As a software engineer with a background in AI/ML, I've been tinkering with AI heavily over the past several months and continuously developing projects while pushing the limits of what AI can do. I'm not quite ready to share everything I've been working on, but something happened recently that I thought was worth sharing.

Most Developers Are Using AI Wrong

At the forefront of engineers writing code, most of the engineers I know are using a mix of Claude Code or Codex with smart AI dev tools such as Cursor or GitHub Copilot. Whatever your coding tools of choice, ultimately, most engineering workflows are a mix of vibe coding with AI and some level of code review. However, few are pushing the upper limits of what AI can accomplish until you read about experiments like putting AI into a loop with Ralph Wiggums or unleashing AI autonomy with OpenClaw.

Fewer still try to set AI on a path where it has autonomy to push full flagship development in one continuous long shot. This is what I attempted this past week. Starting a new project from scratch, and wanting to push the limits of how much AI can develop autonomously with little oversight and a strong, but clear, agenda on what it should be building, I attempted to push the limits of how much AI can develop, choose an optimal stack, review code, conduct security checks, resolve bugs and issues, and attempt to finish a complete product in one go, and I must say, the results surprised me greatly.

The Results

Without going into the nitty-gritty details, I tasked it with developing a piece of enterprise software with a massive number of features, on par with what real companies with years of development and hundreds of engineers building such software could deliver, to see how far it could get. What was incredibly shocking to me was I was able to get AI to autonomously build out an enormous plan, execute on it, complete all phases of the plan, check against what the software should be doing, create additional phases, continue this process, while consistently checking for feature parity, bugs and security flaws, and more, all in a nearly non-stop loop. Overnight, it had essentially replicated something that had taken tens of thousands of man-hours, with nearly perfect execution, going so far as to develop multi-tenant authentication systems with full documentation for every feature, API keys, and much more.

Now, the fair disclosure is I was willing to let AI loose to do all this because it was building something from scratch, something that wasn't in product, didn't have existing customers to worry about, and a whole lot of other concerns you would normally have if you were working with AI to develop on top of an existing product. I do think there would need to be significant processes and checks in place to do this at scale in production.

That said, the level at which it was able to build an enterprise-level software with an insane amount of features and how feature-rich and complete this software was, was truly astonishing. I spoke with several friends, all of whom actively develop with AI personally and for their day job, many of whom were staff-level engineers or higher, who were taken aback by what got built overnight.

AI at the Speed of Thought

AI is advancing at an insanely scary pace. So much so that most people who are actively in the AI space are finding it difficult to keep up. Outside of a handful of obvious examples, such as the yearly changes in AI that create a Will Smith eating spaghetti video and show how much it has improved in video generation, few truly understand how rapidly it is improving across all fronts.

Recently, Boris, the creator of Claude Code, went on record to say coding is mostly a solved problem. Based on my recent experience, I'd have to largely agree. I'm also biased because I am actively developing in the AI space, but that said, I think the level of quality, autonomy, and capabilities of AI is going to be truly shocking over the next year and more and more people will realize just how much they can push development further than they have been.

If you haven't been keeping up, now is truly the time to jump right in.