From agentic Fiverrs to enterprise-grade verticals, builders are racing to define how AI agents are created, deployed, discovered, and monetized.
Online shopping. Stock trading. Vacation planning. If you listen to the most bullish projections about the scope and speed of AI-agent adoption in the coming years, use cases in commerce, public markets, and flight bookings will be a mere subset of those automated through currently nascent forms of artificial intelligence.
But reading such an endstate necessitates building an ecosystem solving for the design, distribution, and monetization of agents. The most prominent builders of foundational models have started to put forth their respective versions of what such an ecosystem may look like. Perhaps unsurprisingly, they tend to be walled gardens. OpenAI’s prototype has obliquely taken the form of ChatGPT’s agentic mode, which can surf the web, run code, make purchases, and more. Anthropic, meanwhile, has rolled out Claude Code’s agent builder, enabling businesses to develop and deploy the agents they see as most pertinent to their end users.
But then what?
Dharmesh Shah is the founder and CTO of HubSpot, and the developer of agent.ai, an online platform for building, publishing, and using AI agents. (The agent.ai website cheekily refers to itself as a “professional network.”) In correspondence with Future Nexus, Shah said agentic marketplaces will eventually take two forms. One path will include a system mimicking the freelance platform Fiverr; under this system, users would find an agent to accomplish a singular task. The other, Shah said, will involve people going to marketplaces to find or hire agents to form a “hybrid team,” in which “agents would be designed to work together and be able to collectively take on higher order goals.”
Shah and agent.ai are building toward that second option. A low-code tool for building custom agents, with the option of sharing them with other users of the website, helps create a network of people and agents simultaneously.
Other entrepreneurs in the space have opted for a different approach to the question of supply and demand for AI agents. Kamil Stanuch and Lukasz Wróbel are behind Job for Agent, which allows users to post work requests, while others submit AI agents they think are up for the job.
“Currently, we operate more like a connection point similar to Upwork or Fiverr but for AI tasks, where the human developers who build and deploy the agents are the ones getting contracted and paid,” Stanuch told Future Nexus. “We facilitate the connection between companies needing automation and the builders who can deliver it.”
Job for Agent started as a “quick experiment” inspired by a Firecrawl job posting for “AI agents only.” The Poland-based team saw an opportunity to build an ecosystem for tasks that could be outsourced to non-human agents.
“We’re seeing a fascinating gap in the market: skilled AI builders seeking applications for their agents on one side, and companies unsure how to leverage the burgeoning power of agentic AI on the other,” Stanuch continued. “Our platform acts as a bridge between the growing supply of AI development talent and the increasing corporate demand for efficiency gains through AI solutions.”
Job for Agent has more than 1,000 vetted AI developers and engineers on its platform, and has processed hundreds of AI agent-focused job opportunities. Stanuch said a broad range of tasks are being posted by companies, which suggests “companies are seriously exploring which tasks can be effectively automated or augmented by AI.” Using Fiverr-like platforms like Job for Agent effectively outsources the R&D of determining where AI agents may or may not be useful by employing developers for discrete tasks and deployments — especially useful for smaller players lacking substantive AI-development budgets.
“What we’re seeing is that for 90% of tasks posted on our platform, you don’t actually need a full-fledged AI agent – a well-designed set of automations will do the trick,” Stanuch said. He added that Job for Agent has yet to determine how it’ll monetize itself, claiming that the “real long-term value” may come from verifying, orchestrating, and managing agents — not just finding an agent.
Yet in the eyes of Siddhant Masson, the founder and CEO of research and diligence platform Wokelo AI, agentic marketplaces may only be useful for a subset of use cases and end users. Enterprises are deeply concerned about data security and privacy, and do not want their data to be used for model training or other purposes beyond their control. Larger players would rather work with tried-and-true platforms or vertical AI companies, with too much on the line to tinker with freelance solutions.
What’s more, vertical AI solutions can keep up with the frequent updates to foundational models. “The underlying models — OpenAI, Anthropic, Gemini — release model updates every few weeks. What you can set up as a small shop internally would be junk a few weeks down the line. So how do you maintain it, do you have the skill set, and are you using best practices? For smaller firms, it doesn’t make sense,” Masson said.
Walled gardens. Agentic Fiverrs. Vertical solutions. Just as today’s AI landscape trifurcates return and scope depending on firm size and R&D budget, agentic AI networks will likewise take different forms according to the bandwidth, budget, and concerns of the end user.
