I’m Still an Individual Contributor. I Just Have a Team Now.

I did not become a people manager.

But I do not work alone in quite the same way anymore.

I notice it most in the work that used to stay stuck in my own head for too long. A draft that needs to be challenged. A product idea that feels close, but not quite there. A message that sounds right until I ask whether a customer would actually care.

That used to be a pretty lonely part of being an individual contributor.

You can ask people for feedback, and you should. But that is not always simple. People are busy. Some have the context but not the time. Some have the time but not the perspective. Sometimes, you do not even know who the right person is to ask.

That last part matters more than we admit.

A lot of professional work does not fail because no one cares. It fails because the right perspective never made it into the room early enough. The customer voice came too late. The technical concern came too late. The editor caught the vagueness after everyone had already aligned around it. The risk review happened after the work had momentum.

By then, feedback starts to feel like rework.

Agents are starting to change that for me.

The useful part is not asking one agent to review something. Yes, that can help, but it still mimics the old model in many ways. One person, one opinion, one pass.

The bigger change is being able to put a small group around the work.

I can ask three to five agents to take different perspectives. One can read it like a skeptical customer. One can challenge the logic. One can look for missing context. One can review the writing. One can take a broader director-level view and ask whether the whole thing is even pointed in the right direction.

Then I can have them challenge each other.

That is what feels different.

I am not just collecting five disconnected opinions. I can ask the group to compare notes, argue through the weak spots, and come back with a consensus. I can ask where they agree, where they disagree, and what they would change if the goal is clarity, usefulness, and credibility.

The final call is still mine. It has to be.

But I am no longer relying on one chat thread, one rushed reviewer, or my own tired read of something I have been staring at for too long.

The old feedback model was always fragile

Most individual contributors know this pattern.

You work on something. You get it to a point where it feels decent. Then you send it to someone and ask, “Can you take a look?”

That sounds simple. It rarely is.

You have to know who to ask. You have to hope they have time. You have to hope they understand what you are trying to do. You have to hope they are willing to be direct. You have to hope they are looking at it from the angle you actually need.

And even when they are helpful, they are one person.

That person may be great at messaging but weak on technical nuance. Or strong technically, but not close to the customer. Or senior enough to see the bigger picture but too far removed from the messy details.

None of that is a criticism of people. It is just how work works.

Good feedback is expensive. Not always in money, but in time, context, attention, and social capital. You do not want to waste someone’s time with something half-baked. You do not want to pull five people into a review before you know whether the idea is worth reviewing. You do not want to be the person constantly asking for “one quick look” when everyone else has their own work to do.

So a lot of early judgment stays with the individual contributor.

You become your own researcher, writer, editor, customer proxy, technical reviewer, and skeptic.

That can work. It can also trap you inside your own assumptions.

Agents are changing the first review

The first wave of AI made it easier to draft. Easier to summarize. Easier to clean up a paragraph or generate options.

That was useful, but it still mostly treated AI like a better blank page.

Agents are different because they allow work to move through a process.

Instead of asking, “Can you make this better?” I can ask for a small review structure.

Something like this:

Give each reviewer a different role.

Have them review independently.

Have them challenge each other.

Then ask for consensus, disagreement, and recommended changes.

That is a different kind of help.

It is not just faster writing. It is structured pressure.

The agents are not perfect. They can overstate. They can miss context. They can agree too quickly if the prompt is weak. They can sound confident when they should slow down.

But when the process is well set up, they can make the work less dependent on a narrow path of thought.

I can see patterns sooner. I can catch lazy assumptions sooner. I can find the sentence that sounds good, but does not say enough. I can spot where I am writing for myself instead of the person who has to read it cold.

This is already showing up in the tools. OpenAI’s Codex subagents documentation describes specialized agents that can run in parallel and return their results back into one response.

This matters because it moves AI from “give me an answer” toward “help me run the work.”

A single chat session can still be useful. I use them all the time. But a single thread can also get narrow. It can follow the same assumption or biases (yes, AI has bias) too far. It can miss the thing you did not know to ask. It can produce a clean answer that still does not hold up when another perspective gets involved.

