Job titles are being sold as software

Something quietly strange happened in enterprise software this month. Two vendors shipped products that are not named after features. They are named after people's jobs.

The org chart got a price tag

Optimizely launched Virtual Teammates, a suite of role-specific AI coworkers. The initial roster: chief of staff, SEO and AI-search analyst, marketing analyst, personalization strategist, and conversion-rate-optimization manager.

Read that list again. Those are not software categories. Those are lines on a job board. Those are titles that exist on real people's LinkedIn profiles right now, attached to real salaries and real mortgages.

The mechanics matter more than the branding. Each teammate can be configured with organizational permissions, connected to specific systems, given scheduled work, and it keeps context across projects. That is four things at once: an access level, a toolset, a calendar, and a memory. Which is, if you squint, a decent working definition of a colleague.

For three years, AI tools have been sold the way power tools are sold. Here is a very fast drill, go build something. Optimizely is selling something closer to a carpenter: here is a thing with a role, a badge that opens certain doors, a standing Tuesday task, and a memory of what it did last month.

The shift is not that the software got smarter. It is that the software started applying for jobs.

Adobe bought a sales team and called it a startup

The second signal landed the same week. Adobe acquired Rilo, a Bengaluru-based AI workflow-automation startup, through IP licensing plus a team transfer. Terms were undisclosed.

Rilo was eighteen months old. Founded in 2025 by two IIT Bombay graduates. Its product let teams build and run workflows from plain natural-language instructions, and it had converged on go-to-market work specifically: competitive intelligence, prospecting, LinkedIn outreach, Reddit marketing, social content, and campaign research.

That is not a feature list. That is a staffing plan for a small revenue team.

And here is the part I find genuinely exciting rather than ominous. Rilo did not win by building a better model. It did not have one. It won by having an opinionated, specific set of workflows that map to how go-to-market work actually gets done, and that was worth acquiring inside of eighteen months. The moat was the opinion, not the intelligence. That is an enormously encouraging fact for anyone building small and specific right now.

What actually changes for the rest of us

The interesting consequence is not "the robot is better at your job than you are." On the evidence available, it mostly is not.

The consequence is that the unit of purchase is being redefined in public. For decades, expert work has been bought in units of a person's time. Hours, days, months, headcount. What both of these launches do is repackage that same work as a role with a defined scope: here is what it can touch, here is what it runs on, here is when it runs, here is what it remembers.

Once a buyer has seen work sold that way, they will keep comparing against it. Anything described as hours gets measured against something described as a scope. That is a communication problem long before it is a capability problem, and it applies to teams, freelancers, and internal departments equally. If you cannot say precisely what your work touches, what systems it reads, what it produces, and on what cadence, someone else will describe your job more clearly than you do, and they will attach a monthly number to it.

The honest counter-position is also getting more valuable, not less. Somebody has to decide what permissions a teammate gets. Somebody has to decide what happens when it is wrong. That job did not exist two years ago and nobody has a title for it yet.

Nobody has solved the supervision part

Which brings me to the most striking thing I read all week, from a completely different world.

At DIA's DoDIIS Worldwide 2026 conference in Tampa, Maj. Gen. Robert Kinney, the agency's chief AI officer, described the direction of travel bluntly: "Building agents, and then agents with agents, is the direction that we want to go." The described end state is an intelligence agent coordinating with separate agents for operations, fires, logistics, communications and planning, reasoning together across one mission problem.

Then officials named the blocker out loud. The tradecraft for responsibly letting agents interact with and control other agents, along with the compliance, security and trust layer underneath it, does not yet exist.

Sit with the symmetry there. A marketing platform is handing an agent organizational permissions, system access, and a schedule. An intelligence agency is describing agents that direct other agents. And the most senior technical voice in that second room is saying, in public, that the governance mechanics for the delegation have not been invented.

The unsolved problem is identical in both rooms, only the stakes differ. Which agent was allowed to do that. Who granted the scope. What data did it touch. Can we replay the decision. That is not a model problem. It is an audit problem, and it is the most interesting unbuilt thing in software right now.

Job titles are becoming SKUs. The supervisor role is the one that has not shipped yet.