A few years ago, when AI coding assistants hit the scene, the headlines everywhere were mainly focusing on the potential employment loss for junior developers.
And that's still a concern that everyone should be aware of, because I don't think we've figured out some kind of solution to that potential issue.
I personally understood that fear back in the day. The code being generated by the older models at the time wasn't typically production-level, but it was definitely functional. It was enough to get a feature request working and at least ready for review.
However, a lot has changed since that time. And my personal concern this year is aimed more specifically at the mid-level developer market.
The Hollowing Out of the Middle
One of the most important roles that a mid-level developer has at a company, has less to do with code and more to do with understanding context.
A developer who has sat through 100 meetings spanning multiple years is a hugely valuable asset to any company. They understand the "why" to certain decisions made and they can remember (usually) the nuance in getting there.
I can't train any model on my meetings, because those meetings live in my head and in impossible to read notes on my desk.
And while, in the past, AI coding assistants were mainly good at scanning a file and generating a few generic function calls, they can now scan entire repo's, email dumps and internal documentation and remember all of it.
Essentially, coding AI agents are becoming pretty decent mid-level developers.
We saw it coming
It would have been a bit foolish to assume that agents weren't going to get smarter, faster, require less power and become more cost-effective. It was definitely more of a matter of "when", than "if".
2 years ago I wrote about what it would take for AI to steal our programming jobs. And truth be told, I thought we had at least 5 years before we should start worrying.
For reference, my main points were:
- Zero hallucinations
- Interconnectivity with other platforms
- Real-time data monitoring
- The ability to perform simulations, before any code execution
- And more power
While we aren't at 100% across the board yet, the progress is staggering. We’re seeing agents that can spin up their own containers to test code and RAG (Retrieval-Augmented Generation) systems that can ingest an entire company's history in minutes.
It turns out my five-year estimate was way off. We aren't waiting for some far-off "future" anymore. We’re watching it happen in real-time through weekly model updates and massive jumps in context windows.
When an AI can dig through a mess of old Jira tickets, forgotten Slack threads, and outdated READMEs faster than a human can, that "institutional knowledge" we relied on starts to feel a lot less like a safety net and more like a commodity.
The "moat" that mid-level developers built around their roles, being the person who "just knows how things work", is being drained by tools that never forget a single commit message.
The Silver Lining (or, The "Enterprise Lag")
But before we all start updating our resumes to become goat farmers, there’s a bit of good news.
Even if the technology is ready today, the world usually isn't. If history has taught us anything about tech trends, it’s that mass adoption moves at the speed of a snail. We’re talking a decade, not a fiscal quarter.
Think about how long it took for "The Cloud" to actually become the standard. There are still major banks and government agencies running on COBOL and local servers because moving to new tech is risky, expensive, and, most importantly, requires a ton of human consensus.
Why we still have time
Trust is earned slowly. Most companies aren't going to hand over the keys to their entire codebase and internal documentation to an AI agent overnight. There's risk, complications, added expenses and expertise requirements. More importantly though, there's not enough people who actually know how to manage an "AI workforce" yet.
Legal and security red tape is real and everywhere. The lawyers haven't even finished arguing about who owns the code an AI writes, let alone the privacy implications of feeding an agent every internal email since 2015.
The "Human" factor is still important. At the end of the day, managers still want a human being they can talk to (or blame) when things go sideways. It's typically much faster for a non-technical PM to go up to a cubicle to get a report, then it is for them to figure out which agent in the orchestration pool has access to the database.
So, while the AI is getting "mid-level" smart right now, we’ve likely got a solid ten-year buffer while the rest of the corporate world figures out how to actually use it without breaking everything.
Take my estimates with a grain of salt.
What can developers do now to prepare?
I think the most important thing to do first, is definitely not panic. Don't drop out of your CS program or abandon your bootcamp work just yet.
Because the truth is that the "human" software engineer role isn't going anywhere. It's just going to get much more difficult. As is true for anything in life.
100 years ago, filming a movie took exponentially less resources than it does today. In the modern era, the role of Director didn't vanish, it just became incredibly complex requiring an understanding that people 100 years ago couldn't even fathom.
The same is true for software engineers. 20 years ago, when I began my coding career, I can attest that being a junior dev was surprisingly easy. You read some docs, you copy/pasted some database code someone else wrote, you make a new page and you bound some fields.
Today though? The "easy" stuff is essentially gone.
Now, even a junior is expected to understand Docker containers, CI/CD pipelines, cloud infrastructure, and how to glue together five different APIs that all have different authentication methods.
We aren't just "writing code" anymore, we are managing complex systems.
And that is exactly where the mid-level developer of the future lives.
If you’re worried about AI taking your job, you have to stop thinking of yourself as a person who converts tickets into syntax. If that’s all you do, then yeah, the agent is going to win out eventually.
Instead, the path forward is becoming the person who orchestrates the AI. Think about it like this:
In the past, you spent 4 hours writing a complex SQL query.
In the future, you spend 10 minutes getting the AI to write the query, and 3 hours making sure that query doesn't accidentally expose sensitive data, kill the database performance, or break the reporting dashboard.
The "work" hasn't disappeared. It has just moved upstream. We are graduating from being the bricklayers to being the site foremen.
You still need to know how to lay a brick, otherwise, you won't know when the AI is building a wall that’s about to fall over, but your value is now in the oversight.
Since AI is getting better at the technical context, you need to double down on the business context.
Why does the company need this feature? Is it to satisfy a specific legal requirement or just to keep a single high-paying client happy?
An AI agent will confidently give you a solution that works technically but might be a disaster for user privacy or long-term maintainability.
The bar for entry is higher, and the middle of the market is getting crowded, but there is still a massive shortage of people who can actually finish a project and maintain it for years.
The learning curve is definitely getting steeper. But if you can stay curious, keep your head up, and learn to steer the tools instead of fearing them, there’s still plenty of room for you in the "middle."