A group of agents can help break that pattern.

Not automatically. Not perfectly. But enough to matter.

The consensus matters

The consensus is where I am seeing the biggest impact.

One agent can be useful. A group of agents can be more useful if they are forced to disagree before they agree.

That is an important distinction.

If I ask five agents the same question and they all give me five polished answers, I may just get five versions of noise. But if I tell them to take different perspectives, challenge each other, and then come back with a shared recommendation, I get something closer to a working review board.

Not a real board. Not a substitute for actual experts. But a useful first pass before I put the work in front of people.

I have started to think of this as a private review layer.

Before I publish something, I can ask agents to read it as editors would.

Before I make a product change, I can ask agents to act as a change review board.

Before I settle on messaging, I can ask agents to take customer personas and tell me what feels clear, what feels empty, and what sounds like vendor language.

Before I trust an answer, I can ask agents to look for hallucinations, unsupported claims, weak reasoning, or missing context.

Consensus does not mean truth.

If five agents agree on something, that does not make it correct. It means the pattern is worth paying attention to. It gives me a signal. It gives me a better set of questions. It gives me a reason to go verify, rewrite, or rethink.

The human judgment does not go away.

If anything, it becomes more important.

This is management, just not people management

Many individual contributors have avoided people management for good reasons.

Some do not want the calendar load. Some do not want the performance review process. Some would rather stay close to the work. Some are better at building, writing, analyzing, selling, designing, or solving than managing a team of humans.

That does not mean they do not want to or don't know how to lead work.

Agent management is becoming its own skill.

You have to know what outcome you want. You have to assign useful roles. You have to give enough context without drowning the process. You have to decide when disagreement is helpful and when it is just churn. You have to know when the group is converging too fast. You have to know when to ignore the recommendation.

That sounds a lot like management to me.

Not human management. Work management.

The individual contributor is no longer only producing the work. They are increasingly designing the process that produces the work.

The best ICs may not be the ones who simply use AI the most. They may be the ones who know how to create better review loops, challenge structures, and decision support for their work.

The skill is not prompting in the shallow sense.

The skill is knowing how to convene the right room before you even know who the right people would be.

There is a governance side to this

This is where my security brain kicks in a bit.

The more agents become part of real work, the more we have to be honest about what they touch, what they can access, and what decisions they influence.

There is a difference between using agents to review a blog draft and using agents to make changes in production systems, approve access, evaluate vendors, or influence customer-facing commitments.

Those should not be treated the same.

Agentic work needs boundaries. It needs access control. It needs review. It needs a record of what happened and why. It needs a clear human owner.

That is also why I am careful with the word “team.”

It is useful as a metaphor, but only up to a point. Agents are not employees. They are not accountable. They are not morally responsible for the outcome. They do not absorb the consequences if the work is wrong.

A recent Harvard Business Review piece made a related point: when people treat AI agents too much like employees, accountability can start to blur.

That warning is worth taking seriously.

Agents can help with the work.

They cannot own the work.

The person using them still does.

That is not a small detail. That is the whole thing.

The value of agents is not that they let us stop thinking. The value is that they can help us think from more angles before we act.

The IC role gets bigger

I do not think this means every individual contributor suddenly becomes a manager in the traditional sense.

I think it means the IC role expands.

The work can become less solitary. The first review can happen earlier. The weak spots can surface before the meeting. The customer perspective can show up before the launch. The security question can be asked before the change request. The editor can show up before the draft gets sent around.

That is a better way to work.

It also raises the bar.

If you have access to a team of agents, “I did not think of that” becomes a little less acceptable. You may still miss things. Everyone does. But you have more ways to challenge yourself before the work leaves your hands.

That is where agentic AI starts to feel real to me.

As a way for one person to bring more judgment into the work earlier.

That matters in writing. It matters in product strategy. It matters in security architecture. It matters anywhere the quality of the work depends on seeing the problem from more than one angle.

Agents can help us see more.

They can help us challenge the work earlier.

They can help us build a better room around the problem.

Then we decide.

That is still the job.

But it is not the same job as it used to be.