Strategic Self-Improvement for Competitive Agents in AI Labour Markets

TL;DR

The paper addresses AI agent labour markets under informational frictions such as adverse selection, moral hazard, and reputation. It introduces Strategic Self-Improving Agents (SSA) with metacognition, competitive awareness, and long-horizon planning, demonstrated through a tractable gig-economy simulation (AI Work) modeled as a Competitive Skill-Based Stochastic Game. The study uncovers macroeconomic patterns and AI-specific dynamics, showing that explicit strategic prompting improves agent performance and market utility, while design choices like bidding mechanisms and performance-based pay materially shape outcomes. It concludes with implications for platform design and policy, highlighting opportunities and limitations at the intersection of ML and economics.

Abstract

As artificial intelligence (AI) agents are deployed across economic domains, understanding their strategic behavior and market-level impact becomes critical. This paper puts forward a groundbreaking new framework that is the first to capture the real-world economic forces that shape agentic labor markets: adverse selection, moral hazard, and reputation dynamics. Our framework encapsulates three core capabilities that successful LLM-agents will need: \textbf{metacognition} (accurate self-assessment of skills), \textbf{competitive awareness} (modeling rivals and market dynamics), and \textbf{long-horizon strategic planning}. We illustrate our framework through a tractable simulated gig economy where agentic Large Language Models (LLMs) compete for jobs, develop skills, and adapt their strategies under competitive pressure. Our simulations illustrate how LLM agents explicitly prompted with reasoning capabilities learn to strategically self-improve and demonstrate superior adaptability to changing market conditions. At the market level, our simulations reproduce classic macroeconomic phenomena found in human labor markets, while controlled experiments reveal potential AI-driven economic trends, such as rapid monopolization and systemic price deflation. This work provides a foundation to further explore the economic properties of AI-driven labour markets, and a conceptual framework to study the strategic reasoning capabilities in agents competing in the emerging economy.

Figures (7)

• Figure 1: Conceptual Overview To study the dynamics and impact of AI agent to economy, we created a simulation that contains the core features of a Labour Market (Right), and examined the capabilities that allow agents to succeed in this competitive economic setting. We identified three domains of reasoning patterns that inform successful agents, which we call "Strategic Self-Improving Agent". These agents operate within an economy shaped by Macroeconomic Factors, Client preferences, and Job Platform mechanics. This paper investigates how these capabilities enable agents to adapt their internal state (e.g., Skill Level, Reputation) and actions to succeed under competitive economic conditions.
• Figure 2: To study the dynamics of AI agents within a labour market, we created a simulated gig platform AI Work, where AI-agents act according to policy $\pi$, and bid for work over real jobs based on a set of latent skills $\theta$ (A). Our simulated market selects bids from agents based on their public rating and price (B). Each turn, agents can choose to bid for work, or train in one of its skills (C). Similar to a real labour market, the only information agents are exposed to is which agents winning which jobs, and their public facing reputation. From our simulation with LLM-based agents, we describe three core capabilities that make agents competitive in this market: 1. Metacognition, where the agent is aware of its own latent skill vector (red), 2. Competitive Awareness, where the agent is aware of its competitiors and market dynamcis (blue), 3. Long-horizon planning, where the agent formulates a coherent plan for its policy over multiple time steps (green). With explicit prompting within the reasoning process, these Strategic Self-Improving Agents demonstrate superior performance in our simulation against other LLM agents.
• Figure 3: Examples of macroeconomic activity from baseline simulations
• Figure 4: The structure of market incentives dictates agent strategy and overall market utility. A: In open tenders (blue), agents underbid each other, leading to lower normalized prices compared to sealed tenders (orange). B: This price-focused competition disincentivizes self-improvement, resulting in less agent training. C: When agents are rewarded with performance-based incentives (green), they progressively increase their investment in training compared to agents reciving fixed rewards (red). D: This increased training directly translates to higher market utility over time.
• Figure 6: Strategic Self-Improving Agents dynamically adapt to market conditions and competitive pressures. A: Price-sensitive clients (dark purple) promote low-bid strategies, while reputation-sensitive markets (orange) drive skill investment. B: As market demand for specific skills shifts (blue vertical line), agents adjust their bidding priority to the new skill (orange) but retreat to their original specialization (purple) when outcompeted. C: Training investments mirror bidding patterns, with agents rapidly shifting focus to newly valued skills before competitive retreat occurs.