This is the daily reality for engineers across HEMA's 750+ stores: looking up how to request an API, or figuring out as a new hire who owns which system, requires hopping between wikis, service catalogs, and IT portals—what they themselves call "portal-hopping." The problem isn't a lack of knowledge; HEMA already had a structured service catalog. The problem is that the "how we actually get things done" part is scattered everywhere, with nobody organizing it.
What this is
As described in a joint AWS-HEMA blog post, the solution: they built HAL, an internal assistant on Amazon Bedrock AgentCore (AWS's enterprise AI Agent hosting platform), consolidated knowledge into one layer, and pushed it via MCP (Model Context Protocol—an open standard that lets LLMs directly call external tools and data sources) into the chat tools and IDEs (including Kiro and Claude) that employees already use. Security is anchored to Microsoft Entra ID; employees don't hold AWS credentials, and permissions inherit from the company's existing identity system.
Industry view
This marks the first time a century-old retailer has deployed MCP in real production and publicly detailed the architecture, since Anthropic open-sourced the protocol in late 2024. Supporters will argue: enterprise AI adoption has long been bottlenecked by "scattered knowledge," and MCP provides a relatively standardized tool-access layer, letting chat tools, IDEs, and Agents pull data directly—avoiding every team reinventing the wheel. AWS bundling Bedrock AgentCore with MCP is a clear bet on this path. Skeptics counter: HEMA itself admits HAL is currently just a read-only knowledge layer—the action layer is next, meaning you can only ask, not do. The MCP ecosystem is still early; the operational cost, security boundaries, and liability for in-house MCP servers remain unanswered. We think the deeper signal is this: the piece repeatedly emphasizes "the knowledge was hard to reach" rather than "the knowledge didn't exist," telling us the real bottleneck is the organization's will to document. AI merely shines a brighter light on an old problem.
Impact on regular people
- For enterprise IT: the internal knowledge entry point is shifting from "web search" to "ask in chat"—provided the company is willing to spend money and people to organize scattered documents into structured knowledge.
- For working professionals: new hires will no longer depend on "asking the person next to you," but that also means the implicit value of those human connections is eroding.
- For consumer markets: consumers won't notice much in the short term; but once HAL upgrades from read-only to actionable, store operations, customer service, and supply chain scheduling could all be rewritten.