Your Next Hire Is an AI Agent. Most Teams Haven't Written the Job Description.
Agents are moving from experiment to workforce. The organizations that get value from them will treat them like a role, not a feature.
An industry survey sponsored by Kantata found that 87% of professional services organizations plan to use AI agents as part of their workforce. The number caught my attention less for its size than for the word workforce. You don't describe a reporting tool that way.
When someone joins your workforce, they get a role and a manager, plus expectations someone checks. In most services organizations I talk with, AI doesn't have any of that yet. The agent gets switched on, operations is excited, and three months later it's unclear whether it's helping.
An AI agent with no job description is not a teammate. It is an unmanaged experiment.
An agent is a role, not a tool
I have spent my career helping organizations define roles in their PSA: who creates projects, who approves time, who sees margin. Agents need the same thinking. The difference is that this team member works around the clock and won't ask a clarifying question unless you design it to.
Before an agent touches your delivery operations, it needs answers to four questions.
What is its scope? One defined job. "Draft the weekly project status summary from logged time and milestones" is a scope; "help with projects" is a wish.
What does it work from? The data it relies on, and whether that data is trustworthy today.
Who owns its output? A named person who reviews, corrects, and answers for what the agent produces.
How will you know it is working? A measure you set before it starts, not after.
If those sound like questions you'd ask before hiring a project coordinator, good. That's the idea.
What changes inside your PSA
Most AI conversations skip this part, and it is where my work lives. Your PSA assumes people do the work, people log time, and time drives cost and revenue. Agents change each piece of that.
Capacity planning. If an agent drafts status reports, builds project plans from templates, or reconciles timesheets, the hours your coordinators and project managers spend on that work go down. Your resource plan still assumes the old hours until someone updates it.
Utilization targets. When admin work shrinks, targets built on the old mix stop meaning what they used to. A consultant who gets four hours back each week either fills them with billable work or loses the gain.
Pricing and revenue. On time and materials work, an agent doing part of the job means fewer billable hours for the same outcome. Some firms are already experimenting with outcome-based pricing tied to client KPIs instead of deliverables. Whichever way you land, that is a commercial decision first, and your PSA has to model it.
Cost of delivery. Agent work doesn't show up on a timesheet. For an accurate view of project margin, you need a way to account for what the agent costs and where it contributed.
None of this requires a new system. It requires deciding how the system reflects a team that now includes contributors without timesheets.
Your PSA assumes each hour of work has a person attached. That assumption is expiring.
Where agents earn their keep today
In the field, agents do well with work that repeats weekly and is easy to check against a standard.
Drafting project status summaries from time, milestones, and project notes
Flagging timesheets that are missing, late, or out of pattern
Spotting forecast variance before the month-end review instead of during it
Checking new project setup against your standards: a missing billing rule, the wrong template, an unassigned role
Preparing a first draft of the reporting narrative for leadership
They struggle with work that depends on judgment and relationships: a scope negotiation, a client escalation, choosing which project gets your senior resource when two are on fire. That work belongs to people. It is also where your best people should spend more of their week, which is the reason to hand the rest to an agent.
Agents are good at noticing. People are still the ones who decide.
Agents amplify the maturity you already have
An agent inherits the discipline of the organization around it. This is where the Services Operations Maturity Model is useful beyond a self-check.
At Level 2, Activation, timesheets are inconsistent and the data is still earning trust. An agent working from that data produces confident summaries of unreliable numbers, and it sounds as sure as it would on clean data. At Level 4 and above, where processes are consistent and reporting has a rhythm, the same agent multiplies what your team can do. I wrote about the data side of this in Your PSA Data Is Ready for AI. Are You?; agents raise the stakes on everything in that piece.
Governance belongs in this conversation too. The Thomson Reuters Institute found that while most corporate departments want their outside firms using AI on client work, fewer than a third know whether their firms are doing so. In services, client trust is the product. Decide now how you will disclose agent involvement, who reviews output before it reaches a client, and how you will trace what the agent did. Those decisions are easier to make before an agent sends something it shouldn't.
Start with one job description
You do not need a platform strategy to begin. Pick the most repetitive task your delivery or operations team handles each week. Write its job description: the scope, the inputs, the owner, the measure. Run it for a month with a person reviewing each output, then decide whether it has earned more.
That is not a technology project. It is the same operational design work that makes a PSA implementation succeed, applied to a new kind of team member. Organizations that do this on purpose will get the promised productivity; those that switch agents on and hope will get more noise in a system that already has plenty.
Hire the agent the way you would hire a person: one clear job, one owner, and a review before it gets more.
Not sure your data and process are ready for an agent?
The SOMM self-assessment places your organization in under five minutes and shows what the next level takes.
Thinking through where agents fit in your operation?
Mapping which work belongs to an agent and how your PSA should reflect it is the kind of operational design I help services organizations with. Let's talk.
Get the next one in your inbox.
PS Playbook is a weekly(ish) read for Professional Services teams who want their operations to actually work.
