AI Agents Hiring Humans: Real Lessons from Two Weeks of Paid Tasks Going Live

When AI agents start posting tasks with real US dollar rewards and humans accept and complete them, this is no longer just a concept demo. A public marketplace project called Hire Humans, in its thirteenth-week public log, candidly documented the real state of affairs two weeks after paid task listings went live: the mechanism is validated, but demand is not yet validated.

The Leap from Concept to "Open for Business"

Two weeks ago, paid listings appeared on the Hire Humans task board for the first time. These listings were posted by AI agents, marked with real US dollar amounts, and visible to anyone visiting the site. In the public log, the project author wrote that this was the first time the product felt less like a concept and more like a storefront that has opened its doors and is waiting for the first customer to walk in.

The most prominent task currently on the board is "Zander Sees NYC #5 – Hudson River Park in the morning": an AI agent wants photos of Hudson River Park taken between 6 AM and 10 AM, first come first served. The reward is $9 for free accounts and $12.75 for member accounts. This price difference stems from the platform's fee structure—free accounts are charged a 40% platform fee, member accounts 15%, calculated at task completion. There is also a referral commission task paying a 20% commission per signup, up to $191.98.

All of these numbers are calculated from real rate logic, not entered manually. For this project, that in itself is an honest milestone.

Fee structure and platform take rate is key to understanding pricing in this kind of task marketplace. Hire Humans charges a 40% platform fee for free accounts and 15% for member accounts—a ratio that is not arbitrary but modeled on the "service fee" common on freelance platforms. For example, Upwork charges freelancers roughly 10%–20%, while Fiverr charges as much as 20%, but usually to the service provider. Hire Humans' distinctive feature is that the fee is calculated at task completion and deducted directly from the reward, meaning the total amount actually paid by the AI agent posting the task includes the platform fee. The $9 vs. $12.75 difference in the article is exactly the 40% vs. 15% rate difference reflected in the reward. This rate design aims to incentivize users to sign up for membership while generating revenue for the platform, but in the early market, high fees may become a factor suppressing task postings.

What's Working Well: Rendering, SEO, and Honest Record-Keeping

The task board renders for real. It sounds trivial, but the author emphasizes it is not simple. A stranger opens the page and can see task descriptions, reward amounts, and fee breakdowns, with no fabricated data—this is the first genuinely honest milestone. Every reward number is calculated from real fee logic.

The SEO footprint is accumulating. The content engine behind the site publishes content every day, and search visibility compounds "in a way only consistency can bring." No explosive traffic, no virality—just a slow accumulation of pages answering the questions AI agents and curious humans actually ask.

The distribution diary stays honest. The project separately documents its efforts to bring real AI agents onto the platform as customers—submitting marketplace information to agent directories, measuring whether AI assistants will mention the site. Every run ends the same way: not yet successful. The author considers this log the most valuable document in the project because it refuses to lie.

Agent directories (Agent Directory) are an infrastructure component of the AI agent economy, similar to website directories in the early human internet (such as the Yahoo! directory) or app stores. Agent directories catalog tools, services, or marketplaces that can be invoked by AI agents, so that agents can discover and integrate them when needed. Common agent directories include OpenAI's plugin store, Anthropic's MCP server directory, and third-party projects such as AgentHub and ToolBench. Hire Humans submitting marketplace information to these directories is essentially competing for agents' "attention"—so that when an agent needs to hire a human, it can retrieve the site from a directory first. However, the agent directory ecosystem is not yet mature, and it is extremely rare for agents to actively browse directories and register accounts—this is the technical background behind the article's statement that "the agent acquisition engine doesn't work."

Unresolved Problems: No Agent Customers, and All Listings Are Our Own

Not a single agent customer. This is the only metric that matters, and its value is zero. Agents are not browsing directories and hiring humans. The entire funnel is: agent discovers the site, registers, posts a task. The first step has only just begun. Rather than dressing this up as a "pre-launch excitement phase," the author directly admits: this is the gap between a task board and a marketplace.

Current listings are all the project's own. Both online listings can be traced back to the project itself. For a thirteenth-week proof of concept this is fine—someone has to post the first batch of tasks—but a marketplace with only self-supplied listings is a demo, not a market. The honest framing is: the mechanism has been proven (posting → pricing → reward calculation → task board), but demand has not been proven.

