The Agent Economy: A New Computational Paradigm
Examining the emerging agent technology stack and its potential to fundamentally transform computational systems
This post is an abbreviated version of a longer research piece on our research platform Kepler
The Agent Economy
The "agent economy" refers to a new computational paradigm driven by AI-powered autonomous agents that can independently perform tasks, make decisions, and engage in transactions on behalf of individuals and businesses.
This fundamental change in how computational systems interact will reshape how work is coordinated, decisions are made, and value is created.
A Brief History of Agents
The concept of autonomous agents traces back to the early days of computer science, from Turing's 1950 work on machine intelligence through McCarthy's 1959 "advice taker" and Hewitt's 1970s Actor Model. While the 1990s saw early implementations through expert systems and Microsoft's Clippy, and the 2010s brought voice assistants like Siri and Alexa, these systems remained limited by pre-programmed functions.
The real breakthrough came with transformer-based language models in the 2020s. Early experiments like AutoGPT and BabyAGI demonstrated the potential for LLMs as reasoning engines, leading to dedicated frameworks like LangGraph and AutoGen. By 2024, Anthropic's Model Context Protocol (MCP) moved beyond simple prompt chaining toward more sophisticated agent functionality.
The introduction of the Agent-to-Agent (A2A) standard earlier this year, backed by over 50 major technology companies, marks a crucial shift from isolated agents to an integrated ecosystem. This standardization enables agents to discover capabilities, coordinate activities, and work together across organizational boundaries—laying the foundation for a new agent economy.
Inflections
At Inflection, we focus on identifying profound shifts in technology and markets that enable novel behaviors at scale. Two key inflections are driving the emergence of the agent economy:
Cultural Inflection: Public sentiment around AI has shifted dramatically. From 2022-2024, most countries increasingly view AI as beneficial rather than detrimental, with some European nations seeing 10% increases in positive sentiment. This acceptance is evident in enterprise adoption: Johnson & Johnson employs agents for drug discovery optimization, while Moody's deploys interconnected agents for financial analysis (WSJ Article). We’re even using them at Inflection to automate our research and sourcing.
Organizations are evolving from viewing AI as tools to treating them as semi-autonomous team members with defined roles and responsibilities.
Technical Inflections: Four key technological advances are converging:
Large models have reached unprecedented capabilities, exceeding 1 trillion parameters
Small Language Models (SLMs) are achieving impressive performance with dramatically reduced computational needs
Specialized architectures like Large Action Models (LAMs) are optimizing for agent-specific tasks
New inference hardware from companies like Groq and Cerebras is enabling real-time agent interaction
These cultural and technical inflections, combined with standardized agent communication protocols, create the foundation for a new computational paradigm.
The Agent Technology Stack
Our full analysis on Kepler examines five critical components of the emerging agent technology stack. Here's a brief overview of the key developments we expect by 2035:
Intent Expression Systems: Today's programming tools remain rooted in explicit instruction-giving. By 2035, we envision systems that understand deep domain expertise and enable true intent-to-execution workflows across scientific research, manufacturing, and business processes. The distinction between programmer and domain expert will vanish as computation adapts to human thought patterns.
Infrastructure: Current solutions from companies like Temporal and Modal provide essential foundations but remain centralized. Future agent infrastructure will enable fluid operation across edge and cloud, with dynamic resource allocation and decentralized data architecture.
Memory Systems: While today's agents struggle with limited context windows, next-generation memory systems will incorporate neuromorphic architectures, in-memory computing, and persistent storage technologies. This will enable more human-like memory characteristics: contextual recall, intelligent forgetting, and dynamic connection-forming.
Compute: The future of agent compute moves beyond general-purpose AI accelerators toward specialized architectures: heterogeneous compute orchestration, custom inference ASICs, and dedicated agent communication processors.
Networking: Agent communication will evolve from today's verbose JSON protocols to optimized binary formats with hardware-accelerated authentication and standardized intent schemas, enabling efficient and secure multi-agent collaboration.
Investment Opportunities
From a deep-tech, pre-seed perspective, we're most excited about infrastructure optimization, specialized memory architectures, compute optimization (particularly for edge devices), secure information sharing protocols, and agent discovery systems.
The most compelling opportunities lie at the intersection of these domains, where advances in one area can unlock capabilities across the entire stack.
Conclusion
Just as the Cambrian explosion was enabled by foundational biological building blocks, the agent economy is being unlocked by advances across the entire technology stack.
This shift from isolated AI systems to interconnected agent networks represents more than an iterative improvement: it signals the emergence of truly autonomous computational systems that can collaborate and evolve without constant human intervention.
The agent technology stack emerging today represents a complete reimagining of our computing infrastructure, and now is the time to engage with the protocols and platforms that will define this next era of computation.
For our complete analysis of the agent technology stack and detailed investment thesis, read the full piece on Kepler.
Tremendous thanks to Lele Cao for notes and comments.
I have been really enjoying reading this piece! Nice work Alex and the team! I strongly recommend to read the full version linked in this article.