Dodgy Coder: The Great AI Arbitrage: making a killing before your client wises up

There is a silent heist happening in the billable hours of every major software agency. We’ve entered a Golden Hour - a brief, high-stakes arbitrage window where the cost of production has cratered, but the price tag remains frozen in 2022.

History is littered with these windows. They are defined by Knowledge Asymmetry: the gap between what a client thinks it takes to build a product and the time it actually takes, using a magical new tech.

We are living through the Agentic AI version of this playbook. In the time it takes a client to finish a discovery meeting on Teams, an agent has already drafted, tested, and hallucinated its way through the first 40% of the codebase. The agency is billing for the month; the machine finished before the coffee got cold.

Here are the most iconic historical parallels to the Great AI Arbitrage:

1. The Gutenberg Gap: The Scribe vs The Printing Press

In the 1450s, a hand-copied Bible took a scribe about a year to complete. The price reflected 12 months of high-level professional labor.

The Tech: Johannes Gutenberg’s movable type press.

The Arbitrage: Early printers were notorious for trying to make their books look exactly like hand-copied manuscripts. They used the same ligatures, the same layouts, and sometimes even had artists hand-illuminate the first letters of a printed page. They charged scribe prices for a product that was essentially mass-produced.

2. The Master Gunsmith vs Interchangeable Parts

In the early 1800s, if a part of your rifle broke, a master gunsmith had to hand-forge a unique replacement. You paid for that master craftsmanship.

The Tech: The transition to interchangeable parts (mass production); machines began stamping out identical components.

The Arbitrage: Government contractors and factory owners keep billing for "fitting and filing" (the manual labor of a master gunsmith) long after the parts were being stamped out by low-skill laborers using jigs.

3. The Master Machinist vs CNC Programming

The CNC (Computer Numerical Control) lathe transition of machine shops during the 1980s is the most direct ancestor to the Agentic AI shift in software development. It wasn't just a change in tools; it was a total transfer of power and a massive, hidden profit engine.

Here is the deeper breakdown of what happened:

The "De-skilling" of the Master Machinist

Before CNC, a master machinist was a craftsman with specific knowledge. They didn't just turn handles on the lathe; they had "the feel" for how metal reacted to heat and speed. This was deep, learned knowledge - it couldn't be easily written down or taught to someone else, and it gave those machinists immense leverage over shop owners.

The Tech: G-code and microprocessors made that deep knowledge redundant. The new machine could replay its tool-paths perfectly every time, regardless of whether the operator understood the physics of the cut.
The Arbitrage: Shop owners realized they could buy one $80,000 CNC lathe (the price of about 12 manual lathes in 1981) and hire a "button-pusher" CNC operator at 40% of the Master Machinist's wage to run it.
The AI Parallel: Today, an agency can use Claude Code to replace a team of senior developer SMEs on a complex legacy project with one AI "prompt-pusher" keeping a zombie legacy project on life support.

The "Ghost Labor" Billing Model

This is where the real money was made. In the 1980s, the market rate for precision machining was anchored to the labor of the machinist.

The Old Reality: If a complex aerospace part took 10 hours to hand-mill, the client was quoted for 10 hours of a senior machinist's time.
The CNC Reality: The machine could churn out that same valve in 45 minutes. Shop owners would literally hide the humming CNC machines out the back, and keep the master machinist hovering out the front for the clients to meet.
The Play: Shops didn't tell the clients. They continued to bill for the 10 hours.
The Result: Profit margins jumped from a standard 20% to a staggering 500% on certain parts, just by exploiting the client's ignorance of the new CNC speed increases.

The Knowledge Monopoly

Historian David Noble argued that CNC machines weren't better when they first came out - manual machinists were often faster for one-off parts. However, management pushed CNC primarily because it removed the knowledge monopoly from the workers.

With CNC, the knowledge moved from the shop floor to the office floor.
It allowed management to dictate the pace of work without the machinist ever needing to be consulted.

The End Game

The end of the CNC arbitrage window provides a brutal preview of what happens when a secret productivity leap becomes a baseline industry standard. When the knowledge asymmetry vanishes, the pricing floor doesn't just sag, it collapses.

Once CNC machines became affordable and CAD/CAM software became user-friendly, the secret weapon became a commodity. Job shops that once made a killing started undercutting each other just to keep the spindles turning. Prices for standard parts plummeted by 60-80% in a few years. The bespoke premium evaporated because precision was no longer a rare skill - it was a machine setting.

The ultimate price-killer was when the giants - the Boeings and Fords of the world - realized they didn't need the middleman anymore. As the tacit knowledge of the shop floor was codified into G-Code, the 'craft' became a 'file' ... and files are infinitely portable. Clients realized they could hire a CNC programmer/operator to run the same machines in-house for a fraction of the agency's markup.

External shops were forced to drop their prices to Cost + 10% just to prevent their clients from in-sourcing the work entirely. The only shops that kept their high margins were the ones that made a pivot.

The Pivot: If you were just "making parts", you were a commodity. If you were providing "Aerospace-Grade Quality Assurance and Material Traceability" you could still charge a premium. You stopped selling the shaping of metal and started selling the certainty of the result.
Master Machinists became Manufacturing Engineers. They traded their manual "feel" for process optimization. A shop might have 5 CNC machines, but they still needed someone who understood the physics of metallurgy to ensure those machines didn't destroy $50,000 worth of carbide tools in a single afternoon.
The AI Parallel: In 2027, writing code will be a commodity. The thing clients will pay for in the future is functional and architectural assurance... did you build the right thing, and build the thing right?

The Generalist Dev Pivot: Breadth Over Depth

Software developers focus will need to shift upwards, onto software architecture, business analysis and the AI tools themselves - using AI tools to drive more productivity for business.

With this shift comes the need for a breadth of knowledge rather than a depth of knowledge. Specialist developers who continue to focus narrowly on just one framework / platform will slowly lose their knowledge monopoly to a machine that never sleeps, doesn't take coffee breaks and doesn't charge for overtime.