ai-agents-are-coming Is DHH crazy? A technical reality check from my little agent factory. Why writing code is becoming less important than deciding what should be built, and why Saturday Night Live is suddenly talking about AI agents.
TLDR: DHH is still sane, but his stand-up skills went through the roof. If you have not seen his Rails World 2026 keynote yet, go watch it right now: https://youtube.com/watch?v=vDjW_dRyKXY. Seriously. It is a masterpiece and quite possibly the most important video you will watch in the next seven days. 🐐🔥🔥🔥
Report from my little slop factory floor #
What can you see above? In my previous post I said that I was looking for a co-founder and described an idea for a control plane sitting above Cezar.
That was last Friday. It feels like ages ago.
The screenshot above is a small, working MVP of that idea. Planning started on September 20 after the Open Mercato HackOn in Wrocław. I genuinely thought getting to something useful would take me around a year. Instead, on Sunday I spent about five hours working on the MVP from an iPhone while visiting a cottage in the Beskid Śląski mountains.
Apparently a year is now a Sunday morning.
The interesting part is not even that I built a control plane quickly. The interesting part is how much the development loop has changed. The code can live somewhere else, the agents can live somewhere else, the compute can live somewhere else, and my interface can just be a phone. The workstation is slowly becoming optional.
There is some good news - Saturday Night Live is back #
Apparently AI safety has become mainstream enough for Dario Amodei to appear as a character on Weekend Update.
Dario Amodei is at the desk to assure that the future of humanity is safe from AI pic.twitter.com/HIem3WP5DX
September 27, 2026 And honestly, I love this. Not because AI risk is not serious - it is. But when something moves from research papers, Hacker News and technical conferences into Saturday Night Live, the cultural transition has happened too. AI agents are no longer some weird developer niche. They are becoming part of normal life.
And the sketch is funny. That helps.
What is next for Ventures Originals? #
The probable co-founder of [Ventures Originals](https://venturesoriginals.com/) is Mateusz Wiatrzyk, founder of our #aveCezar HackOn team:
[https://pl.linkedin.com/in/mateusz-wiatrzyk-91637516b](https://pl.linkedin.com/in/mateusz-wiatrzyk-91637516b)
Mateusz is AI-native, open-minded and, most importantly, willing to actually SHIP things for the future instead of endlessly discussing the future. 🏭🤖💪🏻🚀
The “Orchestration Control Plane” was not a bad idea. Technically, I still like it a lot. Economically, however, I do not currently see a strong enough business case there. I have a working MVP, and it is cool that I can manage more than twenty Cezar instances distributed basically wherever I want, but “cool” is not a business model.
For my actual workload, I went back to running a single Open Mercato Cezar. After testing the opposite extreme, I think one Cezar is probably enough even for a serious power user. So the project status is now: postponed.
Not dead. Not deleted. I just do not see a reason to keep building something because the architecture makes me happy. That may be the most useful business lesson I learned this week.
So naturally I already have another idea #
The next Ventures Originals idea is little-cezar, but only if Open Mercato is interested in collaborating.
Little-cezar would be a tool for children and, more broadly, EdTech. Kids are becoming builders much earlier than we did. Their imagination should not be wasted just because computers historically required years of syntax before they allowed you to make something interesting.
If software creation increasingly becomes a conversation between intent, agents, tools and verification, education should probably focus more on decomposition, curiosity, systems thinking, verification and understanding what is worth building. This is just an idea right now, but I like this one.
DHH is not crazy #
Now we need to talk about David Heinemeier Hansson.
His Rails World 2026 Opening Keynote:
https://www.youtube.com/watch?v=vDjW_dRyKXY is a masterpiece.
I do not mean that I literally agree with every sentence. I mean that I strongly agree with the direction of the argument. And the funniest part for me is the beginning of the talk - portrait painters.
I had never thought about programmers through that analogy before, and it is disturbingly good.
For centuries, if you wanted an accurate image of yourself or your family, somebody had to paint it. It could take an enormous amount of skilled human labor. Photography did not make painting disappear and did not make art disappear. It destroyed the assumption that accurately reproducing somebody’s appearance had to be done by a painter. That distinction matters.
Today almost nobody orders a family portrait that takes one to three years to paint, apart from perhaps billionaires and sufficiently determined nouveau riche. Painting survived. The economic role of the portrait painter changed.
And maybe software engineering is going through something similar.
In 1995, being able to build software by hand was an extraordinarily scarce skill. It was still scarce in 2015. It was still economically important in 2025.
