What is Agent Ops?
Agent Operations (Agent Ops) is the work of designing, deploying, supervising and improving the AI agents that do real work inside a business: agents that research and contact prospects, qualify and route leads, answer customers, update the CRM and move deals through quote-to-cash. It covers the whole life of an agent, from choosing the job it does and the data and permissions it gets, to watching its results, handing edge cases to people, and keeping its cost and risk under control.
Agent Ops is to AI agents what RevOps is to the revenue engine: the people and practices that make agents dependable, measurable and safe once they are doing real work.
Why it exists now
Revenue teams have moved from trying AI to running it in production. Salesforce reports more than $1.5B in Agentforce annual recurring revenue (Q2 FY27, August 2026). Its own support site runs an agent that handles about 32,000 conversations a week with an 83% resolution rate. CEO Marc Benioff has described its support conversations as split roughly 50/50 between agents and humans.
The failures teach as much as the wins. Klarna said its assistant did "the equivalent work of 700 full-time agents" in its first month (February 2024). Fifteen months later it was hiring human agents again and promising customers they could always reach a person. Once an agent is doing real work, someone has to own how well it does it. That ownership is Agent Ops.
What Agent Ops covers
Agent Ops spans eight areas. Small teams give them all to one person; larger ones split them across GTM Engineering, RevOps and a platform team such as Salesforce.
- Use-case design: choosing the jobs agents should do, what "done" means for each, and where a person must stay in the loop.
- Data and access: the CRM data model, enrichment, and knowledge the agent reads, plus the permissions that limit what it can see and change.
- Build and deployment: agent instructions, topics and actions (in Agentforce, Clay, n8n, LangGraph or custom code), tested before release and rolled out in stages.
- Human handoff: escalation rules, the route to a person, and the human-to-agent (H2A) ratio, meaning how many agents one operator supervises.
- Evaluation and QA: test sets, reviewing samples of real conversations, and turning each failure pattern into a fix.
- Monitoring and reliability: resolution rates, errors, drift and outages, with alerts when an agent starts behaving differently.
- Cost and consumption: usage-priced platforms bill per action or credit, so actions per resolved task becomes a number to manage.
- Governance and compliance: audit trails, disclosure that a customer is talking to an AI (required by rules such as the EU AI Act), data retention and brand safety.
Where agents work in a revenue team
Most Agent Ops work in go-to-market teams today falls into six places:
- Outbound: agents that research accounts, enrich contacts and draft personalised outreach for a human or a sequence to send.
- Inbound: qualifying, scoring and routing leads, and booking meetings.
- Customer service: answering questions, resolving cases and escalating to people. This is the most mature use.
- CRM hygiene: logging activity, deduplicating records, and keeping fields and stages accurate.
- Quote-to-cash: quotes, approvals, contracts, billing and renewals.
- Forecasting and insight: summarising pipeline, flagging deal risk and preparing reviews.
The jobs
Agent Ops is a set of responsibilities, not yet a single job title. On AgentOps.careers it shows up in these roles:
- GTM Engineer: builds the automated systems behind sales and marketing, increasingly the AI agents themselves.
- Agentforce Architect / Developer: designs agents on Salesforce, including their topics, actions, Data Cloud grounding and guardrails.
- RevOps AI Automation Specialist: automates forecasting, routing, pipeline hygiene and quote-to-cash, and runs agents across the revenue lifecycle.
- AI Agent Operations Lead / Agentic Operations Architect: owns the agent estate end to end, including quality, cost, escalation and governance.
- Forward Deployed Engineer / Deployment Strategist: puts agents into customers' businesses, a fast-growing title at AI and platform vendors.
How Agent Ops is measured
Good Agent Ops reports outcomes, not activity:
- Resolution rate, alongside escalation rate. One without the other hides problems.
- Repeat-contact rate: did the customer come back because the agent didn't really solve the issue?
- Cost per outcome, or actions per resolved task, on usage-priced platforms.
- Revenue outcomes: meetings booked, pipeline created, speed to lead, forecast accuracy.
- Quality: the share of reviewed conversations that were correct, on-brand and compliant.
- H2A ratio: agents supervised per operator, as a measure of how far the team has scaled.
Agent Ops compared with RevOps, MLOps and DevOps
| Agent Ops | RevOps | MLOps | DevOps | |
|---|---|---|---|---|
| Runs | AI agents doing business work | The revenue process and its systems | Machine-learning models | Software and infrastructure |
| Main question | Is the agent doing the job well, safely and at a sensible cost? | Is the revenue engine efficient and predictable? | Is the model accurate and served reliably? | Is the software shipping and running reliably? |
| Typical tools | Agentforce, Clay, n8n, LangGraph, CRM, eval and logging tools | CRM, CPQ, BI, forecasting | Training pipelines, feature stores, model registries | CI/CD, cloud, observability |
Agent Ops roles open now
GTM Engineer jobs (42) · Agentforce Architect jobs (45) · RevOps AI Automation Specialist jobs (22). The tools named most often in these listings: Salesforce, Python, Apex, Agentforce, Data Cloud, SQL, Clay, LangGraph.
Frequently asked questions
What is Agent Ops?
Agent Operations (Agent Ops) is the work of designing, deploying, supervising and improving the AI agents that do real work inside a business: agents that research and contact prospects, qualify and route leads, answer customers, update the CRM and move deals through quote-to-cash. It covers the whole life of an agent, from choosing the job it does and the data and permissions it gets, to watching its results, handing edge cases to people, and keeping its cost and risk under control.
Is Agent Ops the same as RevOps?
No, but they overlap. RevOps owns the revenue process and systems; Agent Ops owns the AI agents that now do parts of that process. In many companies the RevOps team is where Agent Ops starts.
Is Agent Ops the same as MLOps?
No. MLOps runs the training and serving of machine-learning models. Agent Ops runs agents built on top of models, usually bought rather than trained, and cares about business outcomes, handoffs, permissions and cost more than model accuracy.
What skills does Agent Ops need?
A mix of systems thinking and hands-on building: CRM data models (Salesforce or HubSpot), workflow and agent tools (Agentforce, Clay, n8n, Make, Zapier), enough Python, JavaScript or SQL to glue systems together, prompt and evaluation design, and the judgement to decide what an agent should not do.
What is an H2A ratio?
The human-to-agent ratio: how many AI agents one person supervises. 1:10 means one operator runs ten agents. Employers increasingly use it to describe the size and shape of a role.
Where can I find Agent Ops jobs?
AgentOps.careers lists GTM Engineer, Agentforce and RevOps AI roles, collected daily from company careers pages, at https://agentops.careers/.