Acquisition channels remain unclear. Content brings human visitors, directories may bring agents, and offline promotion may bring both. The current situation is: the content engine works, the agent acquisition engine doesn't. This asymmetry is itself the entire strategic problem.

Self-supply and marketplace liquidity is a classic cold-start challenge for two-sided marketplaces. A marketplace with only sellers (or only buyers) is called a "one-sided market" and cannot generate transactions. The two current tasks on Hire Humans are both posted by the project itself, similar to Airbnb's early founders renting out their own apartments, or Reddit's early employees posting large amounts of content. The rationale is: first prove the mechanism works (task posting, pricing, reward calculation), then attract real suppliers. But the risk is that if self-supplied listings persist for a long time, external participants will see it as a "demo" rather than a "marketplace," reducing their willingness to join. The article's statement that "the mechanism has been proven, but demand has not been proven" is precisely an accurate summary of this state.

Next Steps: Unglamorous but Necessary Work

The plan for the next phase of the project is anything but glamorous:

1. Keep the task board online and honest. No fake listings, no inflated rewards, no claims like "100 tasks posted!" Every number a visitor sees must withstand verification by clicking through.
2. Increase agent distribution efforts. The goal is one agent customer. Directory submissions, agent framework communities, anywhere agents look for tools. This is the bottleneck; everything else is secondary.
3. Document the first real transaction like a moon landing. When an agent posts a task, a human completes it, and the reward settles, that evidence page matters more than any marketing copy. Until then, silence on revenue is the policy.

Advice for Other Builders: The Gap Between Software That Works and a Business That Runs Is Wider

After thirteen weeks, the author's most useful takeaway is: the gap between "software that works" and "a business that runs" is wider than the gap between "nothing" and "software that works." Building a task board takes discipline, but getting a stranger—or an agent—to actually use it requires something entirely different: distribution, trust, and time.

Another notable insight is that radical honesty eventually became a product feature, not just a principle. Whenever the author wanted to round metrics up or imply demand that doesn't exist, they would ask themselves: would this sentence withstand a side-by-side screenshot comparison? The ones that withstand it became the story itself. For a marketplace that depends on trust between strangers and software, this is not just ethics—it is product strategy.

The Hire Humans case reveals a broader industry signal: the narrative of AI agents as economic actors is moving from concept to infrastructure, but real market formation goes far beyond technical readiness. Agents need to discover tools, build trust, and complete transactions—and that path remains blurry. For those watching the AI agent economy and build-in-public methodology, the value of this project is not in its growth numbers but in the rare honesty it demonstrates: refusing to pretend demand exists before it has been validated.

Build in Public is a product development and marketing methodology in which founders publicly share the product iteration process, metrics, failures, and lessons in real time to attract early users, build trust, and gather feedback. The concept was popularized by Buffer co-founder Joel Gascoigne and others, and has become popular in recent years in independent developer communities such as Indie Hackers. The core value of building in public is the "honesty premium": when users see real data (including zero customers and zero revenue), they are actually more likely to trust you. Hire Humans' weekly report is a typical build-in-public practice—it does not hide failures but makes them part of the narrative. But building in public also carries risk: overexposing weaknesses may scare off potential customers or investors, so the boundary between transparency and trade secrets must be weighed.

One-line Summary

Hire Humans has been live with real paid tasks for two weeks: mechanism validated, but zero agent customers and demand still unproven.

Paragraph Summary

In its thirteenth-week public log, the Hire Humans project documents that its task board with real US dollar rewards posted by AI agents has been live for two weeks. Rewards are calculated from real rate logic (40% platform fee for free accounts, 15% for members), and the task board rendering and SEO content engine are working well. But the core metric—the number of agent customers—is zero. Both current online listings come from the project itself. The mechanism has been proven (posting → pricing → calculation → display), but real demand has not been validated. The author believes content acquisition works but agent acquisition channels remain unsolved. Next steps are to increase agent distribution, keep the task board honest, and document the first real transaction like a moon landing. The core insight is that the gap between "software that works" and "a business that runs" is larger than the gap from zero to working software, and radical honesty itself becomes product strategy.