Foundry Works is a significant player in the agent ecosystem because they are one of the few firms bridging the gap between infrastructure (MCP) and commercial delivery (Agency 2.0). They are deeply active in the Model Context Protocol stack, contributing open-source tools that help other developers build more reliable agent-to-tool connections. Their work on foundry-mcp and claude-foundry positions them as a technical leader within the Anthropic/Claude sub-ecosystem.
For those building in the agent space, Foundry Works provides a blueprint for how 'agent teams' can be sold as a managed service. They move the conversation away from general-purpose assistants and toward specific, multi-agent workflows that handle complex tasks like lead routing and document processing. Their existence as an 'AI-native agency' validates the market for agentic labor as a replacement for traditional service-based business models.
Foundry Works is a London-based AI-native agency that represents a shift in how professional services are delivered. Founded by Jason Sibley, who brought a background in sports marketing and Web3, the company is built on the premise that traditional marketing and sales functions can be performed more efficiently by specialized AI agent teams than by human-heavy departments. They do not just build tools for clients; they run the 'marketing machine' as a managed service, deploying agents that write copy, manage social media, and qualify sales leads.
The company’s internal structure mirrors this philosophy. Their public-facing roster includes agents like 'Cleo' for marketing automation and 'Elliott' for trading signals, alongside industry-specific deployments for clients like Lenovo and Brother UK. Notably, they treat these agents as active members of a workforce, categorizing them as 'Live Now,' 'In Training,' or 'Now Recruiting.' This nomenclature reflects a mature understanding of agent reliability: agents are not static scripts but evolving systems that require training and oversight before they are ready for production environments.
Unlike many agencies that operate as simple wrappers for large language models, Foundry Works maintains a significant technical footprint. Their work is heavily oriented toward the Anthropic ecosystem and the Model Context Protocol (MCP). Through their GitHub organization, they maintain foundry-mcp, a project dedicated to building standardized connectors that allow agents to interact with external data and tools more reliably. This focus on MCP is a strategic bet on interoperability, allowing their agents to move beyond closed chat interfaces and into the actual systems where work happens.
They also develop claude-foundry, a framework for AI-assisted development that uses spec-driven development (SDD). This methodology emphasizes creating detailed specifications before implementation, a necessary discipline for building complex agentic workflows that can actually handle enterprise requirements. Their infrastructure also includes ephemeral Docker workspaces designed to isolate coding agents, ensuring that automated workers can execute tasks without compromising the host system’s security or credentials.
Foundry Works is transparent about the limitations of the technology. They publicly retired a voice cloning service for calls, noting that it landed in the 'uncanny valley' and that human interaction still won in that specific domain. This level of honesty is rare in the current AI market and suggests a focus on utility over hype. Their typical engagement is described as an 80% solution today rather than a perfect one next month, a philosophy that prioritizes shipping functional agents into the field quickly.
By focusing on sectors like sports and enterprise technology, they have carved out a niche where scale is high but processes are often repetitive enough for automation. For global brands like Brother, they deploy multiple Partner Account Manager (PAM) agents, a deployment that demonstrates the ability to scale specialized roles that would otherwise require significant recruitment and overhead. They are positioned as a bridge between the high-level capabilities of models like Claude and the messy, specific needs of business operations.
Customized AI agent teams for marketing, sales, and operations.
Foundry Works is hiring.