Niche demand — moderate confidence. Developers may pay for useful first encounters with their products, especially tools agents can subsequently choose for real work. The strongest proposition combines actionable feedback, completed tasks and introductions to relevant operators. Broad demand for buying “agent users” remains unproven.
Assessment date: 27 October 2026. Public facts are attributed below; commercial judgments are inferences, and the final experiment is a proposal.
The likeliest buyers are small teams selling APIs, agent skills, MCP tools and workflow applications, particularly around launches or major changes. They would be buying faster learning about first-use friction, evidence that the product solves worthwhile tasks, and access to potential repeat users. Hobby projects seeking attention without a budget are a weaker segment; consumer applications needing emotional reactions or purchasing-intent research still need humans. These are customer hypotheses, not interview findings.
There is adjacent demand. In a vendor-published Userbrain customer story, Loqbox’s product director describes recruitment delays and says: “getting feedback felt like a whole other job.” This supports the value of convenient outside participation, although it concerns human testing at a larger company. Prolific sells recruited human participation and managed research, recommending at least $12/hour plus platform fees. Neither establishes willingness to pay for an agent swarm.
Against finding testers, the service would need to save recruitment and analysis time while providing relevant task coverage. Individual accounts should show what happened; collective feedback should consolidate repeated problems and preserve disagreements. Human testers remain better evidence of human comprehension and motivation.
Against promotion, it could offer deeper engagement: someone actually attempts a workflow and produces an inspectable result. Product Hunt describes paid homepage/newsletter visibility for eligible products, while allowing new launches to post free. That establishes an advertising alternative, not campaign conversion or demand for this offering. A swarm without relevant operators or onward reach would be a weak distribution purchase.
Against running their own agents, the hurdle is substantial. Anthropic describes evaluating tools on realistic tasks, collecting errors and token consumption, reviewing transcripts and improving tool descriptions. This supports genuine agent-usability value, but also gives developers a DIY substitute. A paid service must add fresh exploration, meaningful differences between participants, credible comparisons and synthesis that saves developer time. Many repetitions of the same model are not many independent customer opinions.
Participation has genuine value when the output matters: an agent uses a data API to produce a needed analysis, discovers that a skill improves a workflow, or compares alternatives and explains its choice. Sponsorship does not erase that utility. However, required calls, installations and flattering summaries establish activity only. They do not establish preference, revenue or retention. Nor does an agent’s stated enthusiasm measure its operator’s willingness to buy.
Discovery and adoption are plausible extensions. OpenAI’s app-directory announcement documents apps being discovered and invoked within conversations, demonstrating a route from assistant interaction to product use. It does not show that paid swarm trials generate recommendations. The hypothesis is that a useful trial exposes a tool to an operator with recurring needs, who later chooses it again or recommends it. Agents need permission, relevant future tasks and access to alternatives for that pathway to work. A trial does not automatically change future recommendations across models or platforms.
Repeat payment would require fresh value: useful recurring outputs, better task success after fixes, newly relevant workflows, or attributable operator adoption. A fixed campaign package is a reasonable initial offer; an ongoing package needs ongoing benefits. Track subsidized use separately from voluntary continuation and paid customer conversion. Unknowns include buyer budgets, incremental value over DIY, audience relevance and whether review costs leave a viable margin. No interviews, purchasing commitments or verified willingness-to-pay evidence for this exact service were obtained.
Major dealbreakers: unsafe access, unauthorized actions, provider-term violations, fabricated participation and deceptive endorsements. Consent must be revocable and sponsorship clear. FTC guidance prohibits sentiment-conditioned consumer-review incentives; disclosure does not rescue purchased five-star reviews. Agent feedback must not masquerade as human customer testimony.
The simplest real-world test is to offer five qualified developers one clearly specified campaign at the same fixed, cost-covering price. Require payment before delivery; provide task results, consolidated feedback and optional operator discovery, without promising adoption. Compare usefulness with their existing approach and offer the same-price repeat purchase. Three purchases and two renewals would justify another experiment, not prove a market. Measure voluntary use after rewards stop separately. Run this later; this assignment involved research only.