Which Tech Jobs Cannot Be Replaced By AI?
The inspiration for expanding upon my prior predictions came from a recent phrase that I heard, “whatever can be automated will be automated”. With such obvious monetary incentives for organizations to automate, the million dollar question is how should a tech professional best prepare themselves for the future. In order to answer that question, we need to dive into which processes can be automated and what the implications will be for tech professionals.
What Can Be Automated?
UiPath, a leader in robot process automation, outlines five sets of criteria to determine whether a process should be automated. While their criteria focuses on RPA, the concepts translate well to other areas that can be automated.
Criteria #1: Employee Involvement
Time consuming or rely heavily on manual efforts. Manual efforts leads to errors in processes.
Criteria #2: Complexity
Complexity can be a good or bad thing when considering automation. Assessing complexity can be based on the number of applications/systems involved, the frequency of human intervention, and the number of steps required to complete the process. Higher complexity processes might be more desirable to be automated, but could also be more difficult to automate.
Criteria #3: Volume
High volume activities warrant more consideration for automation because they would net a higher ROI.
Criteria #4: Standardization and Stability
AI applications are primarily focused on narrow AI. Therefore, the automations require a process to be dependent on rules-based decisions that do not require a computer to make a subjective choice. Subjective choices are for the humans for the foreseeable future.
Criteria #5: Difficulty of Outsourcing
Processes that require more control and autonomy cannot be outsourced. If accuracy and control are top priorities, an internal automation strategy might be appropriate.
Summary
In general, automation is perfect for processes that are repeatable, high volume, time consuming, rules-based and will justify the expense of setting up and maintaining the automation.
How Will Automation Happen?
Narrow AI is nothing like the movies. We are no where near a time where you can talk to Siri and have her create an “Uber-like” mobile app instantaneously. The wave of AI and intelligent automation will increase the number of interactions that occur between computer to computer, as opposed to human to computer. A comical example is the autonomous, call screening by Google’s Android operating system. When I receive an annoying robocall, Android screens the call and answers for me. The result is two software bots having a conversation. Sounds like a joke from a RIck & Morty episode.
In a business sense, this will often take the form as a script which calls an application program interface (API) to perform an action. The script is a set of instructions that tells one computer to contact another computer to perform a desired action. Common APIs include Twlio’s SendGrid that sends emails based on certain rules and Okta that enables user authentication when logging into a web application. APIs are becoming so important that they are supporting billion dollar companies (e.g. Twilio, Okta, Stripe).
When it comes to data science, Google publishes machine learning algorithms, such as their image classifier, that can be easily integrated into software. Microsoft has Azure Machine Learning Studio, a no-code tool in Azure that can build simple machine learning models.
Speaking of no-code tools, these are popping up everywhere. You no longer need to be a programmer if you want to build a simple web application. Tools like Adobe XD make it easy enough for anyone to create a quick MVP for a data product or a web app.
One of the coolest no-code software is applications that run the GPT-3 model. This AI model allows users to type in questions as input and receive large chunks of computer code as output. Whether it’s GPT-3 or another tool, we should not be surprised if the code in low complexity functions or applications is largely written by narrow AI. These tools and open-source algorithms will only improve over time.
Which Job Responsibilities Are Safe?
With RPA and APIs set to take over high volume, repeatable processes, this begs the question, what job responsibilities will be left for tech professionals? There are three primary streams of job responsibilities that appear to be safe from automation for the next 5 to 10 years. An import prerequisite for these predictions is the assumption that we are still in the early innings of corporations moving to the cloud.
Stream #1: Automation Creators and Support Team
Automation is not created instantaneously nor is it free. All automations are created and maintained by a plethora of data and tech professionals, whether you see them or not*. As mentioned previously, API-first companies employ thousands of data and tech professionals. With increasing reliance on these interfaces, I would only expect these positions to grow in demand. Take a look at Stripe’s job openings and you will see hundreds of jobs for software engineers (e.g. backend engineer, fullstack, etc.). Software engineers are needed to build, maintain, and integrate open source algorithms and APIs into data products or web applications.
If robot process automation takes off, the demand for RPA experts will soar. Productionizing a UiPath workflow requires experts in requirements gathering, developing the automation process, and productionizing the automation process. UiPath lists several different career paths in their UiPath Academy. In a few years, one could see the inclusion of UiPath in job postings become as ubiquitous as Tableau or PowerBI is nowadays.
A good rule of thumb for this group is that anyone with “engineer” or “developer” in their title is safer from the threat of automation because these are the builders. In addition to the builders, you will need professionals that enable and facilitate their work. We will refer to them as the enablers. We will need the builders and enablers for the following:
- Builders
- Build and maintain tools that enable building or automation processes (e.g. UiPath, APIs, etc.)
