The Death of Deep Work in the AI Era

Deep work needs two things: a hard problem and a human who will not leave the chair. AI threatens the second one, not the first. Hard problems are still hard. The bottleneck has always been the person, and the person now has a tempting escape hatch three seconds away.

This post is not a Cal Newport rehash — it is about the specific flavour of concentration loss that AI has introduced, and what to do about it in practice, not in theory.

Why AI threatens deep work specifically

Shallow work has always been seductive. Email, Slack, meetings — these have always eaten focus. What is new is that AI provides a fourth escape hatch that masquerades as deep work. You feel like you are being productive, because outputs are coming out, but you are actually in a rapid prompt-tab-copy-paste loop that looks like focus and is not.

The prompt-response rhythm is short. Deep work requires a long rhythm — thirty, sixty, ninety minutes with a single problem. If your default rhythm is fifteen seconds because that is how long a prompt takes, you have trained yourself out of the long one.

What deep work actually buys you

I want to make the case concretely. Almost everything meaningful in my career came from long blocks of uninterrupted thinking, not from fast iteration.

• The architecture decisions that saved months of rework
• The essays that actually mattered to readers
• The product insights that changed roadmap
• The debugging sessions that fixed the real bug, not a symptom
• The strategic pivots that stopped us from wasting a quarter

None of these came from shallow work. All of them came from sitting, uninterrupted, with a problem. That is the output pattern I want to protect.

Protect one 90-minute block a day

Not two hours, not a whole morning — one 90-minute block. This is low enough that it is doable on every day, including bad days. High enough to produce real work. The specific number matters less than its non-negotiability.

The setup for the block is deliberately spartan. Shallow work is happy to expand into any available surface. Give it no surface.

• One tab, no AI, one problem
• Phone in another room, not on silent
• Notes in a plain text file — no autocomplete, no AI suggestions
• A written objective at the top — "by the end of this block, I want to have X"
• End with a written summary of what you learned — not just what you did

What to do in the block

Actual deep work. Which means: thinking, writing longhand or in plain text, reading primary sources slowly, drawing diagrams, doing the hard part of the hardest problem you are stuck on. Not "preparing for deep work." Not "setting up a good environment for deep work." The work itself.

If you hit a wall — and you will — do not reach for AI. Sit with the wall. The wall is what teaches you. If you get unstuck by asking a model, you skip the learning that happens when you unstick yourself. Twenty minutes of staring is worth more than twenty seconds of a clean AI answer, most of the time.

How to make it stick

• Same time every day. Morning is usually best, before the day gets busy
• Tell the people you work with about the block. Defend it socially
• Track streaks. Miss one day, fine. Miss two, fix the pattern immediately
• Measure output per block in "did I learn something new?" — not in keystrokes or tokens

Nothing useful in my career came from fast. It came from long.

Deep work is not dead. It is endangered. In two years, being able to sit with a problem for ninety minutes without breaking concentration will be a rare, paid skill. That is both depressing and an opportunity. Pick up the opportunity.