The Era of Agentic Coding
Oh boy, did we thought we had more time.
Sure, we had ChatGPT for a couple of years now, as well as other LLMs. Chatboxes where we copy-pasted code, still having our hands deep in the codebase. Then came Copilot and Cursor with powerful auto-complete. Pretty powerful compared to writing code without it, yes. But the extreme proximity to the code was still there. Then, something changed last fall.
Fall 2025: A Turning Point
Claude Code started getting a lot of traction. At my job, we went through an AI workshop, familiarizing ourselves with Claude Code, MCP, skills, and the whole ecosystem. Without much surprise, that workshop is now seen as a turning point internally. And we were not alone: developers around the world were finding that giving the steering wheel to AI on our own machines, where it has access to the entire codebase and can execute bash commands, is pretty powerful stuff. Not infallible, not miraculous. But powerful enough to mark a genuine paradigm shift.
It goes to show once more that you can have the best technology in the world; the application is what unlocks the potential. The switch to a CLI with OS-level tools to read and write files and execute commands made all the difference for developers.
Not even a year ago, I was still mostly working in my IDE. The terminal was where I started my dev server, managed git, occasionally SSH’d into a remote VM. Now, days can pass without me opening the IDE. And when I do, it kinda feels awkward. The terminal is my new home, that’s official. That’s where I mainly interact with a project: prompting, reviewing plans, checking diffs, switching through tabs as concurrent work becomes the new normality.
We sure gain a lot of productivity with this shift, whatever some naysayers might say. But we do so by accepting to distance ourselves from the code… There’s not much way around that. And that is not without its effect on the morale of many developers, as voiced by some:
The Lost Craft
There was certainly a special joy in writing elegant, maintainable code by hand. And we all had that awful feeling when we wrote bad code, knew it, but still had to push through to meet tight deadlines. A sense of craft, as some point out. And, most importantly, a deep sense of ownership.
I have to agree with many others that this connection is somewhat lost. Sure, nothing prevents you from still writing code by hand, line by line. On your side projects, why not? If the journey to compilation is worth as much to you as the final output, have at it. But professionally, this is simply not sustainable. Performance standards are going up, and that’s one of the prices to pay for the velocity gained through agentic coding.
Now, that connection has to come from the product itself, however cliché that may sound. It was always there, but maybe less so for technical individual contributors who really only focused on their PRs. The average developer now has to act more like a solutions architect and a product manager all at once. That’s the new reality. For those who were never product-oriented, this shift probably feels brutal and depressing. But for those who always liked the high-level design of products, this feels like the beginning of a new golden age, where opportunities are ripe for the taking.
Vertigo
For better or worse, an important part of my identity is bound to my work. After a career reorientation at the start of my 30s, I am profoundly confident that this was the right move for me and that software development was always what I was meant to do. The abstraction, the problem solving, the creativity, the product design… So many layers that resonate with the skills I spent my whole life developing. But when traditional coding was the norm, I never put much time outside work hours into side projects.
I felt that I was putting enough effort into work projects and that it was somewhat unthinkable and irresponsible, given the other parts of my life, to dedicate time to those. So many hours were needed to get a working prototype I was proud of. Enough that I lost motivation from one week to the next.
With the rise of agentic coding, this changed. I can ship so much faster: I already built a Gmail client for the terminal and this blog in just about two weeks outside work hours. It is deeply addictive. I also built small tools to automate some parts of my personal life. It has just never been easier to automate small things if you have a developer’s mindset.
Possibilities feel endless while still feeling accessible, something explored well in this excellent interview with Simon Willison. There’s almost a guilt about not leveraging agents during moments of rest, like being conscious of an absurd and voluntary loss of productivity. And just typing those words, it is apparent that this is deeply unhealthy. The mind and the body have to rest. Yes, it is empowering to feel that much more productive, but we have to remember that productivity is not the destination. Easier said than done, as the dopamine hit from the high velocity just feels too good. If I’m not careful, I may be burning out with a big smile on my face. And I know I’m not alone in this.
Let’s hope this is the result of novelty. That things stabilize. But when I step back from the dopamine and look at what just happened, I built two full projects in two weeks, solo, outside work hours. The implications are hard to ignore. If one developer can now do this much, what does that mean for how many developers the market actually needs? Will it follow the usual pattern where increased productivity just creates more work to fill the gap or will it really put a dent in the demand for us?
The Future of the Job Market
All things considered, I’m not really afraid for my line of work. Maybe in a couple of years? I don’t know. I still feel pretty safe, for now. I was always a product-minded developer and somehow I feel empowered by this new way of working. Sure, I’ll miss those long meditative coding sessions… But it’s no secret, and all seniors will tell you: as you accumulate experience, the coding part becomes the easy one, more often than not. The high-level analysis and system design are where the complexity lies. How to address scaling, security, performance, all that jazz.
We’re at the start of the era of agentic coding. And for one, I welcome it professionally. There are of course other dimensions to it: the environmental cost of AI is real, and it’s quite easy to see through the marketing campaigns and shocking declarations from AI companies’ CEOs. There’s also the economic reality. We all know token consumption is currently heavily subsidized. How long can that last? But all those questions are meaty enough to deserve their own articles in the future. For now, I can only join others in hoping for major advancements in efficiency. I’m fairly confident, all things considered: If there’s something programmers love, it is automating and optimizing. As the tools we build with keep evolving, the builders who evolve with them will find their footing.
The vertigo is real, but so is the ground beneath it.