now | blog | wiki | recipes | bookmarks | contact | about | donate
* * * back home * * *
pick up the pencil when it makes sense
2026-10-08
A few weeks ago at the Rails World conference, DHH (of Omarchy and 37signals fame) declared it was time for "pencils down".
If you aren't in the loop, what this is referencing is programmers no longer doing the majority of code-writing by hand. Largely, this is becoming truer and truer. We are seeing many agent-driven projects coming along faster than ever before, and more and more people (including me!) working directly with agents to accelerate the development of their software.
I like DHH, and I like agents. I have agents on my local navi machines that I run through ollama. If I need something a little more powerful, I reach for an open model through Ollama's Cloud. It is a very fun, very good workflow. While I work a lot with agents, I don't think I'm going to be fully pencils-down anytime soon, because I still also love the joy of writing by hand as well. For example, this very website has a very explicit rule - it will always be handwritten, with no agents touching it. Our main agent, lain, has its own website, and it is, of course, fully agent-driven. I have not written a single line of code for that site, except for creating the repo for it. That's because that site is a fun experiment: What happens if you hand an agent an empty GitHub repo and a domain?
Instead of going fully "pencils-down", here at xvoidsx, we function in a different fashion, which is through different modes of operation. Let's take a little look at what I mean by that.
We are a very small software/web studio, (currently) made up of two humans and a handful of agents running on our own machines and in the cloud. Over the last little while, as we have finally been pushing out ISOs for navi, our GNU/Linux distro (which I will be doing a more thorough write-up on later in this blog and in the wiki), I have identified the three main ways we tend to function around here.
I'll explore them a little bit here in this section.
I: Manual
The "manual" mode is the mode we all came from if we have been writing code since before the "agentic era". Where we open vim and some docs on the web, and put our hands on the keyboard. In the pencil analogy, this is where the pencil is squarely in the human's hand, chiseling the code, fingers dancing across the keyboard in a flow state.
II: Paired
The "paired" mode, as I like to refer to it, is where the human and the agent are working side-by-side. This can be either the human and agent pair working together across different areas of a repo, or even together in the same file. Both parties have a pencil in hand, maybe bouncing ideas back and forth. "What is the best way to do so-and-so?" This is where the conversation drives the code, and iteration between human and agent shapes the code.
III: Orchestration
This is where the human puts the pencil down, and acts as the "orchestrator" of the symphony. One or more agents are implementing the orchestrator's ideas into the repo, with iteration happening through feedback from the human through prompts instead of the human writing the code by hand.
Agents are a huge new development over the last year or so, and enables a lot of people to accelerate their software development practices, or allows people who have never written code to bring an idea they might have had living in their head for awhile to see that idea spring to life. Ideally, (most) folks won't abandon the idea of learning to code altogether - you should still understand software development best practices, how things like version control works, how to review mission-critical code that is agent-written, and the like.
So, where does this leave us? To continue with the pencil analogy, is picking up the pencil now a dead practice?
I think it largely depends on the project, personally. You're going to have loads of "artisan" projects that refuse any agent commits in them, at all, now and in the future, most likely. There are always going to be anti-AI people, at least in our own lifetimes. Maybe they will become less and less as time marches on and this new AI-accelerated world becomes more the norm. Right now, it is still quite the new concept, no matter how long like it feels it's been around, in reality, it's only been a few years, and the technology only continues to get better and better. Of course, some people are not comfortable with changing how they work just yet, and may not want to at all. That is fine, of course! Everyone is welcome to build their stuff as they see fit.
While I am not the biggest fan of most of the corporate-backed AI platforms for their potential privacy issues, I do like the technology as a whole, and I very much enjoy doing stuff with open-weights models. I run them myself, as I mentioned earlier, through Ollama, and I like checking out other models through Ollama's cloud. A lot of the open harnesses, such as OpenCode (which we made a fork for with our own setup inside, called Wisp for navi), Hermes, Pi, and others, are very good for doing stuff with agents. The herdr multiplexer which doubles as a sort of runtime for agents is another great tool in the toolbox, an agent-aware tmux of sorts that shows you agents' working state, whether or not your agents need your attention, and stuff like that, is genuinely good, and something else we ship in navi.
I like the "modes of operation" approach we have discussed in this post, and it has been serving us well here at xvoidsx. When I feel like the best thing to do is grab the pencil, I pick it up. Sometimes myself and an agent both have pencils in hand, working in different areas across a repo, or even on the same file back and forth. And other times, depending on what's needed, sometimes it is easiest to step back and put my own pencil down, and let the agents do their thing. All depends on what is being worked on, and that is the key.
This very website is my "pencil in hand" place. Everything here is and will continue to be handwritten, because it's my personal spot to play around, write, document things, and experiment. My own little web garden. Something like navi very much is a "paired" situation, and other projects and tools still are an "orchestration" experiment. No matter if the pencils are in our hands or on the table, it is still an amazingly fun time to continue to build stuff, regardless of which mode of operation we're working in at the time.
It's a brave new world, and it's going to be very interesting to see how we continue to go from here!