2026 feels different.
Not because software suddenly stopped mattering.
Because the bottleneck moved.
Keep the happy memories #
If you do not have time for the full DHH keynote, this is a good recap: My god this is such a good speech that every SWE needs to hear. You know what? Every person should hear it
Keep the happy memories, eyes on the reality, be excited about the future. That’s the best that anyone can do pic.twitter.com/KKFUXfesFb
September 25, 2026 I really like that summary: Keep the happy memories. Eyes on reality. Be excited about the future.
Sounds reasonable to me.
I was never really a programmer #
I know how to code. I learned C at university. Later in life I wanted to become a full-stack web developer, which is one of the reasons I still love JavaScript.
C and pointers were difficult for me, and my sophisticated technical opinion around 2012 was basically: why the fuck do I need this pointer?
JavaScript felt completely different. Flexible, expressive, beautiful in its weirdness. I loved its functional side. TypeScript made the whole thing even better.
But I never really built my identity around being a programmer, and maybe that helps now.
Programming languages matter less to my daily work than they did even a year ago. AI agents are already surprisingly good across different languages and stacks, including old systems and languages like COBOL.
So instead of saying “programming stopped being a job”, which is obviously too broad, I would put my actual experience this way:
Around February 2026, writing code by hand stopped being the defining activity of software development for me.
Programming did not disappear. The abstraction boundary moved again.
We used to manipulate machine code, then assembly, then higher-level languages, then frameworks. Now I increasingly manipulate intent, context, specs, tests, tools and agents.
The code is still there. I just touch less of it directly.
“Do nothing” is becoming a technical skill #
This may be the part of DHH’s argument that interests me most.
When execution becomes extremely cheap, the important question is no longer “Can I build this?” More and more often, the answer is simply yes.
The better question becomes:
Should I build this?
My orchestration control plane is a perfect example. A week ago I thought it could become a company. Then I built the MVP, used it, looked at the economics and stopped.
That is not failure. That is progress.
When execution becomes cheap, the ability to say no becomes more valuable. Wait. Observe. Use the boring existing thing. Kill the project before you spend six months polishing it.
In a world where agents can implement ideas faster than humans can properly evaluate them, restraint might become one of the highest-leverage technical skills.
More context from Warsaw #
There is another video worth watching if you are interested in how software development actually looks in the second half of 2026.
Callstack’s panel:
Future of Software Development | Closing Panel at Agent Conf 2026
https://www.youtube.com/watch?v=0Swn0O6HxtU What I like about watching DHH and then watching engineers discuss the same transition is that the question is no longer whether coding agents work.
They work.
The interesting questions are now architectural and organizational: how much autonomy should agents get, how do we review their work, how do we preserve context, how do we control fleets, what stays deterministic and what remains human?
And increasingly: what should never have been built in the first place?
Changes on this blog #
One final thing.
欢迎 🇨🇳
Chinese readers are first-class citizens here.
This blog now has dedicated routes for:
- `/zh-cn/` - 简体中文
- `/zh-tw/` - 繁體中文
This post will live in four languages:
[https://blog.cygankiewicz.com/pl/ai-agents-are-coming/](https://blog.cygankiewicz.com/pl/ai-agents-are-coming/)
[https://blog.cygankiewicz.com/en/ai-agents-are-coming/](https://blog.cygankiewicz.com/en/ai-agents-are-coming/)
[https://blog.cygankiewicz.com/zh-cn/ai-agents-are-coming/](https://blog.cygankiewicz.com/zh-cn/ai-agents-are-coming/)
[https://blog.cygankiewicz.com/zh-tw/ai-agents-are-coming/](https://blog.cygankiewicz.com/zh-tw/ai-agents-are-coming/)
I also worked on Simplified and Traditional Chinese documentation for [Cezar](https://github.com/open-mercato/cezar).
Open source should not quietly assume that everybody worth talking to reads English comfortably. Making software available globally also means making its documentation understandable globally.
See you in the cloud ☁️ #
So that is today’s report from my little factory.
In roughly a week I went from wanting a co-founder for an orchestration control plane, to building the control plane, to discovering that I probably should not build a company around it.
Once, that would have felt like failure. Now I think it might be the opposite.
Implementation costs fell enough for me to test the idea instead of falling in love with a PowerPoint about the idea.
DHH argues for optimism. I would add one thing:
optimism with a kill switch.
Build quickly. Use it. Look at reality. Keep what works. Kill what does not.
Then go outside.
The agents will still be there when you come back.
☁️🏭🤖💪🏻