gtm-agent-fleet

In early 2026, Replicated’s go-to-market team had a capacity problem. The team was small — two AEs, an SE, a director — and too much important recurring work had no durable owner. Inbound leads went stale. Customer channels went unmonitored. CRM hygiene drifted. The usual answer would have been headcount or shifting work to other teams. Instead, we hired agents.

The agents on the GTM team run on OpenClaw: Shelley, an AI SDR who handles inbound response and outbound sequence generation; Sandy, a customer success agent who monitors account activity and produces daily briefings on what needs attention; and Sebastian, a RevOps agent who handles CRM operations, data hygiene, and other workflow maintenance work that revenue teams. They have names, memory, tools, and defined boundaries. More importantly, they have jobs — and those jobs accumulate context over time rather than starting fresh each session.

That distinction is what made the experiment worth running. Most AI work gets framed as individual productivity: draft this email, summarize this thread, answer this question. Useful, but episodic. The question here was what happens when recurring GTM work has a persistent non-human owner instead. The answer is that the shape of the work changes. Inbound follow-up had an owner. Customer signal collection had an owner. Parts of revenue operations had an owner. The humans still owned judgment, relationships, and decisions — but they were no longer carrying every piece of recurring operational work in their heads at once.

The shared infrastructure mattered as much as the individual agents. All three run on the same platform, share common CRM and Slack tooling, and are maintained as a system rather than three separate one-off prompts. When the agents’ Attio skills grew inconsistent — 80% overlap, drifting facts, different formatting — the fix was consolidating them into shared modules rather than maintaining three copies. The management problems that came up — access, boundaries, trust, drift — were real team-management problems, just applied to agents. That turned out to be the clearest proof that treating them like persistent coworkers with narrow jobs is a fundamentally different approach than using AI as a productivity shortcut.