I got put into a meeting with a networking team member and three people from AWS, talking through our AI use and where we want to take it. Nobody assigned me to lead it. About ten minutes into the technical parts, the networking guy turned to me and asked me to take it from there.
Research, fast
I hadn't worked with AgentCore, Bedrock, or Lambda, so the days before the meeting went into learning what each service does and how it maps onto the MCP work I'd built. The problem they solve: my MCP servers have no LLM API endpoint of their own. AgentCore provides one, inside our existing AWS budget rather than a new line item. It's also the missing piece for the PDF-import problem: running a PDF straight through a foundation model skips the extraction ceiling every coded approach hit, and AgentCore is what would let us run that.
I looked at Lambda for hosting the servers, weighing token and cost efficiency against how things run now. Lambda is the answer for keeping the servers running and accessible to the team instead of living on my laptop.
Explaining, not demoing
No live demo in that room, just a natural-language walkthrough of the three MCP servers so far: the PDF-import pipeline and the two dev tools, one for the database and schema, one that uses SVN for recent file changes. The dev tools are already accessible to every developer; what AWS changes is scale. Right now wider use runs into token budget, not access, and Lambda plus a proper endpoint is what makes running them across the team affordable.
The real demo came later, at our internal AI meeting with a senior dev. He framed the value as a step toward proper system documentation, which lines up with the design system work I've started leading. Both are about turning tribal knowledge (schema relationships, commit history, component patterns) into something documented and shared.
Where this is going
AWS pointed the way on two open problems in one meeting: hosting the dev-workflow servers at scale, and extracting data from the irregular PDF formats behind the import problem, which I'm now working on directly with their team. Nobody planned for me to lead that meeting; the results from the last few months made it the obvious call. That's the pattern: the more the AI work holds up, the more of it I'm handed to lead. At this point I run the AI planning and own the infrastructure roadmap.