It’s time to question and revisit how we build software.
Our practices are designed for people writing/reading code, and we are no longer doing that. What got us here isn’t the best way going forward.
The core ideas that stood out to me:
Agents widen the gap between developers.
We used to debate whether the 10X engineer exists. DHH suggests the strongest developers using agents could become 100X or 1,000X more productive than the weakest developers working without them.
He reported producing about 60 times his usual monthly output, measured in lines of code. He also acknowledged that line counts are crude and Rust is more verbose than Ruby.
The implication is that you have to master agentic software engineering to be an outstanding engineer.
Writing code by hand is becoming the exception.
When DHH asked the keynote audience who still writes code by hand, only about five people raised their hands.
37signals stopped writing code by hand. When they find themselves doing so, it’s a signal to improve the agent workflow.
More native apps.
Web apps or hybrid apps made sense because they were easier to build and maintain.
Agents reduce the cost of building native apps so much that even small teams can build multiple native apps. That’s the path 37signals is taking.
We can drastically improve performance.
I spent most of my career working with Ruby and JavaScript/TypeScript because they provided a sensible compromise between developer happiness and performance.
Now we can give performance more weight when choosing a language and build in Rust, Zig, or C++. Syntax is less of an issue when you’re not writing the code.
English is the primary programming language.
We can spend more of our effort on the results/outcomes and let the agents choose the best languages for the job.
Programming is essentially a translation between human language and machine language. Now the machines speak human.
Question programming practices.
Our programming practices evolved for people.
Shared abstractions and principles like DRY may no longer make sense and become bottlenecks when multiple agents are working at once.
Bring Your Own Agent (BYOA).
Apps should let customers bring their own agents. DHH wants to use his butler, not your butler. I agree with that.
I don’t want to use multiple agents (Notion’s, Linear’s, Slack’s, …). I want to use my own agent that has access to all of those tools and knows my context and preferences.
APIs, MCP, and CLIs are table stakes for great products going forward.
You can build your own software.
Small annoyances justify custom software. My perspective is: SaaS is built for the masses. You can’t build a product that perfectly satisfies an individual’s needs when you’re building a product for tens or hundreds of thousands of people.
It made more sense to buy something because the cost of building your own wasn’t worth it. Now the cost is so much lower that building your own custom software is viable.
I built my own bookkeeping software after getting tired of QuickBooks. And my own personal task/project management software. They work better for me.