The Future of Agentic Commerce. Patented in 2016.
How a patent for autonomous AI negotiation agents filed before GPT existed anticipated the agentic AI revolution—and what it means for AI-powered innovation today.
By George Krasadakis — Innovation Strategist, AI Advisor, Author of Innovation Mode 2.0
Highlights:
• US Patent Application 20170287038A1, filed March 2016, described autonomous AI agents negotiating on behalf of humans—an architecture now central to the agentic AI movement
• The enabling technologies (LLMs, function calling, RAG) did not exist in 2016; the vision preceded the infrastructure by nearly a decade
• The same principles of autonomous AI agents now power innovation discovery platforms and are reshaping how organizations identify and pursue opportunities
• Innovation timing requires vision, enabling technology, infrastructure, and market readiness to align—a framework explored in depth in Innovation Mode 2.0
In the past year, agentic AI has become the dominant theme in artificial intelligence. Autonomous AI agents that browse the web, execute code, negotiate on behalf of users, and complete multi-step tasks without human intervention are attracting billions in venture capital. Every major AI lab—OpenAI, Anthropic, Google DeepMind—is racing to build agent frameworks. The consensus is clear: AI agents represent the next frontier of artificial intelligence.
I find myself in an unusual position watching this unfold. In March 2016, I filed a patent describing exactly this architecture—autonomous AI negotiation agents that discover counterparties, exchange offers through multi-stage protocols, operate within user-defined constraints, and execute transactions without human involvement.
I filed the patent (US20170287038A1, "Artificial Intelligence Negotiation Agent") while I was at Microsoft. The core concept described a framework where buyer AI agents and seller AI agents discover each other, negotiate within "elasticity thresholds," coordinate strategies across agent networks, and converge on optimal terms. The system integrated real-time market data, social signals, pricing trends, and inventory information to inform negotiation strategies.
It was a blueprint for agentic commerce—designed nine years before the term entered mainstream discourse.
Although the patent was not commercially implemented at the time—the enabling technologies simply did not exist—the core architectural principles have proven remarkably durable. Today, I apply these same concepts to AI-powered innovation through Ainna.ai and the methodologies I explore in Innovation Mode 2.0.
This is a story about innovation timing, technology readiness, and how ideas evolve across technology generations.
What the AI Negotiation Agent Patent Described
The core premise was straightforward: humans should not have to manually search, compare, and negotiate every purchase. Instead, users would define their requirements—product specifications, price ranges, delivery constraints, quality thresholds—along with their flexibility on each parameter. An autonomous AI agent would then find matching sellers and negotiate the best possible deal.
The system worked bidirectionally. Sellers could deploy their own AI agents with sales objectives, profit margins, inventory constraints, and pricing elasticity. These seller agents would discover relevant buyer agents and engage in automated negotiations.
Several architectural elements from the patent stand out in retrospect:
Elasticity-based constraint encoding. Users did not specify rigid requirements. They defined acceptable ranges, weight factors for different parameters, and "blocker" constraints that could not be violated. This approach to encoding human preferences with flexibility bounds is remarkably similar to how we now structure prompts and constraints for large language model agents.
Multi-agent coordination with shared objectives. The patent described multiple agents working toward shared goals, dynamically reallocating resources based on individual outcomes. This is multi-agent reinforcement learning territory that AI researchers are actively exploring today.
Real-time market intelligence integration. Agents would pull data from pricing databases, social media sentiment analysis, supply-demand signals, and industry news to inform their strategies. This anticipates today's retrieval-augmented generation (RAG) architectures and tool-using AI agents.
Agent-to-agent negotiation protocols. Buyer and seller agents communicated through structured multi-stage negotiation sessions. This is the agent-to-agent interoperability challenge that the AI industry is now working to standardize through initiatives like Anthropic's Model Context Protocol (MCP).
Autonomous execution with human oversight. With appropriate user authorization, agents could complete transactions independently, while maintaining transparency about their actions and decisions—a balance that remains central to responsible AI agent design.
