Introducing clawded.ai
semi◦diffusion: an AI tale of two cities is where I write about the business side of AI: how platforms operate, where demand is moving, and what that means for people building, working, and earning in this market.
Today, I am adding something practical to that coverage. Paid semi◦diffusion subscribers now get access to clawded.ai, a matching tool for remote AI data and evaluation work. It brings together openings across platforms, checks them at the source, and helps you identify which ones are worth your time.
Near the end of the 60 Minutes segment, the optimism meets its objection.
Mercor chief executive Brendan Foody offers a bullish view of employment in an AI economy. The program then turns to MIT economist Daron Acemoglu, who disputes the idea that training artificial intelligence will provide employment on the scale needed to absorb the disruption it could cause.
Earlier, the report examines a more immediate problem: unpredictable hours for the people doing that training. Foody acknowledges the variability and describes much of the work as part time.
For an expert already working through Mercor or micro1, these are different questions. Whether AI creates more jobs across the economy matters. Whether your project has another month of funding matters on Monday.
The connection between them is a business model.
An AI company buys human expertise because something about its system remains inadequate, uncertain, or untested. The expert helps close that gap. Successful work can therefore weaken the original reason for buying it. A task that was difficult becomes routine; an expensive review becomes easier to automate.
That does not mean the entire business disappears. It means the next purchase needs a reason.
Perhaps a model needs harder examples. Perhaps a customer wants to deploy it in an unfamiliar workflow. Perhaps a new release requires another test. Those possibilities can support a substantial industry. They do not automatically support the same people, performing the same tasks, at the same rates.
This is the distinction missing from much of the argument about AI jobs. A platform can win a new contract while one of its experts loses a queue. A client can keep spending while changing the kind of expertise it buys. Growth at the company level need not arrive as continuity in an individual worker’s calendar.
If you already have a platform account, another explanation of how to pass an interview is not enough. The more useful exercise is to read your assignment as evidence of what the customer needs.
Is it paying you to produce an answer, define a standard, investigate a failure, or decide whether a system is ready for actual use? What would make it return? What would let it stop?
These answers determine revenue visibility: how much future income you can reasonably anticipate. A rate describes the price of an available hour. It says little about how many such hours will exist, or whether doing them makes your next assignment easier to win.
Below is the business behind that distinction, followed by a practical framework for judging your next project: where repeat demand could come from, what can erode it, and which capabilities you can carry forward.
What the customer actually buys
Start with the platforms’ own descriptions of their services.