Recap: Software Should Work 2026 | Ben Cornia

Search

Recap: Software Should Work 2026

Published 2026-07-18<br>I attended my first software conference! The conference, hosted in Columbia, Missouri, was named Software Should Work 1. I really enjoyed what every speaker had to bring to the table. There were a couple of themes I took away from the conference and hopefully will convey intelligibly here.

Software is a social contract §

Software is essential to modern society. It runs in our phones, homes, cars, businesses and medical devices. Besides adopting an intensely ascetic lifestyle, it will touch almost everything about our daily life. Software should serve the needs of its users. In my opinion, software should work so well that the user should not even be cognizant of the software itself.

For example, a user shouldn't be worried that when they use an app that it's going to store metadata about them and sell it to third parties. Or that their garage door opener has paywalled features. Or when they get in their car that it has software bugs and will cause them to crash. Software should be invisible and frictionless to its end users.

And I think there are a lot of systems that do work well. The inventions of TCP/IP and Linux are the foundation of modern computing. Most people don't even know what those technologies are and that's a good thing. It means they are working as intended. The problems solved by those technologies have sufficiently been extracted away from the end user.

So what's driving the slop? Is it AI? Is it market pressure? Is it careless and greedy people? The answer is yes to all three. The sentiment that I came away with from Software Should Work is that the future of software is as much in our hands as it ever has been. Corporations may not have our best interest in mind but I believe people, in the aggregate, care about their fellow humans. That's why free and open source software continues to be our best bet for maintaining the social contract of software.

AI is applying immense pressure §

AI is breaking down trust in software as a product. AI is also breaking down trust in our own ability to reason about software . AI is becoming this strange intermediary between software engineers. It's both creating PR's and reviewing them under the guise of our smiling profiling pictures. We're directing each other to "just ask Claude". Like what the hell!?! I want to talk to a human for goodness sake. Not some human-proxy-slot-machine.

AI has turned up the temperature on software engineers. Expectations around our output have increased. Employers are tracking token usage. Teams within the same company are attempting to out innovate each other so they don't get laid off. AI is simultaneously finding vulnerabilities; creating vulnerabilities; and becoming a vulnerability itself. It's taxing maintainers and decreasing trust. As an industry, we are slowly forming an ant mill 2.

Other humans will always be our best resource. It's not to say that AI is inherently bad, but it will regress towards the mean. It's a self-reinforcing feedback loop. I want to experience the highs and lows of human relationship. I want the full human experience. All this to say that being in the same room with people whom I share interests but not necessarily the same values enlivened me.

It reminded me how fun it was to be a software engineer. We get to be excited about the bits and bobs of computers and drawing things on screens. It gave me new hope that there are smart people out there who see AI as an ends to a mean and not an end in and of itself.

Use better tools or build them yourself §

My final and favorite part of the conference was learning all the incredible tools people are bringing to bear to solve real problems. These include: formal methods and systems thinking, type systems, memory safe programming languages, dependency management, reproducible environments, testing and fuzzing, simpler web frameworks, understanding system limits and yes, even AI.

I'm particularly excited about formal methods because I think its a great thinking tool. It's reinvigorated my desire to do a deep dive on P 3. After this conference, I realized I'm not budgeting enought time to think about problems I am facing. Coding, in some sense, is a side effect of us building a mental model of a problem space. The most valuable exercise is the act of problem solving itself. We will continue to face novel problems. These tools help us build a scaffolding around problems so that we can exercise dominion over them.

My perspective has shifted on how I want to use AI moving forward. AI can write a lot of code much faster than me. And it might save me some time right now. But the real value is in my domain expertise; all the tacit knowledge accumulated from talking with customers and teammates. Overuse of AI attenuates my grasp of the domain, so there has to be a balance. AI is great at building throwaway code I don't care about. Which...