What happens when AI codes better than we do?
For a second, let's forget AI is more like a sloppy junior coder. Let's stop feeling like cheating when prompting instead of typing. If AI writes reliable code, what would we offer as coders?
The other day, I found one post from Leon Mika that made me ask those questions.
Maybe writing code is no longer part of my "core offering" at this point in my career. Maybe it is the "judgement, tradeoffs, intents" and all the other buzzwords people throw around when describing a senior software engineer.
Leon's post made me think what we could offer.
Here's my list:
- Talking to end users to find out what they need
- Choosing what features to implement
- Choosing the right tools, stack, and frameworks
- Scoping projects into milestones or sprints
- Finding what to rewrite and when
- Coming up with de-risking plans
- Choosing the right time to scale
- Sharing past mistakes and lessons
- Vetting what to build, buy, or outsource
- Finding cost-effective "cloudification" strategies
What would you add to the list? Let me know in the comments.
Some coders already do those tasks, but soon they'll be everyday work for all of us.
In the meantime, AI is making us rediscover the practice of coding. And that's already a good point for AI.
When AI shines at coding, we need strong product thinking, communication, and other skills I cover in Street-Smart Coding. That's the roadmap I wish I had to become a senior coder.
Top comments (35)
I've always disliked the "you'll have to finish coding and go up the ladder" thing.
Anyway, even if AI takes over (this seems to me something far away right now), coding won't be over. There will still exist the need for coding for computer science research, for instance.
Oh yes, I chased perfect coding when looking for ways to climb up the ladder. But to my surprise, coding and management are different skills. Being good at one isn't a guarantee of success in the other.
Love this perspective! 💡 If AI handles the “typing” part, coders will shine in judgment, decision-making, and connecting with real user needs. Skills like scoping, de-risking, and sharing lessons from experience are the new superpowers AI can’t replace the human context behind the code.
Well, guess what. AI is good at that than too!
I wonder when AI will "take over". I have seen many no-code and low-code tools come and go. Anyone remebers Dreamweaver? Its messy output was a great motivation to learn HTML properly, which back then included designing and maintaining complex nested table layouts and iframes. WordPress theme builders, Wix, Webflow and so many others kept promising to obsolete coding. We're still coding. AI agents "can code" and their output might outperform absolute beginners.
But AI coding discussion has already moved on from enthusiasm to disappointment as bugs, outdated tech stacks and subtle implmentation failures creep into code bases thanks to uncritical acceptance of AI coding worse than the past decade's copy-and-paste-from-StackOverflow code slop. AI won't take over. AI is just another tool.
I do :)
Great point...and chain of thoughts.
Glad to hear I'm not the only one who thinks that...And as a thought experiment?
I think you are right for now! But give it a couple of years. I really think coding itself will be completley replaced. I dont see a reason why AI wont be able to do this. It has all the knowledge and ressourceses existing, it just needs to get trained more and more.
Just my personal opinion, and sorry for my english
That would be the case if langauges and libraries wouldn't introduce breaking changes, deprecations and new best practices and hypes every few years. And most of them are less strict than TypeScript, for example, lacking consistency, documentation and best practice examples. Even when AI moves on gets better training, this won't work unless our industry eventually fixes its fixation on "move fast, break things".
This is a good point. I'd like to think AI like some autonomous cars. They still a pair of hands in the wheel
Recently, I heard a more realistic point: AI will replace coding by hand (or from scratch) in the next 5 years.
at 2001 to 2004 dream weaver was magic tool , then we go to eclips . time was runaway now we at 2026
I think it boils down to two things: (1) making key technical decisions. This requires judgment, experience, and organizational context. It might be as weighty as choosing a backend stack or as nuanced as the right convention to follow for what logic to put in the backend vs the frontend. (2) making quality judgments, especially for non-functional quality. When do you demand changes vs accept imperfect results? How do you hold a high quality bar?
My guess is these areas will continue to require human judgment for some time.
That's my guess too, Derek. Thanks for sharing.
This was very much on my mind yesterday :) Ended up writing it up here: dev.to/derekcheng/managing-unrelia...
Ultimately our job has always been to help businesses figure out how to make computers do the things that will help them make money or save money - and to do those things in ways that are reasonably robust and maintainable etc. Writing code is just one way to do that.
As the commenter in the article already said - this comes down to judgement, tradeoffs, and intent. AI coding tools can help us to write better code faster, and to make better decisions faster. It's up to us to figure out how to use them well.
Love this perspective, Micah. Thanks for sharing.
Great article, Cesar! Your list nails the high-value skills coders bring beyond raw code—things like user empathy, strategic scoping, and de-risking that AI can't replicate yet.
This made me reflect on post where I explored Claude Opus 4.5's code generation prowess (read here )—it's a "virtual senior engineer," but as you say, it still needs human judgment for architecture and tradeoffs.
Coming up with de-risking plans is never gonna get old
Just these days, I witnessed some biz owners/managers panicking with a server upgrade
35 years as a developer. So crazy no one seems to realize only a small percentage of my time was ever writing code. Planning and implementation is most of the job. The code is just the fun part
I like to think coding is the easy part and figuring out what to code is the hard one.
Oh my 35 years damn
Those bullet points already sound like an entry-level Engineering Manager job role
Maybe when AI takes over, we all have to become EM. What do you think?
Actually, it's 2026, and those are the things AI agents do better than generative AI does coding.
Some comments may only be visible to logged-in visitors. Sign in to view all comments.