Skill Depth Versus Breadth, Revisited With Agents

The old debate was: should you be a specialist or a generalist? AI agents have rewritten the question. The right answer in 2025 is different from the answer in 2020. Eight months of daily shipping with agents has given me a sharper view.

Why the Old Framing Is Broken

Before agents, depth was expensive. It took years to become genuinely productive in a specialty. Breadth was shallow by necessity: you could not be expert in seven things. So the advice was 'pick one, go deep, stay relevant'.

Agents change the economics. Depth in execution is now cheap: the agent writes correct Lambda code whether I have written fifty Lambdas or five. What stays expensive is depth in judgment: knowing when Lambda is the wrong tool. That is the new bottleneck.

The New Shape

Breadth in judgment, depth in one or two specialties. That is the shape that wins now. Breadth lets you choose the right tool for the job. Specialty depth lets you trust your judgment when the agent and the docs disagree. The rest is execution, which agents cover.

I have depth in serverless Python and backend architecture. I have breadth across frontend, WordPress, and content systems. The combination ships client work faster than a specialist in any one of them could.

Measuring Your Shape

plaintext

For each domain:
  can I scope a real project in it?            &lt;- breadth check
  can I recognize when the agent is wrong?     &lt;- depth check
  can I debug a production incident in it?     &lt;- depth check

Breadth means scoping. Depth means debugging. These are the two skills agents do not cover.

The Trap to Avoid

Over-indexing on execution depth now is a trap. You will end up with world-class skill in something the agent also has world-class skill in. Your value becomes redundant. The compounding returns are in the judgment layer, not the keyboard layer. See The Protean Career for a pre-AI version of this argument about flexible specialization.

The Honest Self-Assessment

Once a quarter, I ask: what did the agent do this quarter that I could not have done alone? If the answer is 'everything', my depth is atrophying. If the answer is 'nothing', my breadth is atrophying. The healthy shape is a mix on both sides.

Breadth in judgment, depth in taste. That is the thesis.

The Practical Reshuffle

I spent most of my learning hours this year in two specific directions. First, deeper on serverless architecture, because that is where my judgment payoff is highest. Second, broader exposure to adjacent layers like frontend caching and email infrastructure, where I did not need implementation depth but needed enough breadth to scope projects responsibly. Neither direction is the path the 'go deep on one thing' crowd recommends. Both turned out to be correct for the current economic shape of development work.