61 Agents, 7 Departments, Zero New Hires: The Operations System Behind a One-Person Agency

Every agency reaches a structural ceiling. Not a skills ceiling. Not a capacity ceiling. A department ceiling.

A solo founder can execute client work. What they cannot do, not sustainably, is simultaneously run Finance, manage Account Health, screen job applicants, qualify inbound leads, track delivery milestones, monitor system health, and route incoming work to the right queue. Each of those functions requires a person with a specific responsibility, meeting cadence, and institutional knowledge. The cost of that kind of org chart runs $250,000 to $400,000 per year before benefits, office overhead, or the months it takes to onboard someone who actually knows what they are doing.

So most agencies skip whole departments. They run without a Finance function until unpaid invoices stack up. They have no Account Management layer until a client relationship breaks down quietly over weeks nobody noticed. HR is a folder of resumes with no process attached. Internal systems monitoring happens when something is already broken. The departments do not exist because the cost structure makes them impossible, not because the work inside them is unnecessary.

That is the real problem. Not missing people. Missing departments.

The question OpsForge was built to answer: what if the departments existed, but were staffed by agents instead of employees? Not replacing humans in roles they should hold. Building roles that never existed in the first place because the cost to staff them was prohibitive.

The infrastructure underneath all seven departments uses BullMQ for inter-agent messaging. Jobs travel between agents as structured payloads through a single queue (opsforge-agent-jobs). Agents do not call each other directly. They hand structured work to the queue, and the next agent picks it up when ready. This design keeps the system observable and prevents cascading failures from spreading between departments.

Three pipelines cross department boundaries. The lead-to-delivery pipeline runs 29 steps from initial lead qualification through Sales, across Account Management onboarding, into Delivery project setup. The monthly-ops-review pipeline runs 12 steps, pulling data from Finance, Delivery, and Account Management into a consolidated report. The client-lifecycle pipeline runs 4 steps, coordinating status between Account Management and Delivery at key relationship milestones.

Every agent action belongs to one of three classifications. Autonomous: the action runs and completes without any notification. Notify: the action runs, completes, and sends a Telegram message afterward. Approve: the action pauses and sends a Telegram message requiring explicit confirmation before proceeding. Client-facing communications, non-trivial financial actions, anything that crosses a threshold requiring human judgment. The system will not proceed until an approval arrives.

This classification is what makes 61 agents safe to run. Without it, the operator would receive constant noise and every message would compete for attention equally. With it, the Telegram feed is purely signal: an Approve message means something actually requires a decision.

OpsForge shares its technical lineage with MaxReach, the 30-agent content pipeline built on the same architecture. Combined, these two systems put 91 production agents running across two platforms. The 21-page dashboard provides visibility into every department's queue state, active jobs, and recent outputs.

Seven departments now exist that did not before.

Each one surfaces outputs on its own schedule. Sales produces a qualified lead summary without the founder touching the inbox first. Finance flags an aging invoice before the follow-up conversation has to be invented from scratch. Delivery generates a project status summary before the client asks for one. Account Management drafts a check-in message before a relationship goes quiet long enough to become a problem.

That shift, from doing operations to reviewing operations outputs, is the actual result. The founder still makes every strategic decision. Client relationships are still human-to-human. Every output that leaves the system gets reviewed before it touches a client. Agents produce work product, not final decisions.

What OpsForge did not change is worth naming clearly. No agent closes a deal. No agent manages a client relationship. No agent decides whether to take on a project or decline it. The system handles the operational layer underneath those decisions so that the human holding them has time and information to make them well.

Related: MaxReach: 30 Agents Running a Content Pipeline →