Buckle up, gang — it's conspiracy theory time.
I've had this theory percolating in the back of my head for a while now. It was seeded as I started learning more about agents and skills, and throughout this whole conference, it has grown to be fully formed. Let me walk you through it in a bit of a scenario:
One of the quickest and most approachable ways to start customizing how you work with AI is learning how to write Skills. When you're writing your skills, it's critical that you explain when it should be used in very concrete terms. It's also important that the skill itself is clear, concise, and accurate, showing step-by-step how to do the process that it is describing. It should contain links to relevant context, and it should be updated semiregularly as processes change.
"But wait," you say, a little confusedly. "Isn't that just good process documentation? Like, aren't we already supposed to be doing that to help our teammates, new hires, and interns?"
"Ha ha ha!" I laugh agentically. "No, no, no, my sweet, little spring plum. We don't need software engineering anymore — that's what the agents are for."
A few weeks after this conversation, somebody at work cites an article shared at the AI Engineer World's Fair. It reported research that there are a multitude of benefits to agent performance if your codebase is well structured and well architected. If you have a practical, descriptive test suite, up-to-date documentation, a clear dev environment setup, production metrics and observability, and a functional CI/CD pipeline, agents generate correct code more often. They create security vulnerabilities less often. They waste less money, and they can be trusted to not break everything as often.
I see your eyes narrow suspiciously. You say, "OK, hang on a darn minute here. Those are definitely just good software engineering practices."
"Ohoho," I chortle into my AI-rtisanal, vibe-crafted, probiotic high-protein custom kombucha blend. "Absolutely not! Look at the date of the study. April of 2026! This is some high-octane, bleeding-edge research you're looking at here, not some decrepit industry standard." I levitate away as you stand there looking frustrated.
The very next day, I'm wheeling my Onewheel through our open-plan office blasting a podcast from a shoulder boombox, and you catch the AI podcast host talking about "in this new era of AI development, the real winners are using prompt caching and routing to manage cost. If you're not working to get the most out of every dollar you spend on AI, you're already behind."
"THAT'S. IT!" We all turn to see you standing on top of your adjustable sit-stand desk (currently in stand mode for extra elevation). "THE NEW HOT THING IS TO TRY TO OPTIMIZE COSTS AND CACHE YOUR API CALLS IN EFFICIENT WAYS?!?" The top of your head catches fire. You don't seem to notice. You look around wildly at the ceiling and walls. "I SEE YOUR GAME! I KNOW YOUR TRICKS! YOU CAN'T HIDE FROM ME!!!" An intern wheels your desk with you on it out into the hall and asks Claude to "hire another staff engineer. Make no mistakes." Another engineer walks through the door moments later, and your People Ops team gives each other high-fives. The head of People Ops updates a metric with a sigh: "Lost another one to the Conspiracy."
And scene.
There's a secret cabal of veteran software engineers pulling the strings behind the curtain to trick us into doing good software practices by telling us they're cool, new AI skills that we need to adopt or get left behind. I'll take no further questions at this point. Thank you.
(Content note: This is a tongue-in-cheek way of poking fun at the hype-ier parts of the industry, but I'm legitimately really excited about how much of this conference was focused on buckling down and using our new shiny tools with a good, solid, grounded engineering base beneath it.)
Top comments (0)