Aaron Francis

I think AI will make ambitious projects accessible to far more people. It can remove huge amounts of grunt work and let someone build a useful internal tool, automate a tedious process, or explore an idea without first becoming a professional programmer. Agent use will probably become ordinary office work, much like spreadsheets did: not everyone becomes a software engineer, but many more people can shape software around their own needs. That does not mean expertise, judgment, or taste disappears. Rough personal software can be tremendously useful even if it would never meet the standard for a production system serving thousands of people. Those are different contexts, and confusing them creates problems. You still need humans to decide what is worth making, recognize when the result is bad, and verify important work. The practical future, to me, looks less like handing everything to one infallible machine and more like orchestrating tools: stronger models directing other models, separate agents reviewing results, and better memory carrying context across conversations. Used that way, AI should not merely help us do the same work faster. It should expand the size of the things we believe we can attempt.

The biggest change would be evidence that AI cannot reliably move beyond impressive demos into sustained, context-rich work. If agents keep forgetting essential details, fail under ordinary real-world complexity, or require so much human checking that they do not actually remove grunt work, I would lower my expectations substantially. Conversely, dependable long-term memory and consistently strong verification would push me further in the optimistic direction. If agents could preserve context across projects, coordinate effectively, and catch one another’s mistakes without creating a new pile of supervision work, that would make them far more useful. The key question is not whether a model can produce one dazzling answer. It is whether people can trust a whole workflow enough to make ambitious things with it repeatedly.