Guide

T-shaped skills mean depth in one stem (you can go deep enough to be dangerous and trusted) plus a crossbar of literacy in adjacent areas so you can collaborate and switch contexts. Pure generalists risk being "junior at everything." Pure specialists risk brittle careers when the specialty's market cools.

You can change stems once or twice. You cannot change them every year and still claim depth.

Walking through the shape

Specialist paths that tend to stay scarce as of the mid-2020s: inference/GPU serving, serious distributed data systems, security, certain ML specialties, platform/SRE at scale. Depth takes deliberate years — see paths like inference engineering or platform engineering.

Generalist paths thrive at startups and in product engineering where the bottleneck is integration and judgment, not novel algorithms.

A practical approach:

  1. Build a reliable generalist base (ship features, own production, communicate).
  2. Add a stem that the market pays for and that you can stand studying on weekends.
  3. Keep the crossbar alive with reading and the occasional adjacent project so you do not fossilize.

Markets swing. As of 2025–2026, AI-adjacent and infrastructure skills are loud; that will not always be true in the same way. Transferable skills that survive fashion cycles:

  • Clear written communication and design docs
  • Production ownership and incident judgment
  • Ability to learn a new stack quickly because you understand fundamentals
  • Taste for simplicity and maintainability
  • Reputation for finishing

Fashion skills are still worth adding as a T-stem when the market pays — just do not bet your entire identity on one framework or one vendor.

Example

Two résumés after a downturn. A: "5 years React only, no production ownership story." B: "Product engineer who owned billing authenticity end to end, reduced incident rate, wrote the migration plan others used; also fluent in React." B interviews worse on trivia maybe; B wins more offers because outcomes travel.

Charity Majors' Engineer/Manager Pendulum is useful even if you stay IC: careers are decades long, and the stem can shift once without becoming a new person every year.

Common mistakes

  • Junior at everything. No stem deep enough for anyone to trust you.
  • One vendor as identity. When that market cools, the résumé does too.
  • Changing stems yearly. Depth never forms.

Check yourself

  1. What is the stem vs the crossbar?

Stem: depth the market pays for. Crossbar: literacy in adjacent areas so you can collaborate and switch.

  1. After a downturn, why does "owned billing end to end" travel better than "5 years React only"?

Outcomes and production ownership survive fashion cycles; a single framework may not.

Key takeaways

  • T-shape: one scarce stem plus a useful crossbar.
  • Build the generalist base first, then add a stem you can stand studying.
  • Transferable craft (writing, production, finishing) outlasts fashions.

Further learning

Charity Majors' Engineer/Manager Pendulum talk:

The Engineer/Manager PendulumInfoQ

Next: evaluating startup offers.