code, but they can see how it’d apply to the position. A language like R would be a tertiary skill: a skill which the average practitioner has trouble tying back to day-to-day life. Most DBAs never think of using R for anything (although I’m trying to change that in my own small way).
Now, skills move over time. As Eugene points out in his post, I sincerely believe that administrators who don’t understand Powershell are at a serious disadvantage and that there will come a time that database administrators entirely lacking in Powershell scripts will be superfluous. We’re probably the better part of a decade away from containerization technologies like Docker having the same impact, but it’s ramping up as well. On the other side, an example of a technique that differentiated good from great database administrators a long time ago was the ability to lay out files on disk to minimize drive latency. SANs and later SSDs killed that skill altogether.
I wouldn’t describe these skill shifts as fluid, but rather tectonic; they don’t change overnight but they do significantly alter the landscape when they happen, and you don’t want to be on the wrong side of that tectonic shift.
So What’s The Answer?
In my eyes, the answer is to build your skills along one of two paths: the polyglot or the specialist. The polyglot knows a little about a lot but has a few major points of depth. A polyglot database developer might know a lot about writing PL/SQL and tuning Postgres queries, but also has enough experience to query Lucene, write some basic Informatica scripts, and maintain a Python-based ETL project. At many companies, a broad slice with good depth in a couple skills and relatively little depth in several skills is enough, and for our polyglot developer, it keeps doors open in case the market for Postgres developers flattens out for a few years or our developer wants to go down a new road.
In contrast to the polyglot, a specialist developer is elite at certain skills and knowingly ignorant of most others. A specialist SQL Server query tuner is in the top 1-3% of all developers at tuning SQL Server queries and knows all kinds of language and configuration tricks to squeeze percentages off of queries which take milliseconds or even microseconds, but might not know (or care) much about the right way to automate taking backups. You go to the polyglot to solve general, overarching problems but go to the specialist because you have a particular problem which is beyond the polyglot’s skill level.
In case the parallel isn’t completely clear, this model fits with the model for medical doctors: you have Primary Care Physicians/General Practitioners (PCPs or GPs) and you have specialists. The PCP knows how to diagnose issues and sees patients with a wide range of maladies. Sometimes, the PCP refers a patient to a specialist for further diagnosis or action. As an example, a PCP might stitch up a patient with a nasty gash, but that same PCP won’t rebuild a shattered femur; that PCP will refer the patient to a specialist in that area.
Is This Really The Right Direction?
A couple days before Eugene’s post, I had a discussion with a person about this topic. She was getting back into development after a couple years doing something a bit different, and one thing she noticed was the expectation of employees being more and more polyglot. Her argument is that we, as IT professionals, have a lot to do with this, as there’s a bit of a race to the bottom with developers wanting to learn more and willing to spend more and more time learning things. This makes IT jobs feel like running on a treadmill: you expend a lot of effort just trying to keep up. And this shows in how job titles and job role expectations have changed, including the concept of a data scientist (which I’ll touch upon at the end).
I’m not sure I agree with this assessment, but it does seem that more positions require (or at least request) knowledge of a range of skills and technologies, that it’s not enough to be “just” a T-SQL stored procedure developer in most shops. So to that extent, there seems to be a combination of developers innately moving this direction as well as job roles shifting in this direction.
To the extent that she is correct, there’s a good question as to how sustainable this strategy is, as the platform is expanding ever-further but we don’t have any more hours in the day. But at the same time, take a step back and this is nothing new: database developers are already a subset of all developers (much as we are loathe to admit this sometimes), so these categories are themselves cases of specialization. But let’s shelve that for a moment.
Anecdote: ElasticSearch And Me
It’s time for an anecdote. A few months ago, I started running a predictive analytics team. Our first project was to perform predictions of disk growth based on historical data. No big deal at all, except that all the data was stored in ElasticSearch and our DBA team wanted the results in ElasticSearch as well. My experience with ElasticSearch prior to this assignment was practically nil, but I went into it eager…at the beginning…
There were days that I wasted just figuring out how to do things that would take me five or ten minutes in SQL Server (particularly around aggregating data). In that sense, it was a complete waste of time to use ElasticSearch, and throughout that time period I felt like an idiot for struggling so hard to do things that I intuitively knew were extremely simple. It took a while, but I did muddle through the project, which means that I picked up some ElasticSearch experience. I’m definitely not good at writing ElasticSearch queries, but I’m better than I was three months ago, and that experience can help me out elsewhere if I need to use ElasticSearch in other projects or even to give me an idea of other ways to store and query data.
This is one of the most common ways that people learn: they muddle through things because they need to, either because the work requires it or because it’s a challenge or growth opportunity. If you’re able to take the results of that muddling through and apply it to other areas, you’ve got a leg up on your competition. And I think it’s easier to form quality systems when you have a few examples—it’s easier to reason accurately from several scenarios rather than just one scenario.
Summarizing a rather long blog post, I do agree with Eugene that “data platform” is a very broad space, and expecting someone to know everything about it would be folly. But that’s not unique. “Programmer” is extremely broad as well, but we don’t expect embedded systems developers to write databases (or even write database queries) or design responsive web applications. Doctors and lawyers specialize to extreme degrees, as do plenty of other professionals, and I see no reason to expect anything different from data platform professionals. I do believe that unless you are at the extreme right end of the distribution for certain skills (and can thus be a top-end specialist), you want to err in the direction of being broader than deeper, as it reduces the chances of getting caught in a sudden paradigm shift (remember how cool Web Forms was for about 4 years?) and risking your livelihood as a result.
One other point I want to make is that the broadness of this space shows the power of teamwork and complimentary skills. There’s an argument that a good scrum team is made up of a bunch of generalists who can all fill in each other’s roles on demand. I think that concept’s garbage for several reasons, one of which is that you often need specialists because specialists fix problems that generalists can’t. So instead of having a team of generalists, you have a team of people with different skills, some of which overlap and some of which complement each other: you have one or two data specialists, one or two UI specialists, one or two “backbone” specialists (usually .NET or Java developers), one or two QA specialists, etc. This says to me that there’s less a difference in kind than a difference in degree, even between the polyglot type and the specialist type: you can be a polyglot with respect to other data professionals (because you’re using several data platform technologies and are working across multiple parts of the stack) while being a specialist with respect to your development team (because you’re the database person).
Coda: Data Science
One bit at the end of Eugene’s post is that he’s interested in digging into data science. For a post criticizing the impossibility of living up to a definition, this did elicit a knowing chuckle. The problem is that the term “data scientist” is a microcosm of the issues with “data platform professional.” To be a data scientist, you should have development skills (preferably in multiple languages, including but not limited to SQL, R, and Python), a strong background in statistics (ideally having worked through PhD level courses), and a deep knowledge of the context of data (as in spending years getting to know the domain). I saw the perfect t-shirt today to describe these people.
There are very few people who have all three skill sets and yet that’s what being a data scientist requires. It’s the same problem as “data platform professional” but at a slightly smaller scale.