Imagine a wholesaler with product information scattered across several files. It wants a contractor to build a way for its support team to answer customer questions. If the brief says only “build an AI chatbot,” the contractor may deliver a fluent interface without agreement on who checks product facts, approves answers or handles a price change.
OpenAI published its overview of The Rebundling of Freelance Work in the AI Age, a report produced with Upwork, on November 10, 2026. The report cover gives November 2026 as its date. It examines Upwork postings through 2025 alongside categories of ChatGPT activity from a different period. Among postings with at least two tasks, historically uncommon task pairs accounted for a larger share of pairs in AI-related postings than in other postings. That is a reason to examine handoffs across skills, rather than buy by skill label alone. OpenAI–Upwork full report, pp. 1 and 11–12; OpenAI's November 10 overview
The report also finds that posted fixed-price budgets for AI-related Upwork jobs are about 52% higher than for non-AI-related jobs after accounting for work subcategory and posting month, using budget data through April 2026. The authors suggest broader scope may partly explain the difference. These are posted budgets, not actual payments, returns or proof that AI caused prices to rise. Full report, p. 9 and methodology on p. 18
The platform datasets do not match the same users or projects, and neither represents the whole labor market. The results cannot establish a trend in Thai hiring or settle whether AI replaces freelancers. Full report, introduction and limitations A more useful question for a business owner is how to specify the handoffs among people, source information and software.
Start with the work to be delivered
Fictional example: A wholesaler wants a knowledge base for its support team to answer questions about product models, warranties and ordering. The support manager owns the answers. IT controls document access and the test environment. The contractor organizes approved material, designs retrieval and draft answers, and hands over an editing guide. The contractor is not authorized to use real customer data or unapproved price lists.
Before requesting quotes, the support manager selects current documents, records the owner of each set and prepares common questions with approved answers. IT separates test material from production systems and defines how access will be revoked at handover. The company must provide this starting point; the contractor should not have to guess which file is authoritative.
The scope should name verifiable deliverables: source documents and their versions, how each answer cites its source, a test-question set, recorded test results and a handover procedure. Tie payment and delivery dates to those agreed deliverables and revision rounds, rather than the phrase “uses AI.”
Test both answers and the decision to stop
The support manager prepares two groups of test questions. The first can be answered from approved documents, such as the warranty of a model still on sale. The second lacks a reliable answer or contains conflicting information, such as the price of a discontinued model. A proposed pass condition is that the first group's answers agree with the specified document version and show their source. For the second, the system says it cannot confirm the answer and refers the question to a staff member instead of guessing a price.
The contractor supplies test results, document versions and an issue list. The support manager checks the wording; IT checks that the tool can access only approved material and that access can be revoked. Both sign off before acceptance. If an answer is wrong or cites an old document, stop using it, correct the source or retrieval method, and rerun the same tests before the support team uses it.
Before the next request for proposals, write a one-page brief with the answers needed, data owners, excluded data, handover evidence and what happens when the system cannot answer. It gives bidders a common scope to price and the company a concrete basis for acceptance.
AI-generated conceptual illustration of product information, draft answers and human review; it does not depict customer work.