Why the Technology Was Not Ready in 2016
In 2016, the AI infrastructure required to implement this architecture reliably did not exist.
Natural language understanding was primitive. Users could not describe what they wanted in conversational terms. They would have needed structured input forms with dropdown menus and checkboxes. The friction of configuration would have killed consumer adoption.
Reasoning capabilities were limited. The AI agents I envisioned needed to make complex trade-offs, understand negotiation context, interpret counterparty signals, and adapt strategies dynamically. The machine learning of 2016 could not handle this level of nuanced reasoning.
Tool use and function calling did not exist. Modern AI agents can call APIs, execute code, browse the web, and interact with external systems through standardized interfaces. In 2016, this required custom integrations for each connection.
User trust was not established. Consumers were not ready to delegate decisions to autonomous software. Even today, building this trust requires demonstrating competent, transparent AI behavior over time.
The transformer architecture that would enable large language models was published in 2017, a year after I filed the patent (Vaswani et al., 2017). GPT-1 arrived in 2018. GPT-3 in 2020. ChatGPT in late 2022. The capabilities we now consider standard—natural language interfaces, complex reasoning, tool use, multi-turn planning—did not exist when the patent was filed.
Why Agentic AI Is Now Achievable
The technology gap has closed. Large language models can understand nuanced requirements expressed in natural language. They can reason about trade-offs, maintain context across extended interactions, and generate appropriate actions. Function calling and tool use allow AI systems to interact with external APIs, databases, and services reliably.
The infrastructure ecosystem has matured. Vector databases enable semantic search. Embedding models represent preferences in computable form. Agent frameworks like LangChain, AutoGen, and CrewAI provide scaffolding for multi-step task execution. The building blocks I would have had to create from scratch in 2016 are now commoditized.
Agent-to-agent communication protocols are emerging. Model Context Protocol (MCP), function calling specifications, and agent communication frameworks are establishing the foundation for AI agents to interact with each other and with external systems—precisely what the patent envisioned for buyer-seller agent negotiations.
From Agentic Commerce to AI-Powered Innovation
The architectural principles I explored in the 2016 patent—autonomous agents, intelligent matching, multi-source intelligence integration, elasticity-based preferences—extend far beyond commerce. They apply wherever humans need to discover, evaluate, and act on complex opportunities.
Innovation is one such domain. Organizations struggle to systematically identify opportunities across fragmented information landscapes—emerging technologies, market shifts, regulatory changes, competitive moves, partnership possibilities. The cognitive load of continuous opportunity scanning exceeds human capacity. This is precisely where AI agents can transform organizational capability.
This insight led me to build Ainna.ai—the AI-powered opportunity discovery platform that applies agentic AI principles to innovation. Ainna continuously monitors diverse sources (patents, research publications, market intelligence, news, funding data), identifies relevant opportunities based on user-defined innovation priorities, and surfaces actionable insights. The platform embodies the same core concepts from the patent: autonomous agents operating on behalf of users, integrating real-time intelligence, matching opportunities to explicitly defined parameters and preferences.
Where the 2016 patent imagined AI agents finding the best deal on a product, Ainna's agents find the best opportunities for innovation—whether that's an emerging technology to watch, a potential acquisition target, a research partnership, or a market gap to exploit. The architectural pattern is the same; the application domain has evolved.
AI Agents and the Future of Corporate Innovation
In my forthcoming book Innovation Mode 2.0: Designing Innovative Companies in the Era of Artificial Intelligence (Springer, 2026), I explore how AI is fundamentally reshaping corporate innovation—not just as a tool, but as a transformative force that changes how organizations discover, evaluate, and pursue opportunities.
AI agents represent a critical capability in this transformation. Consider the traditional innovation funnel: opportunity identification, screening, concept development, business case creation, and execution. Each stage has historically been constrained by human attention and cognitive bandwidth. AI agents can operate continuously across each stage:
Discovery agents monitor technology landscapes, patent filings, academic research, startup funding, and market signals to surface relevant opportunities that human scanners would miss.
