Human Reverse Engineering

I spent a decade watching good teams spend most of their week on work that didn't need them: the same decisions, the same replies, the same coordination, over and over. The usual answer is to buy ten tools and wire them together. It never quite works, because none of them know how your business actually thinks.

When AI got good enough, the bottleneck moved. The hard part was no longer doing the work; it was explaining the work precisely enough that something else could do it the way you would. That's a human problem before it's a technical one, and it's the one HRE is built to solve. So I started encoding my own work, one piece at a time, and the method came out of doing it for real.

Most AI automation dies the same way: it looks great in a demo, runs once on a clean input, and quietly falls apart the first time it meets a real customer.

The fashionable version of this is the agent that does everything. It's a great demo and a bad system, because nobody encoded what the expert actually decides, or where the thing has to stop and ask. The model was never the problem. The missing judgment was.

HRE is the boring opposite. You don't hand the work to an autonomous black box and hope. You encode the decisions, draw the line AI must not cross, and keep a human on the other side of it. Less impressive in a demo, still standing a year later.

The method is deliberately boring, because boring is what survives production.

You take one process and decompose it completely, down to every decision a human makes inside it. For each decision you make the reasoning explicit: what the person looks at, what they weigh, what would change their mind. Then you draw the line most automation skips, what AI may decide and what it may not, and you route everything on the wrong side of that line back to a human: a question, a checkpoint, a draft to approve.

What's left is runnable. The judgment that used to live in someone's head now lives in software, with the human kept exactly where they add value and nowhere they don't. Naming the limits out loud is what makes the rest safe to hand over.

And it isn't one-and-done. Every correction a human makes feeds back into the encoding, with how to improve each piece documented rather than left to memory. The system sharpens over time instead of drifting, which is the difference between automation that ages well and automation that rots.

ITzWorking Forge is HRE applied to one of the hardest things an agency does: onboard a prospect and put a real price on what they're asking for.

It started where everyone starts, an advanced web form. We'd already wired an LLM into parts of it: suggesting what a prospect might need from their description, pre-filling the rest, quietly filtering out the ones with no budget. A good form. Still a form.

But pricing a project honestly was never a form. Done properly it's a qualification call, then a sales-and-technical conversation, at least two meetings before anyone can quote you without guessing or selling you something that breaks later. That's the premium part, and it's slow, expensive, and full of friction for the person on the other side.

So we encoded the conversation itself. The Forge helps you think through what you actually need, prices it on the spot, and lets you shape the scope, add a feature, drop one, watch the number move. No guessing, no chasing ten quotes, and the price you see is the price you pay.

Two meetings and a salesperson, replaced by a dashboard that carries the judgment, not just the math. The senior conversation everyone assumes needs a human in the room now runs self-serve.

ITzWorking Forge is the method applied to a single problem. Nexus, the operating system behind ITzWorking, is what happens when you keep going: it started basic and absorbed my work one piece at a time, each one reverse-engineered and handed over, how I write code, the decisions I take with clients, how I reply to them, how I run projects. Today the agency runs around 90% inside it. The last 10% is human on purpose: that's where the judgment and the quality live, and the difference between a service that feels premium and one that feels automated.