There is a growing assumption that if AI makes employees more productive, organizations will need fewer managers. If software can be generated faster, progress can be tracked through digital tools, and routine updates can be summarized automatically, why maintain the same management structure?

It sounds logical until the actual work of managing people is examined.

A manager does far more than collect status updates, assign tickets, and make sure deadlines are met. Good management requires understanding technical risks, recognizing when information is incomplete, helping people make decisions, resolving conflicts, and taking responsibility when those decisions have consequences. AI can assist with parts of that work, but automating individual tasks is not the same as replacing the judgment required to connect them.

The arrival of coding agents makes this distinction especially important. Engineers can produce more software in less time, but that also changes how work should be divided, how quality should be assessed, and how teams coordinate. At the same time, managers are being asked to adopt new tools, improve productivity, oversee larger teams, and help employees deal with uncertainty about their own careers.

The question is no longer whether management will change. It already is.

The more useful question is whether organizations will use AI to make managers more effective—or simply expect fewer people to carry more responsibility.

A common misunderstanding about engineering management is that technical skills become unnecessary once someone moves into leadership. In reality, technical knowledge helps managers assess risk, challenge assumptions, and determine whether a proposed solution is realistic.

Consider a project that is expected to take two weeks. An engineer using an AI coding agent produces a working prototype in two days. The result looks promising, and leadership wants to commit to a launch date immediately.

A manager without sufficient technical understanding may interpret the prototype as evidence that most of the difficult work is complete. An experienced manager knows that working software is only one part of the equation. Security, integration, testing, maintainability, deployment, and unexpected failure cases can still demand significant effort.

AI changes the speed of implementation, but it does not make every engineering problem equally simple.

It also changes the amount of work one engineer can attempt independently. Tasks that once needed several people may now be manageable by a smaller group, while other projects become more ambitious because implementation is less expensive. Managers must reconsider how work is divided, which dependencies matter, and where additional review is necessary.

The challenge is to avoid treating AI-generated output as proof of productivity. A feature that appears complete but introduces security weaknesses or creates maintenance problems may transfer the cost of today's speed to the team responsible for tomorrow's system.

Technical understanding allows a manager to distinguish between a fast demonstration and a dependable product. It does not require personally writing every line of code, but it does require knowing enough to ask the right questions and recognize when a confident answer needs further examination.

AI is becoming increasingly useful for summarizing work. It can review project documents, examine digital activity, organize updates, and help a manager prepare for a one-on-one meeting. Used carefully, these capabilities can reduce administrative work and give managers more time for meaningful conversations.

But visibility and understanding are different things.

A summary might show that an engineer completed several tasks during the week. It may not explain why the work took longer than expected, which technical compromise was necessary, or whether another team created a dependency that delayed progress. Those details often emerge through conversations, observation, and follow-up questions rather than from a collection of digital artifacts.

This matters because management is partly a process of signal detection. A manager needs to separate important developments from background activity, identify emerging problems, and recognize when something that appears minor could become consequential.

AI can help surface possible signals, but its summaries necessarily depend on the information available to it and the way that information is interpreted. If a crucial conversation never appears in a document, the system may have no way to account for it. If a summary omits an unusual detail, a manager may overlook the very clue that would have changed the decision.

There is also a risk of losing situational awareness. Tone of voice, hesitation, body language, silence, and the difference between what someone says and what they seem reluctant to say can all provide context. None of these signals guarantees that an interpretation is correct, but they can indicate where further conversation is needed.

When a meeting is reduced to a transcript and summary, some of that context may disappear. If managers rely entirely on automated notes, they may spend less time paying attention to the people in the room and more time reading what the system decided was important.

The better approach is to use AI to reduce the burden of collecting information while preserving the human work of interpreting it.

Few management responsibilities demonstrate the difference between assistance and accountability more clearly than performance reviews.

AI can help organize evidence, retrieve examples of completed work, and compare documented contributions against an established competency framework. This can be useful because managers are vulnerable to recency bias: the tendency to give disproportionate weight to recent events while forgetting earlier contributions or difficulties.

Used as a memory aid, AI can make a review more complete.

The risk begins when the tool moves from gathering evidence to deciding what the evidence means.

A performance review is not merely a summary of activity. It is a judgment about an individual's contribution, development, strengths, and areas for improvement. It can influence compensation, promotion, future opportunities, and the employee's trust in their manager.

A polished assessment may sound reasonable while overlooking context that changes the interpretation. A missed deadline might reflect poor planning, an unrealistic commitment, or an unexpected dependency. A highly visible project may have received more attention than less visible work that was equally important. A list of accomplishments cannot always reveal how much support someone needed or how effectively they helped others succeed.

The manager must therefore remain responsible for the judgment.

This is also a matter of trust. Employees expect their manager to know their work well enough to explain an assessment, provide meaningful feedback, and defend the conclusions reached. If the review reads like an automated summary with no evidence of personal consideration, even a favorable evaluation can feel dismissive.

AI can help a manager remember more. It cannot take responsibility for the relationship that follows the decision.

If AI can accelerate implementation and reduce administrative tasks, managers should theoretically have more time for people. In practice, that outcome is far from guaranteed.

Some organizations are using expected productivity gains to justify larger teams per manager, fewer management positions, and new expectations that managers also contribute directly to technical delivery. At the same time, managers are expected to encourage AI adoption, assess its usefulness, address employees' concerns, and keep work moving while the organization is changing around them.

