Early-career engineers are learning the craft in an environment where the surface they touch is often an agent, not a codebase. The instinct is to treat this as a rupture in how engineers grow. It isn't. The principles that grew engineers before agents still hold: gradually widening the scope of what someone owns, pairing to transmit judgment that doesn't fit in a doc, feedback loops tight enough to build intuition, real exposure to production. Those are how humans learn a craft. What's changed is the mechanics underneath each one, and the mechanics are where growth can stall.

Scope no longer means "size of the diff." Pairing no longer means "two people at one keyboard." The tightest feedback loop in a junior engineer's day is now the loop between them and their agent, and the shape of that loop is doing more of the teaching than most teams have noticed. Apply the old principles through the old mechanics and you get engineers who ship fast for a year and then plateau, or worse, look senior on the surface and can't hold up under a real incident.

Drawing on how we've been growing engineers inside Netflix through this transition, I'll lay out a working framework: what to keep, what to rebuild, where the failure modes hide, and how to tell the difference between an engineer who's actually growing and one who's just moving fast.

What you'll leave with:

• A concrete model for scoping work to a growing engineer when the agent is doing the typing.
• A rebuilt definition of pairing for the agentic loop, and where it belongs in a team's week.
• The failure modes that look like productivity and aren't, with signals to watch for.