What we shipped
- A software catalogue of every agentic component we are building, laid out in swim lanes with its product requirements, features and vertically-sliced tasks, so agents can build against it around the clock while staying inside spec.
- An agency MCP that lets the team reach all of it from any AI surface. Ask for a client's hero keywords, when their QBR is, or what they asked for on the last call, or push an update like "add this keyword to their profile", and get an accurate, current answer.
Why we built it this way
As the number of agents starts to grow and as AI gets better at working autonomously in loops, we need one place that holds the plans, specs and future ideas so our team and our agents can help grow our systems together based on their real lived experience.
The catalogue keeps 24/7 agent work bounded by our own requirements and specs. The MCP gives people a single, natural way in, rather than a scatter of scripts and dashboards.
How it benefits clients
- Eventually the same tooling points at the client: "where is the link to this?", "why did these rankings rise and those fall?", answered immediately.
- Consistency. Deliverables are assembled from tested, shared components rather than rebuilt from scratch each time.
What's next
- Opening a version of the MCP surface to clients directly, so they can query their own account in plain language.
- Integrating more account management features with the MCP so a client can ask it anything and get an immediate answer or escalate to their account manager. Want to know why a ranking changed? Need to check if a technical update has been made already? We want to be able to answer that in real time with our MCP.
Technical innovations
The main issues with the current state of AI is memory, portability and access across systems. Project management is one tool, analytics another, call notes a different format all together so getting them to talk to each other is a massive challenge.
Building a medallion database system and an agency MCP gives our team access to everything in a single place. It also allows us to understand the most common requests and issues and use the queries to make the system better over time.
At the moment the MCP is used for ad-hoc issues as our preference is planned in deliverables inside of our frameworks and HIL/AFK workflows however it provides wide access to everything across clients and agency.
How it fits the agentic journey
This is the manifesto's operating model made concrete: everything is a framework, work breaks into primitives, and the humans plan while the machines run. The platform is where that belief gets turned into a real working network of interconnected agents that are augmenting and speeding up our teams work.