With the advent of AI, the software engineering job is evolving, but our responsibilities remain.
Producing code is no longer the hardest or most valuable part of the job.
Solving problems still very much is.
That’s where the real shift is happening.
AI won’t replace problem solvers.
But it will compress demand for pure code writers. It will make pure code writing cheap, and eventually optional.
Which side you’re on will largely determine your near future.
So, how can you lean in the right direction, today?
Be an actor, not a spectator
Watching from the sidelines will make you irrelevant faster than the actual impact of AI on the industry.
Actors drive outcomes; spectators only consume outputs.
This is where careers start to diverge.
If you’re an actor, you’ll compound your value.
Spectator? Anyone else can do it, even AI can do it.
You’re interchangeable and eventually… replaceable.
There are two main ways to be an actor today:
Tame and wield AI effectively
Learn to use it well, in ways that are efficient for you.
Best practices, workflows, skills, commands, prompt engineering, local setup, limitations …
Treat it as a power tool, not a magic box.Expand your problem-solving surface area
Work on codebases you’re not familiar with.
Use AI as a translation layer across languages, frameworks, and paradigms.
Apply your problem-solving skills, your judgement, your taste and reasoning, to a much wider surface area.
Be intentional and explicit
Never blindly accept AI output.
Review it. Sanitise it. Improve it.
Use your expertise, and yes, use AI again to refine it.
Follow a clear, repeatable process, make it better, and stick to it:
- 🤖 Plan | 👨‍💻👩‍💻 Chaperone
- 🤖 Execute | 👨‍💻👩‍💻 Chaperone
- 🤖 Verify | 👨‍💻👩‍💻 Chaperone
Any step can be done by AI.
None should be fully delegated to AI.
You remain accountable. You supervise. You decide.
Your (new) job: tame the tool
AI is just that. A tool.
Your job is to know how to use it effectively, reliably and safely.
It is nothing more than an extension of your expertise. Not a replacement of it.
People (used to) go to the circus to see lions.
But, at the end of the day, the tamer is the one getting paid.
In the past, staying relevant meant learning new languages, paradigms, frameworks, or platforms.
Today, it also means learning how to work with AI at a high level, because it only amplifies the expertise, taste, and judgement you already have.
For engineers early in their career, this means doubling down on fundamentals.
AI is only useful if you know what “good” looks like. It won’t give you opinions or instincts.
Skipping fundamentals and jumping straight into AI is how you become a fast, confident producer of bad code.
This is how you build experience: study opinionated work, internalise the trade-offs, and develop your own standards.
Do not delegate your responsibility
Delegating responsibility is how you make yourself irrelevant, slowly but surely.
Anyone can ask AI to generate code. My brother, a sushi chef, is doing it.
It can look, and be, production ready. But without judgement, standards, and review, it will progressively turn into unmaintainable slop. (Sorry bro')
Very few can steer it precisely, validate its output, and make it consistently useful.
AI can produce good code, engineers produce systems.
That's where your problem-solving skills matter most:
- understand the problem
- figure out the right solution
- orchestrate its implementation
- own the outcome
A good exercise to enforce this is to be able to precisely explain the problem, the solution and implementation.
Unless you really understand them, you won’t be able to explain them to someone else.
Invest in the right things
Experimenting with core models and first-party tooling is great.
Obsessing over every new thin wrapper or orchestrator built on top of them is not.
The pace of change is overwhelming. No one truly keeps up with everything, and that’s fine.
What matters isn’t tracking every new tool, but working from first principles: pick a vanilla LLM, understand its core capabilities and limits, and experiment deliberately.
Remain alert and open.
Do not over-invest in niche wrappers, orchestrators, or other products that merely patch temporary gaps in current agent capabilities.
Many of those gaps will disappear within months, with a simple agent update.
It’s the equivalent of mastering jQuery in the 2010s without understanding JavaScript.
Retain the methodologies and the workflows.
Not the particularities and specificities of the tool you’re using this week.
Invest in learning how to use LLMs in general, how to steer them the best, what standards to apply.Not just which button to click.
Think prompt engineering, skills, AGENTS.md, commands, things that transfer across tools and generations.
My optimistic hot take is, if anything, AI will ultimately raise the overall level of code quality.
It will widen the gap between bad code vs good code.
Given how easy it becomes to build instead of buying, new products will need to be exceptionally good to justify being acquired or paid for.
It raises expectations. It raises the value of engineers who think, reason, and decide.
Top comments (1)
This is a very encouraging post. It’s easy to feel worried about AI, but I like the idea of focusing on things AI can't do, like understanding the "why" behind a project.