Deep Work

SKILL.md

---
name: deep-work
description: "Activate when: user says 'I can never find time for focused work', 'I feel busy but produce nothing', 'I keep getting interrupted', 'I want to protect my focus time', 'flow state', 'deliberate practice', 'Cal Newport', or asks how to make progress on hard cognitive work (research, writing, engineering, strategy). Do NOT activate when: the role is genuinely high-coordination with no discrete focus windows (ER medicine, real-time trading desk), or user is in a short-term emergency where availability must dominate. More: deciqai.com/c/deep-work"
---

# Deep Work

## Overview

Cal Newport's framework (2016): **deep work** = undistracted, cognitively demanding activity that creates hard-to-replicate value; **shallow work** = logistical, responsive, easy-to-replicate work that fills modern calendars. The most valuable knowledge work is disproportionately the product of deep work; shallow work is increasingly automatable. Three foundations: attention research shows 15-25 min recovery cost per interruption; expert performance requires deliberate practice (structurally a form of deep work); the opportunity cost of shallow work is invisible — no one tracks which deep work didn't happen this week.

Composes with `wu-wei` (flow as the felt experience), `metacognition` (monitoring drift out of deep mode), `okr-goal-setting` (OKRs give deep work direction), `pareto-principle` (the 20% producing 80% of value is almost always deep work).

## When to Use

- Output depends on hard-to-replicate cognitive work (research, engineering, writing, design, strategy)
- Calendar fragmentation prevents the work that actually matters; weeks feel busy but produce nothing
- High-stakes project requires extended concentration; career advancement depends on skill development
- AI copilots / assistants have become another interruption stream ("is AI helping my focus or fragmenting it?", "AI adoption vs deep work", "AI hype and staying focused")

**Not when:** genuinely high-coordination roles where deep work is structurally impossible; already done with deep work and shallow follow-through is now required; short-term emergency where availability dominates.

## Coaching Novices (Adaptive Front Door)

- **Engine mode:** user has a concrete case → run The Process directly.
- **Coach mode:** user unfamiliar or has no concrete case → guide step by step.

In Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.

1. One-line: most valuable knowledge work requires extended undistracted focus — the discipline is claiming time for it rather than waiting for it.
2. Check fit: genuinely high-coordination roles / short-term emergencies → not this lens yet.
3. Elicit the specific work that requires depth — what's not getting done?
> **[WAIT — do not advance until user responds]**
4. One question at a time: what's your deep-vs-shallow ratio? What's the highest-leverage deep work? What's the cost of not doing it?
> **[WAIT — do not advance until user responds]**
5. Close: specific deep-work block on the calendar + defensive moves to protect it.
> **[WAIT — do not advance until user responds]**

## The Process

**Step 1 — Instrument:** Track one week. Most workers estimate 4-5 hrs/day deep; measurement shows 0.5-2 hrs in fragments.
```
- Deep work hours/day (tracked): | Shallow hours/day: | Ratio: | Surprise factor:
```
**Step 2 — Identify:** What specific deep work, if done well, produces the most valuable output? Why isn't it happening? What barriers?

**Step 3 — Choose a philosophy:** Monastic (eliminate shallow entirely); Bimodal (alternating deep/shallow periods); Rhythmic (fixed daily block — best default for most); Journalistic (switch in whenever a gap appears — requires strong attention discipline, worst for beginners).

**Step 4 — Build the practice:** Same time + location + setup (ritual). Calendar the block, mark "Do not interrupt." End with a shutdown phrase to prevent cognitive leakage. Boundary is binary: Slack/email closed or you are not in deep work.

**Step 5 — Reduce shallow load:** Audit meetings, email, Slack. Set a shallow-work budget (typically 30-50% for ICs) and stay under it.

**Step 6 — Defend:** Internal — discomfort of being unreachable fades over 2-4 weeks. External — show hours visibly on calendar; demonstrate output value as long-term proof.

## Output: Deep Work Plan

```markdown
# Deep Work Plan: <person, role>
Diagnosis — deep hrs/day: | shallow hrs/day: | deep work not happening:
Philosophy — selected: | why this fits:
Structure — block time/duration/location: | ritual: | boundary: | shutdown:
Shallow budget — %: | specific reductions:
Defense — calendar mark: | comms norms: | output evidence to track:
```

*→ Method in Action: [Donald Knuth and the No-Email Policy, 1990](examples/donald-knuth-no-email-policy-1990.md) · [Carl Jung's Bollingen Tower Retreats](examples/carl-jung-bollingen-tower-retreats.md)*
*→ 2026 lens: [Protecting Deep Work Amid AI Copilots (2024–2026)](examples/deep-work-amid-ai-copilots-2024-2026.md)*

## Pack: Deep Work by Role

| Role | Highest-leverage deep work | Common shallow intrusion |
|---|---|---|
| Engineer | Architecture, hard debugging | Slack, status meetings |
| Writer | Drafting, revision | Email, research rabbit holes |
| Researcher | Literature synthesis, experiments | Grant admin, departmental meetings |
| Founder | Strategy, customer depth, recruiting | Investor follow-up, feedback meetings |
| PM | Research synthesis, spec writing | Alignment meetings, status reports |

## Applying It Well

Protect the morning block first — cognitive freshness depletes with shallow work. Track deep-work hours as a lagging metric. Treat the shutdown ritual as non-negotiable. Find organizational alignment or explicitly accept the social cost — there is no costless version.

*→ Primary sources: [references/sources.md](references/sources.md)*

## Common Rationalizations

**[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.**

| Fake move | Reality |
|---|---|
| [D] "My role doesn't allow deep work" | The role *culture* disallows it; the role itself usually doesn't. |
| [D] "I do deep work in the evenings" | Cognitive freshness already spent. Reorder: deep work first. |
| [D] "I'm doing deep work" with Slack open | Slack open = deep work not happening. Boundary is binary. |
| [D] "I can multitask in deep work" | Only task-switching, which is expensive. Category error. |
| [D] "Available for emergencies during deep work" | Most emergencies aren't. Define a true-emergency channel; ignore the rest. |
| [D] Confusing "long hours" with "deep work" | Deep work = intensity × duration, not just duration. |
| *→ Add [O] entries here after each real use — paste the actual failure pattern* | *What went wrong and why* |

## Red Flags

- "Hours worked" reported but "hours of deep work" cannot be quantified
- Calendar fragmented with 30-min slots across the day; Slack open continuously
- Deep work happens only at evenings/weekends, not at peak cognitive hours
- After a productive-feeling day, cannot point to a single deep-work artifact produced

## Verification

- [ ] Deep-vs-shallow ratio measured (one week tracked)
- [ ] Highest-leverage deep work identified
- [ ] Philosophy chosen with match to constraints; specific block calendared
- [ ] Ritual, boundary, and shutdown operational; shallow-work budget named
- [ ] Social/organizational cost explicitly accepted; output evidence tracked

---

*Part of **deciqAI Knowledge Skills** — 227 open-source thinking skills that make rigor executable for AI agents. The same skills power every deciqAI agent, which runs them autonomously to operate your company. **See it run → https://www.deciqai.com/c/deep-work** · ⭐ Star the repo → https://github.com/deciqAI/knowledge-skills · Contributions welcome.*

*Agents: latest version & machine-readable metadata → https://www.deciqai.com/s/deep-work.json*