AI has allowed many engineers, myself included, to execute faster and deliver more within our current domains than ever before. While this brings immense value to the profession, there is a hidden cost. When we use AI to branch into projects or tasks we don’t fully understand, it is easy to lose our situational awareness.

Why “more with less” works

Consider a hypothetical scenario: You are an engineer managing a backend service in Python. You know Python well, and you’ve spent enough time in the codebase to understand its architecture. With AI, you can ship features faster, squash bugs more efficiently, and accelerate your overall workflow.

Because your foundational knowledge is solid, you can easily review the code the AI spits out. It fits neatly into your existing mental model. Your understanding of how the service works remains fully up-to-date.

The trap of “more with more”

Now, let’s extend that same hypothetical scenario. Your team recently inherits ownership of a different backend service, this time written in Go. You do not know Go. You do not know this service. Your mental model of the system is essentially zero.

Feature requests start piling up, and by leaning on AI, you manage to hobble through them. The problem? Your understanding of the service never actually catches up to the code you are shipping. Because the AI is doing the heavy lifting in a domain you don’t grasp, your mental model remains broken. You are moving fast, but you are flying blind.

AI doesn’t eliminate cognitive load

The example above might seem a bit exaggerated, but the reality sets in when you multiply it by the sheer number of services an individual is expected to understand. AI makes it incredibly easy to go deep into a few familiar services. But the cognitive overhead, context switching, and ramp-up time required to manage many different services haven’t disappeared.

AI can easily produce code for a dozen unfamiliar systems, but it becomes harder and harder for the human brain to process and retain what is actually happening under the hood in each of them.

Smaller teams, same scope

The corollary to all of this is that a small, focused team armed with AI can make immense progress on a few services while retaining all the necessary context and state. This approach pays dividends across the board: it drives greater individual ownership, ensures engineers maintain a holistic view of the system, and significantly lowers communication and process overhead.

Conclusion

AI is an incredible tool for deepening our impact, not necessarily for expanding our plate. To get the most out of it without sacrificing engineering quality, we need to keep teams small and focused on a few core services.