Why Replacement Is Harder Than It Looks
Disclaimer: This blog was AI Assisted. Just like AI levels up junior programmers, it can level up junior writers. Just like in both cases, I take responsibility for the output.
There is clearly a sentiment of “building AI Agents” and “making an app in a weekend” that seems unrealistically easy, but certainly stokes the AI-replacing-engineers hype. It can also stoke the fear in the programmers who have built their careers going deep into their tech stack.
If you are a software engineer that has not fully embraced AI yet, it is okay to ignore some of the noise, like jumping on the latest trend or model benchmarks. What you shouldn’t ignore, however, is its ability to automate things that we couldn’t automate reliably before. We engineers should embrace it and apply it as we would any other technology, library, or pattern in our toolkit.
There is a lot of fear that AI will replace developers and there is going to be some truth to it. However, there are a few truths that the AI hype train is overlooking. Here are some reasons why you should remain cautiously optimistic.
Automating Jobs away is nothing new
Prior to AI, we were the ones automating the jobs away. How successful were we in that? The technology to automate away most jobs already exists, but it takes an upfront investment and the project needs to succeed past human capabilities in order for the job to be completely replaced. AI replacing programmer jobs is no different than writing automated integration tests to replace your QA team. Some companies invest in it, but most companies don’t, and even the ones that do still need a technical development and QA team to maintain it.
There is a big difference between a Proof of Concept app and a production grade app. It does not always get there.
For now, we are still the ones doing the automation.
Trust and Security
Companies are hesitant to turn over all their proprietary data to AI. Companies with a lot to lose are not going to want to take a human out of the loop, and they are going to want that human to be someone that deeply understands how AI works and the output that it is producing.
Scarcity Makes Experts / Obsolescence Breeds Opportunity
Much of the same way COBOL programmers remained in demand because of its prevalence in financial systems, the future may hold that same type of regard for those that understand code. We often don’t realize how much tacit knowledge we hold about code, and how that knowledge will still be valuable in a world that will still be run on software, even if AI is doing most of the leg work.
We have spent time learning things, things that people may not be motivated to learn anymore. With every tech professional focused on learning AI, who is investing their time in the old stuff like Cloud infrastructure, Query Tuning or Containerization?
As we are starting to see for the first time drops in CS enrollment, with new engineers having the AI crutch inhibiting them from needing to get in the weeds. I’d say your coding knowledge is still a valuable asset to have.
More Projects under budget / within scope
When cost to develop goes down, more projects may get the greenlight. More companies will choose to build vs buy. This may lead to an increase in these types of projects:
- Rewriting Old apps
- Automation of processes that wasn’t possible pre-AI
- cleaning up and maintaining the AI Slop
The software development process doesn’t change. The process still requires experienced developers. We know software engineering is still about tradeoffs. We know the better of two bad solutions. We know that moving faster and writing more code is not always better.
Agents are just Applications. Applications need engineering
“Agents” is a term that gets thrown around very loosely. “In the future we expect an agent to do that”. People have different definitions for an agent, but an Agent is just an app, and an app still needs to be built! Prior to the code being written, there are still decisions, and after the code is written there are more. Where should it run? what is the best model to use?
The only difference is, instead of deterministic code, you are now working with nondeterministic code. But the goal remains the same. It is also your job to figure out whether an AI Agent is the right solution to the job. There is speed, cost and risk associated with AI Agents, an engineer is best suited to make that call.
Programming may be going away, but engineering is not.
Ignoring AI is not going to make it go away, and confidently embracing it will make you more valuable.
There are plenty of studies out there showing we have not yet seen the return on investment of the billions being funneled into AI spend. I believe it’s because a large percentage of programmers have yet to fully embrace it, and a larger percentage of programmers have not fully unlocked what its capable of. AI Projects are being force fed from the top down rather than the bottom up. They are diving in from a marketing sense but nibbling around the edges in an engineering sense. There is still a lot of untapped potential with AI, and I think it will be us programmers that actually build the thing that provides value, not just the shiny demo.
If you have a traditional programming background, you are going to need to lean into all the skills you’ve gathered outside of just learning coding syntax. Those skills are the perfect complement to these AI tools. The engineers that figure this out first will be the ones that matter most.