Modern development environments are speedrun simulators. Autocomplete fills in your functions, AI chats debug your code, and Slack is never more than three seconds away from shattering your concentration. The tools work. The work, increasingly, does not.

What used to be long sessions of focused engineering are now a mosaic of micro-interruptions. On paper, productivity is up. In practice, we are losing something harder to quantify: focus, endurance, and independent reasoning.

The Dopamine Drip of AI-Powered Work #

It starts small. A suggestion from your AI assistant here. A ChatGPT query there. A Slack ping in between. The moment-to-moment flow of work gets chopped into fragments, each one punctuated by a helpful tool that resets your mental state like a browser tab crashing and reopening mid-scroll. Collectively? A distributed denial of deep work.

You sit down with purpose. Your AI assistant, like an overeager lab partner, offers a solution before you’ve finished framing the problem. It seems plausible, even helpful. You take it. Then you wonder about another part of the code, so off you go to consult the oracle. As the answer materializes, a coworker drops a fire emoji on a message from fifteen threads ago. You respond, of course. When you return to your editor, your stack trace is gone, and so is the thought that led you there.

Multiply this cycle across a full workday, and your mind ends up cluttered with what psychologists call “attention residue.”¹ Each interruption leaves behind a fragment of unresolved context, clouding the next task and dulling your mental edge. It often takes more than twenty minutes to recover full focus. Over time, the damage compounds. Your day becomes a chain of shallow thoughts stitched together by interruptions that feel urgent but rarely are.

Offloading Thought, Losing Practice #

There is also a deeper cost that is harder to spot in the moment. AI tools are not just distracting. They are thinking for us. And just like people who grew dependent on GPS became worse at spatial reasoning, engineers who outsource too much problem-solving risk losing their fluency in the craft.

This has consequences. Not dramatic ones. Just quiet erosion. The kind that turns senior engineers into high-functioning rubber ducks and junior devs into plugin operators. You forget syntax. Then architecture. Then why you liked writing code in the first place.

And no, it’s not just you.² Studies are catching up to the anecdotal vibe: offloading too much thinking to machines makes you worse at thinking. The curve is subtle, but the trend line isn’t flattering.

It does not arrive all at once. Instead, it creeps in over time, as you gradually stop doing the hard parts of the work yourself. Why memorize function signatures or wrestle with architectural trade-offs when a chatbot can suggest something that looks about right? It feels efficient in the moment. But the cost is paid in rusted instincts and faded mental maps.

The Illusion of Speed #

Perhaps the strangest irony is that AI often makes us feel faster while quietly slowing us down. In a recent study, experienced developers using AI assistants were 19 percent slower at completing familiar tasks.³ They had expected to move quicker. What they encountered instead was a series of partial suggestions and plausible but imprecise code that had to be vetted, adjusted, and reworked.

The process felt easier. And because it felt easier, it felt faster. But it was not.

This is the illusion AI creates: reduced cognitive strain masquerading as productivity. Editing someone else’s paragraph usually feels lighter than writing your own, even if the result takes longer. The same applies here. We accept imperfect answers from machines because they save us the discomfort of thinking deeply from scratch.

That’s the trap. Ease is not the same as effectiveness. Just because something feels smooth doesn’t mean it’s the best route. Sometimes, the extra effort was the actual work.

When Critical Thinking Becomes Optional #

The risk goes beyond convenience. It touches our judgment. You see a suggestion. It looks right. You click accept. It might be fine. It might also contain the sort of subtle issue that gets mentioned in a post-mortem four weeks later.

More and more, engineers report deferring to AI suggestions not because they trust them, but because they seem confident. The output is often polished. It arrives quickly. And it saves you from having to stare into the void of a blank file or an ambiguous design problem.

But this passivity is dangerous. Studies show that people are more likely to accept AI-generated recommendations, even when those recommendations contradict their own knowledge.⁴ The presence of a machine in the room changes how we weigh ideas. We stop asking “Is this right?” and begin asking “Could this be good enough?”

Once that threshold shifts, you are no longer reviewing suggestions. You are approving them by default.

The Case for Deliberate Friction #

This does not mean we should abandon our tools. Far from it. Generative AI offers real leverage, especially for repetitive tasks or quick lookups. But we need to reintroduce friction into our workflows, because friction is where understanding lives.

Deep work does not survive in 40-second increments. It needs breathing room. It needs silence. It needs your brain to stop multitasking like a hyperactive squirrel.

Some developers might schedule AI-free focus blocks, giving themselves time to engage in uninterrupted, original thought. Others batch their questions and AI interactions, resisting the urge to pepper them in real time. A few try deliberately analog exercises, like sketching out designs by hand or solving problems without opening a browser. These are not acts of nostalgia. They are acts of mental preservation.

The goal isn’t to reject AI. It’s to remember that you are allowed to think.

Final Thought #

We aren’t becoming less capable. We’re becoming more incentivized to stop trying so hard. AI offers speed, convenience, and a dangerously appealing alternative to struggling. But struggling is where understanding lives. It’s where intuition grows. It’s where craftsmanship hides.

AI should be a co-pilot, not the driver. You don’t learn to navigate by sitting in the passenger seat with a map you never read.

So maybe the next time your AI assistant pipes up with a suggestion, you say, “Not now, I’m thinking,” and see if your brain still remembers how to stretch.

Will it be slower? Probably. Will it be messier? Definitely. Will it be yours? Absolutely.

And that is still worth something.

Of course, all of this assumes you’re trying to focus. Vibe coding is another story entirely. That strange and joyful state of frictionless improvisation, fueled by instinct and rapid feedback, might just be the one domain where the chaos works in our favor. But that, perhaps, is a different kind of deep work, and a different essay altogether.