The Self-Improving Business
There’s a term I’ve been sitting with that I think names where all of this is headed: the self-improving business. I searched for it online and found nothing. Zero memetic real estate. That tells me we’re early, which is exactly where you want to be when a concept is this important.
Here’s the progression. Right now, the frontier conversation in AI is about self-improving AI: systems that get better at their own tasks over time without human intervention. Andrej Karpathy’s auto-research repo is a good example. The AI conducts research, evaluates its own results, and iterates. It’s impressive. But it’s still a tool improving at a single function.
The next level is a self-improving business: an organization where AI systems know the business well enough to train and integrate their own specialized models, identify bottlenecks, and optimize operations continuously. Not just one AI getting better at one thing, but an interconnected system of agents that collectively make the entire business smarter over time.
I want to be careful here, because this is not the pitch I lead with when talking to most business owners. Most businesses aren’t anywhere near this yet. They’re still figuring out how to use ChatGPT effectively or how to build their first internal tool. That’s fine. You have to walk before you can run. But internally, for those of us building at the frontier, the self-improving business is the north star. We help people get to the place where their businesses are self-improving.
What This Looks Like in Practice
I’m already seeing early versions of this in the wild. At a panel I moderated at SXSW, multiple builders described systems that are clearly approaching this threshold.
Ryland’s SwitchBooks is building toward fully autonomous bookkeeping. The system doesn’t just execute accounting tasks; it checks its own work, loops until correct, and operates with increasing independence. His stated goal (“completely make bookkeeping autonomous”) is essentially the self-improving business applied to accounting.
Jordan described an automated client pipeline that genuinely startled me. A client submits a request through Linear. A specialized agent team (frontend, backend, security, architect) picks it up, builds the solution, opens a GitHub PR, runs QA, deploys it, and then monitors client comments to trigger fix loops automatically. “I replaced myself as an engineer,” he said. The system doesn’t just build software. It responds to feedback and improves its output without Jordan touching it.
Rostam described a self-improvement loop where his system gathers annotation data and uses an LLM as a judge to update its own system engineering. The AI evaluates its own performance and rewrites its own instructions. That’s not just automation. That’s a system that learns.
Michael built a social media analysis skill that stores performance data over time and learns what content strategies actually work for each client. The longer it runs, the better it gets. No human has to teach it the patterns. It discovers them.
The Spiritual Dimension
Here’s what I find deeply compelling about this concept from a faith perspective. God designed creation to be fruitful and multiply, to grow and compound. A self-improving business is an expression of that design pattern applied to organizations. You plant something, you steward it faithfully, and it grows in ways that exceed what you could have achieved through effort alone. That’s not a new idea. That’s how vineyards work. That’s how the kingdom works.
The caution is the same one that applies to any powerful capability: a self-improving business still needs a dictator of truth. The system can optimize, but someone has to define what “better” means. Someone has to set the values. Someone has to make sure the optimization function is pointed at something worth optimizing for. Without that, you get a system that improves at the wrong things very efficiently. That’s not progress. That’s drift at scale.
The Honest Timeline
“This is the worst AI is ever going to be.” I keep saying this because it reframes the entire conversation. If the AI available today (which is already remarkable) is the floor, then the self-improving business is not a fantasy. It’s an inevitability for any organization that takes AI seriously over a multi-year horizon. The question is not whether your business will become self-improving. The question is whether you’ll be the one steering it when it does, or whether you’ll be watching from the outside while competitors figure it out first.
The gap between where most businesses are and where they could be is staggering. But the path is clear. Start with the basics. Build your tools. Encode your truth. Give your AI systems room to learn. And keep your eyes on the north star: a business that doesn’t just use AI, but genuinely improves itself.
Key Takeaway
The self-improving business is one where AI systems know the organization well enough to train their own models, identify bottlenecks, and optimize operations continuously, and it’s the inevitable destination for any company that takes AI seriously.
References
- Karpathy, Andrej. auto-research. An example of self-improving AI that conducts and evaluates its own research.