The Agentic Economy Needs a Market for Work

AI agents frequently operate within confined environments, capable of executing specific tasks like writing code, conducting research, or generating documents. However, these agents often struggle when faced with challenges that extend beyond their toolset. As agents begin to receive modest budgets and authorization to perform actions, the landscape of work will evolve.

An agent that cannot resolve an issue independently will have the capability to hire another agent, compensate a human, or offer a reward to an individual who can fulfill the task. This newfound ability will facilitate the formation of an agentic economy - a marketplace where individuals and agents collaborate on tasks, results, and payments.

The concept can be illustrated with a software agent tasked with identifying a bug in a library that its owner does not maintain. In today's systems, the agent would open an issue and wait for assistance. However, a more advanced system could enable the agent to create a task, complete with a failing test, clear acceptance criteria, and a monetary reward.

Another agent with the necessary expertise would then address the task, submit a patch, run the required checks, and receive payment upon successful completion. Similar scenarios can be extrapolated to other domains. A research agent may require a structured dataset, a small business may need product images resized and tagged, and a community may offer incentives for translating public information or verifying the functionality of a set of links.

Although these tasks vary in nature, they share a common attribute: the desired outcome can be articulated clearly and validated through digital evidence. Agents do not need to assume the role of employees for this arrangement to thrive. Instead, they require access to well-defined tasks accompanied by a tangible reward, sufficient information to assess the task's validity, and a dependable mechanism for disbursing payment.

Establishing trust within these transactions is crucial. Interacting with an unknown party presents a familiar predicament. The buyer desires assurance that the work will meet expectations before disbursement, while the worker seeks certainty regarding payment prior to delivering the results. Traditional human marketplaces alleviate this dilemma through escrow services, reputation systems, customer support, and arbitration mechanisms.

While these systems prove beneficial, they often involve slow decision-making processes, private deliberations, and complexities that hinder software integration. To accommodate agents effectively, the marketplace must make the agreement explicit from the outset. The task should clearly outline the expected result, the verification process, the allocated funds, and the consequences when a submission is accepted or rejected.

The evidence submitted should be easily accessible for inspection. Payment should be released only after the result has been validated, rather than relying on a post-delivery promise. Smart contracts present a fitting solution, as they can securely hold funds and enforce predefined settlement rules. Although smart contracts cannot guarantee fairness in every judgment or determine the quality of every task, their primary benefit lies in ensuring that financial transactions and stipulated rules remain immutable throughout the task.

The initial reliable markets will predominantly focus on tasks with deterministic verification processes. Code can readily pass a test suite, data can conform to a specific schema, and a file can be validated against a required hash. Likewise, an on-chain action can be verified through an associated event. However, tasks necessitating human judgment, such as writing, design, strategy, and research, often require human evaluation, ordered decision-making, and a structured appeals process.

Treating these judgments as straightforward would foster misplaced confidence. Instead, the market should acknowledge the inherent subjectivity and maintain transparency. The emergence of such markets will fundamentally alter how work is discovered. While work is currently sourced through traditional channels like job boards, client relationships, personal networks, or platforms catering to human profiles, these methods are less effective when tasks are small, urgent, or easily verifiable.

An open task market presents work in a public format that software can comprehend. Agents can search for tasks based on skill sets, rewards, deadlines, verification methods, or entry costs. A new participant can gain a competitive edge by producing the desired result rather than first establishing a professional profile or audience.

People can post a task without anticipating whether a person or an agent will carry out the work. This approach may foster a more meritocratic and efficient marketplace, albeit contingent upon maintaining visibility into its limitations. It is important to note that one wallet does not equate to one independent person. The emphasis on speed alone may not always yield the optimal outcome.

A verifier may contain bugs or exhibit biases towards certain types of solutions. Transparent market structures facilitate the identification of these issues rather than obscuring them. Open rules empower participants to scrutinize tradeoffs and develop enhanced alternatives. Discovery holds equal significance as settlement. A funded task that remains undiscovered by capable agents is inconsequential to the market's functionality.

Public feeds, APIs, repositories, and standardized labeling can transform each task into a discoverable entity for agents without relying on human intervention. Over time, agents could amass a track record of completed outcomes and leverage their earnings to fund subsequent tasks. The envisioned realization of this future is relatively straightforward.

An individual or agent can define a task and allocate the necessary funds. Eligible participants will discover the task through the platforms they currently utilize. They can submit work along with digital evidence, adhering to pre-established rules. A designated verifier will assess the result. The payment status will only be considered final following successful settlement confirmation.

This iterative process can serve as the foundation for more extensive coordination efforts. An agent can decompose a goal into smaller tasks, allocate funds to specialized agents, amalgamate their contributions, and deliver a cohesive outcome to its owner. Alternatively, a person can establish the objective, budget, and boundaries without micromanaging every intermediate step.

Contributors from any corner of the globe can contribute when the task and evidence are digital in nature. However, challenges remain, including failures, disputes, flawed specifications, and attempts to manipulate the system. Rather than obscuring these issues behind a confident interface, the market must address them with conservative permissions, clear resolution mechanisms, transparent verification processes, and candid communication regarding the actual payments made.

Open-source infrastructure assumes paramount importance at this juncture. If agents are to leverage a protocol for task discovery and financial transactions, developers must prioritize the creation of robust and dependable infrastructure.