Self-improving company

A self-improving company is an organizational model reimagined as a set of recursive, self-improving AI loops that continuously improve without requiring human intervention at each step. The concept represents a fundamental rethinking of what a company is, rather than merely how it uses software tools.

Core concept

Tom Blomfield advocates for this model, arguing that companies should be understood as interconnected AI feedback loops rather than collections of human workers augmented by software: "you can reimagine what a company is as a set of recursive self-improving AI loops".

Contrast with the Copilot model

The self-improving company concept explicitly contrasts with the Copilot model — the framing of AI as a productivity assistant layered on top of existing human workflows. Blomfield characterizes this as a fundamentally flawed starting point: "I think that is actually a broken way of thinking about AI". Where the copilot framing keeps humans at the center with AI as support, the self-improving company inverts the structure so that AI loops operate autonomously across the organization.

Foundational prerequisites

A key prerequisite for building a self-improving company is making all organizational activity legible to AI systems. This requires recording everything — emails, decisions, communications, processes — because "if it is recorded, it happened to the AI". This accumulated organizational data forms what Blomfield calls the company brain: "all of your data, all of your emails, your DMs, the skills, the know-how, that is like the company brain".

The residual role of humans

Even within this model, humans are not eliminated but repositioned. They operate primarily at the edges where AI cannot yet function — handling novel situations, ethical judgment, and high-stakes moments: "Human beings reach into places the models can't go yet". This frames human involvement as a temporary boundary condition rather than a permanent structural feature of the organization.