Evaluation agents assess opportunities against organizational strategy, capabilities, and risk parameters—applying the same "elasticity threshold" logic from the patent to determine strategic fit.
Connection agents identify potential partners, acquisition targets, or collaboration opportunities by matching organizational needs with external capabilities.
Synthesis agents integrate insights from multiple sources to generate comprehensive opportunity assessments, business cases, and recommended actions.
This is not speculative. These are capabilities that can be built today with existing AI infrastructure. The question is no longer whether AI agents will transform innovation management, but how quickly organizations will adopt them and who will build the platforms that enable this transformation.
What This Teaches Us About Innovation Timing
This experience reinforced a principle I explore extensively in Innovation Mode 2.0: being right too early is functionally equivalent to being wrong.
The patent described a technically sound architecture for a real problem. But the enabling technologies were not ready, the infrastructure did not exist, and user behavior had not evolved to accept AI delegation. No amount of engineering could have overcome these gaps in 2016.
Successful innovation requires four elements to align:
1. Vision — seeing the opportunity and understanding the problem deeply enough to design a solution
2. Enabling technology — having access to the technical capabilities required to build the solution
3. Infrastructure — ecosystem support, platforms, and complementary services that the solution depends on
4. Market readiness — user willingness to adopt and trust the new approach
Missing any one of these creates a timing gap that cannot be bridged through effort alone. The 2016 patent had vision but lacked enabling technology and market readiness. By 2025, all four elements are aligning for agentic AI.
What should innovators do when they see the future clearly but cannot yet build it? Document your thinking rigorously. Protect intellectual property where appropriate. Build the components you can with available technology. Stay engaged with the problem space. And be ready when the enabling technologies arrive.
The Road Ahead
Agentic AI is no longer speculative. Within the next few years, we will see autonomous AI agents negotiating prices, managing subscriptions, and executing purchases on behalf of consumers. We will see AI agents transforming how organizations discover and pursue innovation opportunities. We will see multi-agent systems coordinating complex business processes that currently require extensive human orchestration.
Challenges remain. Trust and accountability frameworks need development. Agent-to-agent protocols require standardization. Privacy concerns around AI agents accessing sensitive information need addressing. The competitive dynamics of AI-versus-AI negotiation are largely unexplored.
But these are engineering and policy challenges, not fundamental barriers. The core architecture I described nine years ago is now achievable. The timing, finally, is right.
And that's the ultimate lesson: good ideas don't expire. They wait for their moment.
———
About the Author
George Krasadakis is an innovation strategist and AI advisor with over 20 patents in artificial intelligence and machine learning. He has held senior technology and innovation leadership roles at Microsoft, Accenture, and GSK. George is the author of Innovation Mode 2.0: Designing Innovative Companies in the Era of Artificial Intelligence (Springer, 2026), founder of Ainna.ai (the AI-powered opportunity discovery platform for innovators), and founder of The Innovation Mode consulting practice. He is based in Dublin, Ireland.
References
Krasadakis, G. (2016). Artificial Intelligence Negotiation Agent. US Patent Application 20170287038A1. Filed March 31, 2016. Available at: patents.google.com/patent/US20170287038A1
Krasadakis, G. (2026). Innovation Mode 2.0: Designing Innovative Companies in the Era of Artificial Intelligence. Springer.
Vaswani, A., Shazeer, N., Parmar, N., et al. (2017). Attention Is All You Need. Advances in Neural Information Processing Systems, 30.
Anthropic. (2024). Model Context Protocol (MCP). Available at: anthropic.com
OpenAI. (2023). GPT-4 Technical Report. arXiv preprint arXiv:2303.08774.
Wooldridge, M. (2009). An Introduction to MultiAgent Systems. John Wiley & Sons.
Topics: Agentic AI, Autonomous AI Agents, AI Negotiation, Agentic Commerce, Multi-Agent Systems, AI Patents, Innovation Timing, Large Language Models, AI Strategy, AI-Powered Innovation, Innovation Management, Corporate Innovation, Ainna.ai, Innovation Mode
