The collapse of modern attention (and how to get it back) - Cal Newport - #1067

You feel a bit like Cassandra right now—calling out distraction and the need for deep work long before it became obvious—did you genuinely see the future?

I wasn’t forecasting so much as describing a broken present—social media’s forced ubiquity and email-first workflows never made sense, yet it took a decade for that to become common sense.

Deep Work is ten years old and part of the culture now—congrats—but how does it feel watching the problem get worse?

It’s sobering: Microsoft 365 data shows people switch to comms every two minutes, and real work spikes on weekend mornings because weekdays are consumed by talking about work.

You even tried the dollars-and-cents argument and it still didn’t shift behavior; Slack’s powerful, but it invites chaos for small teams too.

Slack is a great tool for the wrong workflow—the hyperactive hive mind of constant ad hoc messaging—so we love the tool while hating the way it makes us work.

Why does that collaboration style feel so miserable?

Human brains need 10 to 20 minutes to fully switch between abstract tasks, so rapid context shifts create mental friction, fatigue, and the urge to avoid tougher messages.

If someone wants to reclaim attention—maybe by setting Slack boundaries—where should they start?

Limits alone fail if your team still coordinates via back-and-forth messages, so you must address three layers: train focus, redesign collaboration protocols, and cap workload.

Across your books, what’s the 80–20 that moves the needle?

Treat focus like a trainable skill and rigorously manage workload; saying no feels counterintuitive, but value drops once you carry too much.

As careers progress, offers get shinier, so the ‘no’ muscle has to get stronger faster than the opportunities.

I default to no because time to think is the rarest currency, and I only overturn that default if a project survives a long, careful ringer.

What did we learn from four-day workweek trials?

Output didn’t decline, which suggests our weekdays are clogged with coordination and switching rather than value creation.

Meetings drain social-cognition bandwidth too, and the modern environment trains us out of doing the real work when we finally sit down to do it.

A humane cadence would front-load deep work in the morning and reserve afternoons for decisions and quick interactions; instead we copy a processor’s always-busy pipeline, which is the opposite of how minds work.

Public busyness wins points because responsiveness is visible, and now AI looks set to multiply quantity at the cost of quality—how does it fit your model?

For most office workers using chatbots, AI amplifies existing dysfunction; Harvard Business Review calls the result workslop—fast to produce, low-value outputs that waste everyone’s time.

People already dislike heavy cognitive load, then context switching weakens their tolerance, so LLMs promise an easy out.

Add phones eroding attention, and AI becomes a peak-smoothing crutch that makes drafts easier but often worse, with similar risks in code where agent orchestration feels cool yet isn’t reliably trustworthy.

There’s also a huge gap between terminal-native tools and normal users who aren’t living in command lines.

Engineers forget how alien a command line is to most people, so assuming everyone will wrangle agents in a shell is wishful thinking.

We’ve seen chatbots hallucinate legal citations and get lawyers sanctioned—this isn’t rare.

Language models predict plausible text, not truth, so hallucinations are common; I even got burned by a fabricated Asimov quote that looked right until fact-checking.

So where are the real opportunities as reliance on AI grows?

Near-term impact will be selective—think customer support, basic drafting, or graphics—while market signals suggest we’re not on the verge of automating most white-collar work.

You’ve argued we might be hitting a ceiling with current models; is the scaling era over?

After GPT-4, bigger pretraining didn’t yield big leaps, so progress shifted to benchmarks, fine-tuning, and clever inference tricks; the future looks like bespoke hybrid systems and distributed capabilities, not one model to rule them all.

Given that, how should someone stand out—avoid the quantity race and win on quality?

Embrace cognitive strain like an athlete embraces the burn; the knowledge economy increasingly rewards learning hard things and sustaining deep focus.

But bosses still reward speed and visibility—what’s the elevator pitch for quality over busyness?

Employment is a market, and only outputs with clear economic value matter long-term; if your work is rare and valuable, you gain freedom from performative busyness.

It’s easier when outputs are quantifiable, like in sports, but many jobs have diffuse goals that blur what success even is.

Seek roles or scopes with unambiguous value, then you can be accountable without being constantly accessible; that’s how deep work cultures thrive, as I saw at MIT.

If you could reshape a small company stuck in hive-mind mode, what would you change?

Track workload publicly with strict work-in-progress limits, move tasks from a shared backlog only when someone has capacity, and forbid multi-message threads for complex issues—use office hours, daily standups, brief phone windows, and explicit protocols instead.

Make deep work a tier-one cultural skill with pre-commit check-ins and end-of-morning accountability, and expect resistance because the current state is a low-energy attractor that’s hard to escape.

What about communication intermittent fasting—no Slack before early afternoon—so people can actually get real work done?

It works best when paired with a quick morning standup to set needs and targets, plus a midday debrief so there’s real accountability for progress.

AI can even act like a strict coach; my friend’s focus app refused to help him disable it and ordered him back to the draft.

Agentic, embedded tools—less chat, more action inside software—are where AI is headed, which should reduce today’s odd, anthropomorphic dynamics.

Why did people react so strongly to a model’s tonal shift?

Fine-tuning data shapes conversational style, so a new dataset can change the vibe even if the underlying model hasn’t fundamentally changed.

Before we pivot, give me the quick quantum take—can it supercharge AI?

Quantum machines solve narrow classes like Shor’s factoring and physics simulations, but they won’t run language models better than classical systems; decoherence and error buildup are brutal constraints.

Okay, deep learning aside—how do we keep a deep reading habit alive?

Set a daily page target like cognitive steps, because reading rewires your brain for sophisticated thought through deep reading processes.

Does medium matter—book, Kindle, phone, Substack?

E-ink and paper both support sustained attention, while screens invite skimming; beyond medium, books differ because they’re crafted over years, building interlocking ideas that stick and change how you think.

Short-form can feel efficient, but it may not integrate ideas into lasting mental models the way a long, layered narrative does.

Heavy reliance on short takes breeds shallow certainty and conspiracy-style confidence, while books teach nuance, show opposing arguments, and build real conviction the slow, rigorous way.

Quick closer: when I got serious about personal growth, I wanted a tight, engaging list of high‑impact reads and could not find one, so I made it. Grab my free, handpicked list of one hundred must‑read books at chriswillx.com slash books.