That combination creates competing demands.

Managing a larger team can reduce the time available for individual coaching. Taking on more implementation work can crowd out hiring, feedback, planning, and coordination. Introducing AI without adequate guidance can leave managers responsible for outcomes they have not been given the time or training to support.

It is understandable that some managers may question whether the role remains worthwhile. When an organization expects someone to perform the work of an individual contributor while simultaneously managing an expanding group of people, the problem is not necessarily the manager's willingness or ability. It may be an unrealistic job design.

The distinction matters because management work is often less visible than the output of an individual contributor. A developer can point to a feature. A manager's contribution may be reflected in a conflict resolved before it derails a project, a struggling employee who receives timely support, a hiring decision that improves the team, or a priority clarified before months of effort are wasted.

Those outcomes are real, even when they do not appear as a line of code or a completed ticket.

Organizations that overlook this work risk removing the very support systems that help technical teams perform well.

For years, writing software required enough time and effort that producing a working implementation could demonstrate commitment. An engineer who built a solution was often showing initiative, technical ability, and a willingness to take responsibility.

AI changes that signal.

When generating code becomes much easier, a person can build a parallel implementation simply because using an existing system requires coordination with another team. A project that once demanded substantial effort may now be easy enough to start without asking whether the work belongs elsewhere.

The immediate result may look productive. A working feature exists, and the team has avoided waiting for another group.

The longer-term consequences can be less appealing. Two teams may maintain overlapping systems. New dependencies may be introduced without the owning team's knowledge. Security assumptions, data access rules, and architectural decisions may be bypassed because the agent can use whatever resources it can reach.

This connects to Conway's Law, the observation that a system's design tends to reflect the communication structures of the organization that builds it. Teams often develop software around their responsibilities, working relationships, and established channels of coordination.

AI does not inherently understand those organizational boundaries. It can generate a technically plausible solution using a system that another team owns, without recognizing the history, commitments, or support obligations surrounding that system.

The issue is not that every team must seek permission for every implementation detail. Excessive coordination can also slow progress. The challenge is to distinguish between harmless independence and decisions that create shared operational, security, or maintenance responsibilities.

As implementation becomes cheaper, organizations need clearer ownership, well-defined interfaces, and better ways to discover existing capabilities. Otherwise, faster development can produce more duplicated work and more conflict over who is responsible for what.

The strongest use of AI in management is not to remove human judgment from every process. It is to reduce the repetitive work that prevents people from applying judgment where it matters.

AI can prepare meeting notes, retrieve examples of past work, summarize project changes, draft initial plans, and help identify possible risks. These are useful capabilities when their limitations are understood and the results are checked.

Managers still need to decide what matters, resolve conflicting priorities, evaluate trade-offs, develop people, and take responsibility for decisions. Those responsibilities do not disappear simply because information becomes easier to collect.

A summary can help identify a performance concern, but a conversation is needed to understand it. A coding agent can produce a solution, but someone must determine whether it fits the system and its requirements. An automated planning tool can identify dependencies, but people still need to negotiate ownership and make choices when priorities conflict.

The goal is not to preserve every traditional management ritual. Some status meetings, manual reports, and repetitive administrative tasks should absolutely be reconsidered. The goal is to preserve the purpose behind useful management practices while finding better ways to achieve it.

That requires experimentation rather than blind loyalty to old processes—or blind confidence in new tools.

Technical judgment, information processing, and people skills have always mattered in engineering management. AI makes the relationship between them more visible.

Technical judgment helps a manager understand what can be built, what might fail, and which shortcuts are likely to become expensive. Information processing helps the manager find the right signals, question incomplete summaries, and make decisions without confusing activity with progress. People skills help turn those decisions into coordinated action while maintaining trust, clarity, and accountability.

None of these capabilities works well in isolation.

Technical knowledge without communication can lead to decisions the team does not understand. Information without judgment can create the illusion of certainty. Strong interpersonal skills without technical understanding can make it difficult to recognize risks before they become serious.

AI can strengthen all three when used deliberately. It can help managers learn unfamiliar technical concepts, organize information, prepare for difficult conversations, and explore alternatives. But it can also amplify weak management when its output is accepted without scrutiny or when organizations mistake greater visibility for deeper understanding.

The real opportunity is to make management more thoughtful, not merely more automated.

For someone leading a team today, that means resisting two tempting extremes: assuming AI will make management obsolete, or assuming the role can remain unchanged while every other part of software development transforms.

The practical path lies between them. Use AI to remove low-value work. Reconsider processes that no longer fit the speed of development. Protect the time required to understand people and systems. Make decisions that can be explained, and remain accountable for their consequences.

The central lesson is straightforward: AI can help an organization move faster, but speed alone does not create alignment, sound decisions, or trust. Those outcomes still depend on how people are led, how information is interpreted, and who takes responsibility when a decision matters.

The manager's role may change considerably in the age of AI. Its value, however, is not measured by how many tasks a manager personally completes or how many reports a system can generate. It is measured by whether the team can make better decisions, deliver dependable work, learn from mistakes, and work together without losing sight of the people behind the output.

Hit reply and tell me: what is one thing someone else noticed about your work that you could not see yourself? I read every answer.

One more from today's sponsor, which helps keep Tiny Big Spark free:

Keep going, one small spark at a time.

Erwin