Software teams grew around specialist roles
Building software usually meant dividing work among specialists. One person built the interface, another managed cloud infrastructure, and someone else worked on data or automated tests. Each specialty took years to learn, so this division made sense.
It also created handovers. A decision about the interface affected the data behind it. A change to a deployment process depended on what the company’s cloud platform could support. The more people owned separate pieces, the more effort it took to keep the whole system pointed in the same direction.
Generalists tended to follow how the pieces connected instead of staying inside one technical specialty. Until recently, the cost of moving between those pieces was high. Knowing what to do was not enough. You also needed enough time and specialist skill to implement it.
AI lets one engineer cross those boundaries
That cost has fallen sharply. A single week can include work on a system for testing AI outputs, a fix for a slow document process, a dashboard for cloud resource use, an application configuration, and a frontend redesign. AI helps with understanding each area, comparing possible approaches, and producing the implementation.
Each problem still requires enough depth to understand it, decide what a good result looks like, give the AI the relevant context, and review what it produces. Then the work moves to the next problem.
The mechanics of delegating implementation are covered in Becoming a manager of agents. The change now goes beyond managing more coding work. AI allows one engineer to cross boundaries that used to require a handover to another specialist.
The scarce input is company context
AI can usually propose a sound technical answer. The harder question is whether that answer belongs in this company, for these people, at this moment.
Consider an internal tool that needs cloud infrastructure. An AI agent can create the files needed to run the automated deployment pipelines. What it may not know is that the platform team has not yet provided everything those pipelines require. Conversations with the people building the platform reveal that constraint. Creating the technically correct files now would still be the wrong decision.
The same applies to the frontend. AI can build one, review it, and replace it with technology that will be easier to maintain. It does not hear direct conversations with the engineers who will use the tool. They care about insights that help them in customer conversations. Most of the technical work happening in the background is clutter to them. That knowledge changes what belongs on the screen and what should stay hidden.
If AI had all of this context, it could make many of the same decisions. Today, that context is spread across conversations, relationships, past incidents, company plans, and half-finished systems. The value comes from finding it, judging it, and turning it into clear direction.
The need to understand delegated work is explored in Outsourcing thinking, not understanding, while Engineering is still hard covers why choosing what matters remains difficult. The additional point here is that this judgment can now travel much further across the system. AI supplies much of the specialist implementation needed along the way.
Generalists can now carry more of the system
A generalist in this setting enters whichever problem currently limits the system, understands it well enough to make a decision, and uses AI to carry that decision into working software. This requires going deep before moving on, rather than knowing a little about everything and leaving the details to others.
Deep specialists still matter when a problem reaches the edge of current knowledge or when a wrong decision has severe consequences. Their experience can quickly show which solutions have failed before and where hidden risks lie. Many projects, however, no longer need a different specialist to own every layer of implementation. A small team of people who know the company and take responsibility for the complete system can cover much more ground than before.
This favors engineers who understand the systems around the code. They know whom to ask, which constraints are temporary, what users actually care about, and how one local choice affects everything else. AI gives those people enough reach to act on that understanding directly.
The work may keep moving upward. If AI gains the full context of a team, the human role may shift toward aligning the department. If it gains the context of the department, the work may shift toward the business unit or the company. At each level, someone still has to draw out what people need, resolve competing interests, and turn that understanding into a shared direction. For now, generalists are especially well suited to that work.