# How AI is changing my job: Tracer bullets

> As implementation gets cheaper, integration, validation, restraint, and judgment get more valuable.

_Published May 12, 2026 by Vinicius Brasil._

---

AI is rapidly changing software engineering, and with any kind of
technological evolution, we need to sort the wheat from the chaff. On one side,
Claude Code's creator Boris Cherny [says that "coding is largely
solved"](https://www.youtube.com/watch?v=We7BZVKbCVw). On the other, [devs like
Mario
Zechner](https://mariozechner.at/posts/2026-03-25-thoughts-on-slowing-the-fuck-down/)
argue that real software will always require human judgment, restraint and
sustainable pace.

Amid all the signal and noise, there's you and me: engineers who heavily use AI
in their day-to-day, but aren't swept up in the frenzy.

First, let me clear things up about my view on AI assisted software
engineering:

- I do use Claude Code.
- It has made me more productive.
- I get frustrated with it sometimes.
- I'm constantly experimenting with new workflows and techniques.
- I also see engineers using these tools carelessly, creating growing piles of
  tech debt that hopefully Claude Opus 3001 can fix.

Philosophical arguments aside, let's get to the code: how AI is changing my job
as a Staff Engineer.

## Tracer bullets

Now that AI does the typing, engineers have more time to think about other
parts of the stack: product, infrastructure, performance, databases. This is
where **tracer bullets** come in handy, as opposed to proof of concepts or
prototypes.

Imagine you are building an online multiplayer game. Instead of prompting an AI
with, "As an experienced Rockstar Games developer, build GTA VI and don't miss
the deadline," you start with a tracer bullet: two cubes in an empty map. One
player moves, and the other player sees that movement in real time over the
network. No graphics, no missions, no physics engine. Just the smallest
possible version that proves the core system works end-to-end.

A tracer bullet differs from a PoC or prototype: its goal isn't to look
impressive or prove a big idea, but to verify the full path works end-to-end.
As a rule of thumb:

- Tracer bullets optimize for stack integration
- PoCs optimize for feasibility
- Prototypes optimize for visualization.

None of this is new, by the way. I remember when I first read The Pragmatic
Programmer, Clean Code, and Mythical Man-Month. These books may feel distant in
a fast-changing industry like ours, but they teach lessons that are even more
important now. In fact, there's even a [dedicated AI agent skill for The
Pragmatic Programmer](https://www.skills.sh/wondelai/skills/pragmatic-programmer)
that includes tracer bullets.

AI is not replacing software engineering. It is changing where engineering
effort is spent. As implementation becomes cheaper, integration, validation,
restraint, and judgment become more valuable.

This is a series on what AI is actually doing to software engineering. We're
all figuring this out together. If that's useful to you, follow along.