When agents can write specs, code, and documentation, the traditional roles of IC and PM are under pressure. “What do we need them for?” is a question that surfaces in orgs experimenting with agentic workflows.

The answer: they add value where agents can’t. Strategy, judgment, coherence, learning. But that value has to be reinforced—made visible, rewarded, and embedded in the workflow. Otherwise, people drift into agent-wrangling or busywork, and the org loses the upside of having humans in the loop.

What ICs and PMs Add (That Agents Don’t)

Strategy.
Agents execute. They don’t decide what to build or why. The PM sets direction; the IC makes technical decisions that shape the system. Reinforcement means ensuring those decisions are made explicitly and valued—not lost in the rush to “let the agent run.”

Judgment.
Agents produce; humans evaluate. Is this output good enough? Safe? Aligned with our goals? The IC and PM are the evaluators. Reinforcement means treating evaluation as core work—not an afterthought—and building in time and recognition for it.

Coherence.
Agents optimize for the local task. They don’t ensure the whole system hangs together. The IC maintains architecture; the PM maintains product consistency. Reinforcement means making coherence a responsibility that’s visible and rewarded.

Learning.
Agents don’t learn from outcomes in the same way humans do. The PM and IC interpret what shipped, what worked, what didn’t—and feed that back into the next cycle. Reinforcement means protecting time for reflection and ensuring that learning informs the next round of agent directives.

Reinforcing These Roles

1. Name them explicitly.
Don’t assume everyone knows what “strategy” and “judgment” mean in practice. Define the behaviors: “PMs own the prioritization decisions. ICs own the architectural decisions. Both own validation of agent output.” Make the roles legible.

2. Protect time for them.
If the calendar is full of execution-focused meetings, strategy and judgment get squeezed. Reinforcement means carving out space: “no-meeting blocks for deep work,” “reflection time after each sprint.” Structure supports the behavior.

3. Reward the right things.
If performance reviews focus on output volume (tickets closed, PRs merged), people optimize for that. Reinforcement means measuring and rewarding strategy, judgment, and coherence—even when they’re harder to quantify.

4. Model from the top.
Leaders who visibly spend time on strategy, judgment, and learning—and who talk about it—reinforce that these are valued. Leaders who are always “in the weeds” or only celebrate ship velocity send the wrong signal.

The new loop isn’t “agents replace humans.” It’s “agents execute; humans direct and validate.” Reinforcement is what keeps that loop stable and effective over time.