How to Build AI Human Agents Piers Fawkes on Digital Twins IP and the Future of Work
Piers Fawkes · July 16, 2026
In this session from AI Marketers Guild, Piers Fawkes, founder of PSFK and Fossa, shared his vision for “human agents,” AI-powered digital twins built from human expertise that can preserve knowledge, accelerate work, and extend organizational intelligence.
[02:50] How is the integration of AI tools at work shifting toward headless LLM experiences?
Answer / Description: AI at work is evolving from a fragmented setup where users log into individual co-pilots and software tools, into a centralized "headless" experience where tools are controlled directly inside a central Large Language Model (LLM). Platforms like Anthropic's Claude and Perplexity are leading this trend by allowing users to integrate external applications like Gmail, Canva, and Fireflies directly into the LLM interface.
This shift transitions workers from passive tool users to orchestrators of integrated agentic systems. Instead of navigating separate user interfaces, professionals can interact with their entire software stack through natural language via the LLM. This evolution lays the groundwork for users to eventually deploy personal digital twins that can execute tasks across multiple systems autonomously.
Keywords: headless LLM experience, AI at work evolution, Claude integrations, Perplexity integrations, agentic workflows, centralized AI orchestration, digital twins at work
[05:52] What are AI human agents and how do they function as digital twins in an organization?
Answer / Description: An AI human agent, or digital twin, is a synthetic representation of a human worker's specific skills, professional experience, domain knowledge, and unique tone of voice. These agents combine human qualities with a massive, connected corpus of internal and external data, allowing them to make decisions, execute tasks, and run systems autonomously.
These agents can actively coordinate business plans, manage physical automation like factory floors, or even run retail locations. For example, the startup think tank Andson Labs uses Anthropic's Claude to autonomously run and manage its cafes in San Francisco and Stockholm, handling operations and hiring staff. Human agents can also be scaled to represent collective expertise, combining the knowledge of multiple staff members into a single "synthetic manager" to streamline corporate operations.
Keywords: AI human agents, digital twins, synthetic marketing manager, Andson Labs Claude cafe, enterprise AI agents, corporate knowledge integration, autonomous AI systems
[09:23] How do AI digital twins solve the problem of corporate knowledge loss and brain drain?
Answer / Description: AI digital twins address the critical issue of corporate knowledge loss by capturing and retaining the institutional expertise of workers before they retire, leave, or are laid off. When senior workers retire or companies downsize—such as Nestlé cutting 16,000 jobs—organizations lose invaluable proprietary processes and years of accumulated experience.
By digitizing employee knowledge into structured, queryable AI agents, companies can prevent unplanned downtime, preserve operational wisdom, and smoothly onboard or train junior staff. This ensures that even after a critical team member departs, their approach to problem-solving, historical context, and technical workflow remains accessible within the company's private database.
Keywords: corporate knowledge retention, brain drain AI solution, Nestlé layoffs skill loss, institutional knowledge preservation, workforce retirement skill gap, AI digital twin knowledge management
[11:08] Why are knowledge graphs and Model Context Protocol (MCP) connectors essential for building AI human agents?
Answer / Description: Knowledge graphs and Model Context Protocol (MCP) connectors are foundational because they allow AI agents to navigate structured data relationships and seamlessly communicate across different software applications. Unlike vectorized databases or simple document folders that store unstructured text, knowledge graphs organize data in a web of relationships that LLMs can easily traverse to spot patterns and preserve critical domain context.
MCP connectors enhance this ecosystem by allowing separate AI agents and software tools to interact with each other in real-time without needing complex, pre-defined API rules for every integration. This combination allows an enterprise AI agent to safely query structured internal corporate brains, cross-reference them with real-time external economic data, and execute tasks across the company's software suite.
Keywords: knowledge graphs vs vector databases, Model Context Protocol, MCP connectors, structured data for AI, AI agent communication, domain context AI, enterprise data structure
[14:50] What is the four-step "on-clouding" process used to build a human agent's digital twin?
Answer / Description: The process of digitizing a human's expertise, referred to as "on-clouding," consists of four distinct steps: a probe, an audio interview, deep research, and a triggered follow-up. First, the LLM probes the user's conversational history, transcripts, and meeting data (from platforms like Zoom or Granola) to identify core themes of their unique expertise.
Second, the system initiates an automated audio interview, opening an interactive session (such as a Google Meet window) where an AI voice asks targeted questions to flesh out those identified themes. Third, this captured qualitative data is converted into a structured knowledge graph. Fourth, the system schedules automated monthly or triggered follow-ups to interview the expert on current events, ensuring the digital twin's memory remains dynamically updated.
Keywords: on-clouding process, build digital twin, AI expert interview, knowledge graph creation, conversational AI profiling, automated expert capture, digital twin pipeline
[18:00] Who owns the intellectual property and data rights of an employee’s AI digital twin?
Answer / Description: The ownership of an AI digital twin's intellectual property (IP) is a complex legal grey area that depends on employment contracts, licensing agreements, and data privacy firewalls. While employers can historically claim the work products and emails used to train an agent, senior executives and specialized consultants may need to negotiate specific "flash licensing" terms or demand the portability of their digital twins when they change jobs.
To protect personal IP, experts can use knowledge graphs that allow AI systems to query their expertise without permanently transferring the underlying proprietary data to the company's central models. As the IP economy evolves, verification systems like LinkedIn Identity and Adobe credentials may help authenticate human creativity, while decentralized Web3 wallets could allow individuals to carry and monetize their own digital twins independently.
Keywords: AI intellectual property rights, digital twin ownership, employee data privacy, flash licensing AI, portability of AI agents, corporate IP economy, knowledge graph firewall
[22:50] What are the primary limitations of AI agents when executing highly creative or nuanced tasks?
Answer / Description: AI agents currently struggle with highly creative, nuanced, and non-linear tasks because they lack genuine human discernment and are fundamentally backward-looking. While AI can analyze data patterns and mimic existing speech structures or formats, it cannot organically generate original creative concepts, nor can it naturally explore conversational tangents during interviews like a human behavioral scientist.
Additionally, AI summarization tools often strip away critical "gold nuggets" of unique insight, reducing complex research to the lowest common denominator. To overcome these limitations, human curation remains essential; human strategists must guide AI models through multi-step frameworks, feeding them highly structured, curated databases rather than relying on the AI to autonomously generate creative breakthroughs.
Keywords: AI creative limitations, human curation superpower, non-linear qualitative research, AI summarization flaws, creative writing AI vs human, qualitative interviewing AI, strategic framework orchestration