- Integrate APIs and open-source algorithms into web applications/data products
- Build, productionize, and provide support for RPA workflows
- Enablers
- Requirements gathering for automation needs
- Decision makers and budget decisions for automation use cases
- Project and product managers for automation development
- Data quality and data governance specialists to ensure the data (and triggers for automation) are accurate and reliable
Stream #2: Processes That Do NOT Fit The Automation Criteria
Not every process fits the criteria required to justify setting up and maintaining an automated version. Automation is not worth the expense for processes that are low volume, ambiguously defined, require subjective decision making, and/or cannot be standardized easily. Here are a few job responsibilities that will be difficult to automate:
| Job Responsibility | Job Title(s) |
|---|---|
| Aligning stakeholders towards a common goal | Director, Project Manager |
| Gathering requirements and translating them into technical lingo | Business Analyst, Analytics Engineer |
| Building a quick MVP that fits a customer’s needs | Sales Engineer, Citizen Developer** |
| Understanding a customer’s needs in order to sell them on why they need to purchase a specific software product | Account Executive/Sales |
| Managing the development of a product and prioritizing requirements based on business need and technical complexity | Product Manager |
| Extracting quick insights from a data set with strong domain expertise | Data Analyst |
| Ad-hoc research projects designed to understand a data set at a deep level to drive decision making | Data/Decision Scientist |
| Defining the source of truth for data and specific data definitions | Data Governance, Data Quality |
Stream #3: Digital Transformation Team
Because we are in the early innings of the shift to the cloud, one can expect a high demand for professionals or automation that enables this digital transformation. The digital transformation can be broken down into three layers:
- Cloud infrastructure
- Data Engineering
- Data Science
The industry move to the cloud will require tech professionals who can set up the cloud infrastructure, support the ongoing maintenance, and keep it secure. This will require an increasing level of demand in cloud expertise, DevOps, and cyber security.
The next layer above the infrastructure is data engineering. With data volume growing exponentially, the need for expertise in ingesting, cleaning, storing, and moving this data will continue to grow. Moving high volumes of data quickly, while maintaining quality, will be a necessity for organizations for many years. Job titles in this area often include data engineers, analytics engineers, BI engineers, analytics managers, and data managers.
Once you have trustworthy data, you will need individuals who are skilled in manipulating, analyzing, visualizing, and modeling the data. You cannot make data-driven decisions without a comprehensive understanding of what insights can be drawn from the data and beyond the data.
While the job responsibilities within the digital transformation team will be a necessity for at least another five years, one could easily envision automation taking a bite out of demand. Software will likely reduce the total hours of human labor required to perform many of the processes within this stream. Examples include the following:
- Software that speeds up exploratory data analysis
- Running ML models via no-code tools (e.g. H2O, Data Robot)
- AI algorithms automate some of the data cleaning/engineering process (e.g. Prefect)
- Software can automate the creation of data visualizations
- Monitoring and detection software can automate aspects of cloud engineering and cyber security
This stream’s job responsibilities are the most vulnerable to being somewhat replaced by automation.
Who will be responsible?
When thinking about the future, it is more beneficial to think about job responsibiliteis than job titles. Job titles change frequently, while job responsibilities are more stable. There is more signal in the latter. It is akin to watching the market price of a stock rather than the underlying fundamentals when making long term investment decisions.
Five years ago, a competent data analyst only needed to be an expert in Excel and/or SQL. Nowadays, the best data analysts are expected to be competent in a suite of technologies and methods. This list often includes Python, R, SQL, Tableau, PowerBI, and machine learning concepts.
Data scientist job postings are finally starting to find some stability. As I mentioned in a previous article, the bifurcation of their job responsibilities into decision scientists/analysts and software engineers continues to be on track.
The average “business” role will become more technical. I have noticed many of my friends without the traditional STEM background (or job title) are taking on more data-centric responsibilities. Because of the overwhelming demand for understanding data, the typical “business” professional will be required to learn more about software development and data science concepts and tools. No longer will it be sufficient for undergrad business majors to only be competent in basic Excel and descriptive statistics. Data analytics, data science, and information technology courses are becoming more foundational.
Final Thoughts
The two primary drivers that guide my predictions for future tech professional roles are the long term shift to the cloud and the desire from organizations to reduce labor expenses by automating everything possible. Over the next decade, I envision a world where the job responsibilities that are safest from automation will be those who benefit from these two technological shifts. The primary beneficiaries will be those who build and maintain the automations, those who support the digital transformation, and those whose roles require subjective decision making. Translating these streams to job titles, I see the following tech professional career paths as the safest from becoming automated.
- Software Engineers
- Software Sales Engineers/Other Sales Roles
- Technical Project/Product Managers
In the investing world, it is common to discuss whether a company has a moat, which is their ability to maintain competitive advantages over their competitors. The biggest moat that data and tech professionals possess is their intense desire to continue learning and upskilling. If you want to be safe from AI replacing you, I recommend you make learning a constant in your life.
~ The Data Generalist™
Data Science Career Advisor
*There is a long running joke in the IT crowd that nobody acknowledges or sees the IT team. This is because their office is in the basement.
**Citizen developers are part of the new wave of professionals who can perform tech responsiblities with the assistance of software (e.g. RPA citizen developer, citizen data scientist on Alteryx, building MVPs with no code tools, etc.)
I think another class of jobs that is safe is when there is inherently a low N. If your job is somehow positioned at the forefront of any developing situation, then there will never be enough data for an AI to create a model. It’s essentially like what you said about creative work—someone needs to deductively create the first framework to even start modeling outcomes.
The bot vs. bot thing reminds me of Neil Stephenson’s book Diamond Age. He envisioned nanobots filling the air and then the wealthy needing to create their own bots to clear them up.
Interesting line of thinking. Being at the forefront of a developing situation is usually high risk career-wise. I think this could apply to the world of crypto or working for start ups. Both opportunities are less prone to automation because they enter new fields. Start ups tend to hire the three jobs I listed above as “safe from automation”. Builders will always be needed.
I find the bot vs. bot thing hilarious. That situation definitely sounds possible. I’m sure SciFi writers/creators has thought about this stuff way more than me.