OpenAI is quietly paying specialists up to $500 an hour to teach ChatGPT the nuances of hundreds of occupations, from commercial aviation to soil science, in a project that acknowledges the uncomfortable reality that the AI may eventually replace the very workers training it.
Inside a project known internally as Stagecraft, thousands of freelancers are building what amounts to a comprehensive occupational simulator for ChatGPT. Run through data-labeling startup Handshake AI, the initiative has deployed between 3,000 and 4,000 contract workers to create detailed task simulations spanning over 400 distinct job titles, according to documents reviewed by Business Insider. The work pays at least $50 an hour for general contributors and up to $500 hourly for domain experts.
The scope is striking. A spreadsheet obtained by journalists lists everyone from commercial pilots and emergency medicine physicians to sculptors, pharmacists, and agricultural managers. Each contractor is asked to develop a professional persona and write prompts that mirror the actual decision-making their job demands. We are not talking about teaching a chatbot to write polite emails. This is about capturing the specialized reasoning that makes a soil scientist or an airfield operations specialist valuable.
For the past two years, the AI industry's race centered on brute-force capability: larger models, more parameters, longer training runs. That era produced impressive generalists, systems that can draft legal memos one minute and debug Python the next. But general competence only takes you so far in enterprise markets where companies pay for domain-specific accuracy.
Stagecraft signals where the competitive frontier now lies. OpenAI is investing real capital, potentially millions of dollars given the contractor headcount and hourly rates, to build occupational depth into ChatGPT. If you are a hospital network evaluating AI tools, a model that understands the specific triage decisions an emergency physician faces is far more valuable than one that simply knows medical terminology. The same logic applies to agriculture conglomerates, logistics firms, and media companies.
This trend extends beyond OpenAI. As Bloomberg recently noted, data-labeling platforms worldwide have pivoted from basic image tagging and sentiment classification toward highly specialized tasks requiring postgraduate degrees and professional licensure. The companies building foundation models have realized that the next leap in AI performance will not come purely from scaling compute, but from feeding models expert-caliber training data rooted in real professional workflows.
The Uncomfortable Economics of Training Your Replacement
One contractor, speaking anonymously, captured the tension plainly: everyone involved understood they were essentially building the system that might displace them. That awareness is not misplaced. A 2023 analysis by Goldman Sachs estimated that generative AI could automate roughly 300 million full-time jobs globally, with knowledge workers in fields like legal, finance, and administrative support facing the highest exposure.
Yet the picture is more layered than simple replacement. Agricultural managers are not interchangeable with chatbots, nor are emergency physicians. What Stagecraft appears to be building is a form of occupational augmentation: a model that can assist specialists by handling routine analytical work, drafting preliminary assessments, or surfacing relevant data during time-critical decisions. The economic disruption comes not from full automation but from reducing the number of junior specialists needed to support senior decision-makers.
Handshake's involvement adds another dimension worth watching. The San Francisco-based company built its brand as a career platform connecting young professionals with employers. Its expansion into AI data labeling represents a strategic pivot that leverages its existing network of skilled workers while tapping into the exploding demand for expert training data. Other startups including Scale AI, Surge AI, and Remotasks are chasing the same market, but Handshake's deep roster of early-career professionals across industries gives it a structural advantage in sourcing niche expertise quickly.
The broader implication for startups and enterprise strategists is clear. The value premium in AI is migrating from general-purpose models toward systems that demonstrate verifiable competence in specific professional domains. Companies building vertical AI applications, whether in healthcare, agriculture, logistics, or creative industries, should watch how quickly models trained on Stagecraft-style data can close the gap between generic AI output and genuine expert-level performance. The contractors building this data may be training their own replacements, but the companies that figure out occupational specialization first will likely capture the next wave of enterprise AI spending.
