The Quiet Promotion: Why Agentic AI Moves Your Role Upstream

The most frequent question I hear when I run sessions on agentic AI is some version of the same thing. "What does my role look like when this is real?" Business analysts ask it, project managers ask it, developers ask it, IT leaders ask it. And the honest answer is not what most people expect.

Your role is not going away. It is moving upstream.

That distinction matters. Agentic AI automates the production of artifacts. Requirements documents. Status reports. Code modules. Test cases. Research summaries. The work that has historically filled your days is increasingly work that a well-directed agent can produce in a fraction of the time. What does not automate — at least not yet, and arguably not at all — is the thinking that has to happen before the artifact and the judgment that has to happen after.

The professionals who thrive in the agentic era will be the ones who spend less time producing and more time directing, validating, and governing.

Business analysts: from documenting to architecting

The business analyst role has historically been anchored in artifact production. Requirements documents. Use cases. Process maps. Stakeholder interview summaries. A good BA could fill a week producing the artifacts that captured a single conversation.

Agents can draft those artifacts now. A well-configured agent with access to meeting transcripts, source documents, and your organization's template library can produce a first-draft requirements document in minutes. It will not be right, but it will be directionally correct, and the BA's work shifts from producing the draft to validating, refining, and completing it.

The deeper shift is upstream. BAs in the agentic era spend more time on the questions agents cannot yet answer. What is the actual business problem? Who are the stakeholders whose needs are not represented in the documents? What is the organizational context that changes how this requirement should be interpreted? The BAs who move into this work will be more valuable, not less. The BAs who stay focused on artifact production will find themselves competing with software.

Project managers: from chasing to deciding

Project managers spend an enormous amount of time on status. Chasing team members for updates, compiling reports, formatting summaries for leadership, tracking budgets and burn rates in spreadsheets. None of that is high-judgment work, and all of it is work an agent can do.

What agents cannot do is decide. They cannot judge whether a risk is acceptable, which tradeoffs are appropriate, how to have a hard conversation with a stakeholder, or whether a team is struggling because of workload or because of something deeper. These are the judgment calls that define project success, and they are exactly what project managers will spend more time on when the agent handles the status mechanics.

There is also a new responsibility: managing token budgets. Agentic projects have variable, usage-driven costs that traditional project financial models do not anticipate. PMs in the agentic era need fluency in FinOps for AI, which is a skill most organizations are not yet teaching.

Developers: from writing to architecting

Developers have been on the front lines of this shift for two years. Coding agents like Claude Code, Cursor, and GitHub Copilot can now generate entire modules from a well-specified description. Anthropic has publicly stated that power-user software engineers should expect to spend $100K to $300K annually in token costs when coding with AI assistants, which is one way of saying the work is fundamentally changing.

The developers who adapt fastest are the ones who stop identifying as code writers and start identifying as system architects. The valuable work is not producing the function. The valuable work is designing the system the function belongs to, deciding which components the agent should build and which require human authorship, reviewing the agent's output with an eye toward maintainability and security, and integrating the pieces into a working whole.

Code review becomes a much more intensive activity. When an agent produces hundreds of lines of code in minutes, the bottleneck is not writing. It is reading, understanding, and confirming. Strong technical judgment becomes the differentiating skill, not typing speed.

IT leadership: from approving to governing

The most significant role shift may be at the leadership level. IT leaders have historically made approval decisions about projects, vendors, and strategic direction. The agentic era adds a new category of decision: what agents are allowed to do, on whose behalf, with what oversight, and with what accountability.

This is governance work. It looks more like regulatory compliance than traditional IT leadership. Defining scope boundaries for agents. Establishing audit trail requirements. Setting escalation policies. Deciding which processes are appropriate for autonomous execution and which require human approval gates. Measuring AI-driven outcomes in ways that do not mislead executives into believing they have captured savings they have not actually captured.

IT leaders who build this governance muscle early will find themselves in much stronger positions when their organizations try to scale agentic deployments. IT leaders who treat agents as "just another tool" will find themselves on the receiving end of preventable compliance issues.

What to do about it now

The good news is that the shift is gradual enough to prepare for. A few practical steps.

Build genuine fluency with generative AI first. Before you can direct an agent, you need to understand how to direct a model. Use Claude, ChatGPT, or Gemini as a first resort for tasks, not as a curiosity. Ethan Mollick's advice to "try everything with AI first" is the fastest path to the intuition that matters.

Develop your judgment, not your production skills. The artifacts you produce are becoming commodity. The judgment you apply to them is becoming scarce. Invest in the thinking.

Pick one workflow in your current role and ask what it would look like if an agent did it and you validated the output. Not someday. This quarter. That exercise, repeated over time, is how you actually prepare for a role shift that is already underway.

Your role is not going away. But the version of it that existed five years ago is. The professionals who see that clearly and move accordingly are the ones who will be quietly promoted into the roles that matter next.