The anatomy of the process

George Djabarov · August 11, 2026

My goal, stated plainly: a process that can build and maintain a large, complex, high-quality project with very few skilled people, as few as one. The kind that would traditionally take 50 people several years. Can a handful build it, and build it faster?

Spec-driven development, AI as an advanced assistant, "it's all process." People treat these as competing answers. They're not. It's the totality of how the pieces fit together, and the scarcest ingredient is human expertise.

A spec is required, but nowhere near sufficient on its own.

Someone with real domain expertise has to design the architecture and the engineering process. AI can assist, but the human drives it and owns the definition. The more experience that human has, the faster and better the result, and the fewer expensive corrections later. No different from leading a human team. The design needs enough detail, no gaps, no contradictions. It should capture both the what and the how, at the right level. Don't micromanage your AI any more than you would a good engineer. And AI can't tell you which tradeoffs actually serve the product. That judgment is yours.

From there, the design converts to implementation phases and tasks mostly unattended. But it's not a one-shot deal. The design, the phasing, the tasks and bugs are a living thing. Design docs are never perfect: gaps, bugs, sometimes outright contradictions. Both the design and the work evolve with the project.

So where do humans sit? They own the high-level requirements and product goals, and give feedback on the result, mostly by filing bugs. They're not in the core loop. But a mostly autonomous system needs a way to pull things up for a human: to resolve conflicting requirements, correct design drift, and call the balance when goals compete.

And the process itself? It brings structure, retains progress, and controls cost. AI doesn't have the context to remember everything, so you need mechanical bounds to keep it focused and lock in durable results, so later work doesn't reinvent them. The process is also what makes it reliable. Left alone, AI does genuinely silly things, like spawn a process-fork storm that kills the machine. Running unattended for long stretches needs real guardrails.

No single piece is critical. The way they fit together is what makes it practical. The more I build, the more it looks like this works, provided you have the right ingredients, human expertise the most critical one. If you don't know what you want, no machine can build it for you.

Making software will stay intellectually demanding, complex, and slow. You'll just be able to do it with far fewer highly skilled people.

That's where I've landed so far, and I'd like to hear where it breaks. I'm especially interested in AI pushed past the assistant role, running on its own with no human in the critical loop, not "help me write X, I'll review it." If that's you, what's carrying the most weight, and where does it fall apart?

Two links if you want to go deeper: Promise, the language and toolchain built this way, and BASE, the white paper on the process itself.

First published on LinkedIn on 11 August 2026.