Should we just build more?
Al is helping us be more productive, but the question is what should we actually do with it?
Traveling by train is still one of my favorite ways of transport because of the possibility to do whatever I like during the trip. It takes longer than flying in many cases, but I can get so much more done (not necessarily work-related), like coding, reading a book, and even catching up on Youtube videos I wanted to watch for a while (while train WiFi is - of course - always flawless…I’m a happy YouTube Premium subscriber, so downloading videos before the trip is super convenient).
This time I was watching Andrej Karpathys recent interview at Secquoia Capital about how he sees Agentic Engineering and its future. I admire Mr Karpathy because I think he has a unique way of transporting concepts in a way that many people understand (I intentionally don’t say developers because I’ve seen people from other professions learning so much in his videos). His breakdowns of how GPT works and frankly most of his other content is great. He also coined the term vibe-coding.
Agentic Coding Opens Endless Possibilities
However, in that interview he gave I couldn’t help but feel like there are profound pieces missing. I don’t want to spoil too much, but he describes the future impact of agents, how he sees them interacting in the future and how they’ll be able to produce even more and be even more efficient at it at the same time (this is vastly simplified, you should watch the talk yourself, it’s worth it).
And I think this is what tripped me up here: producing more. More code, more things to build, more usage, more tokens. This seems to be where this is all headed. AI enables us to do so much more. All the side projects (and he mentions that he has plenty, he truly is one of us) that have always just been lying around can now finally be created with the help of Claude Code, Cursor, or Codex.
The side-project you’ve always been thinking about? Let’s just ask Claude to build it. Your manager is pressuring you to ship that feauture? Just grab a coffee and ask Codex to implement it while you’re away. You’re asked to be tokenmaxxing anyways, right? And while we’re at it, have you also experienced that anxiety, that whenever you leave your desk and there’s no agent running…there’s this itch that you might be wasting time?
Not Everything Should Be Built
What I am missing in this communication is the point of: Why? I’m not doubting that technology is moving at a rapid pace and as an enthusiast - I freaking love it! There’s something profoundly impressive when you ask an agent to implement a feature and it just nails it. It is satisfying, exciting, mind-blowing and just incredibly cool.
But this also leads to us just mindlessly doing things instead of maybe stepping back and asking ourselves: do we actually need this? Does every side-project actually deserve to be built or might it actually be fine to not have the next cheap SaaS clone vibe-coded over the weekend that you’re throwing away anyways? Do we really need to create another slop site, just to share it on Twitter to demo how powerful our OpenClaw setup is?
It’s so easy to just kick of tasks, jump onto a different project and continue this forever. It gives us an incredible dopamine rush and we’re feeling crazy productive orchestrating our army of agents. But even not talking about the fact that models are still incredibly verbose in the code they write (this is a topic for a different time) I would vouch that this is not what the best usage of these tools is.
Where Should We Go From Here?
Please don’t get me wrong: I’m not saying we shouldn’t use AI or that it’s not an incredible tool for us. But what I’m vouching for is: let’s use its powers to build actually useful things. It is an incredible time to be alive and working in this space with developments and changes and unthinkable innovations happening this fast.
But so far I’ve been disappointed in the things AI has been used for. Yes, people are cranking out more code, GitHub is swamped with bots, productivity (as in: token usage) is at an all-time high. But are we really solving the problems that we’re having? As a friend has asked me a few days ago: Has AI cured cancer, yet?
I think we’re diving deep into the technology before stepping back and analyzing what to best use it for. Because all of this comes at a cost. There’s the first thing coming to mind: tokens are not free. They are subsidized by AI companies to drag us deeper into using them (which I don’t think is a bad thing, it’s very natural).
But there’s also the cost of us losing our MOAT as developers. Everyone can vibe-code. Everyone can continue prompting until there’s eventually enough code to catch every little edge case. But this doesn’t mean that it’s good code that has a reason to exist. I love agentic coding and the tooling around that and I hope we can find a better way to use it for awesome things that are worth building and that are solving real problems for real humans.