Managing Builders

Everyone at every company will use code to build applications. That changes the engineering manager's job more than anyone's.

At Vendasta I want our engineering managers to mentor and guide the whole company, beyond R&D. When sales, support and operations start building software, someone has to teach them how to build it well. That someone is the engineering manager.

Here is how I see the role changing.

| | Before | Now | | :-- | :-- | :-- | | What the team produces | Reliable code and features | Solved user problems | | How the team is shaped | Frontend and backend specialists | Full-stack builders | | The bottleneck | Typing and finding answers | Decisions and context | | What mentors teach | Syntax and best practices | System design and verification | | The main risk | Bugs and delays | Bloat and the illusion of competence |

Builders own the whole slice

AI coding tools are wearing down the walls between frontend, backend and operations. One engineer can now scaffold an entire product. The assembly line of backend, then frontend, then QA is going away. A manager now runs a team of builders who each own a vertical slice end to end.

You cannot read every line of AI-generated code. So the manager moves from code reviewer in chief to architecture reviewer in chief. The value is in spotting the flawed design before the AI writes a thousand lines on top of it.

Teams should review intent as well as syntax. Ask why the AI chose a pattern. Asking whether it compiles is the easy part.

The missing middle

This is the trap I worry about most. AI lets a junior engineer skip the struggle. The hours of debugging that build intuition just disappear.

A junior can now ship senior-looking code without senior understanding. So managers have to create the struggle on purpose. Swap "here is the fix" for "explain how this handles the edge case." Watch for the person who is leasing their thinking from the AI. If the AI went away for a day, could they still ship?

Volume is a trap

Lines of code and commit counts are now meaningless. AI produces unlimited code in seconds. Story points inflate too, because AI makes easy points easier.

Measure the time from a raw idea to a working feature in a customer's hands. And reward deletion. AI code tends to be verbose and generic, and it sprawls. The builder who uses AI to make a system smaller deserves credit.

The manager is the human in the loop

Teams have style guides for their languages. They will need style guides for how they prompt, so output stays consistent across the team.

Individual output is rising. Teams will get leaner and be expected to do more with the same people. The manager's main skill shifts from execution to prioritization.

Three lanes

If everyone builds, everything cannot run in the same place. I think about three lanes.

1. The platform. The product our customers use. Secure, audited, with direct access to code and data, built by R&D.
2. Internal production. The systems we run the business on, built on our own platform. Data comes through approved interfaces. R&D and trained citizen developers build here.
3. Prototypes and experiments. Anyone at the company, any tool, with data reached safely through approved interfaces.

The engineering manager is the person who helps work move from the third lane to the second, and from the second to the first, when it earns the move.

Code is about to become everyone's job. Teaching it well is the engineering manager's.