Why optimising tasks in isolation makes knowledge work exhausting

Executive overview

Knowledge work keeps getting more draining even as tools improve. The culprit is the isolated optimal mindset: evaluating each request in isolation and asking "what's the optimal thing to do here?" — without considering the full human context.

Computer security researchers faced the same trap with password rules and solved it by shifting to human-centric security. The same shift fixes broken email and meeting culture.

The fix is not optimising individual tasks — it's designing workflows around the actual experience of the humans doing them.

The isolated optimal mindset

• Looks at a single behaviour in isolation and asks: what is the optimal response?
• Concludes: answer emails instantly (it only takes 3 minutes); accept all meeting invites (your time was free anyway).
• Misses cumulative cost: 200 context-switches per day, no recovery time, inequitable — only those who can work nights can keep up.
• Password rules are the canonical example: technically correct, practically ignored because the human overhead is too high.

Human-centric security — the source of the fix

• A new subfield that watches real people in real contexts rather than reasoning about optimal behaviour in the abstract.
• Finding: users don't struggle because they can't create complex passwords — they fear forgetting them across multiple devices and systems.
• Solution: meet people where they are — pre-install and train password managers, remove passwords entirely where possible.
• Lesson: what's optimal for the task is often damaging for the human.

Applying a human-centric lens to work

• Email: viewed in isolation, instant replies are optimal; viewed holistically, they create cognitive fragmentation and constant anxiety from the growing inbox.
• Meetings: auto-scheduling feels efficient for the organiser; it destroys the receiver's ability to do sustained work.
• Alternatives (office hours, docket-clearing meetings, standing pre-scheduled slots) are less locally optimal but dramatically better for the human experience overall.
• Goal: work that is effective and sustainable for humans, not maximum throughput on isolated tasks.

Book update — writing voice and information density

• Rewrote the first half of the new book after finding the original voice too dense and culturally self-referential.
• New voice: speaks directly to a reader who broadly wants to improve their life, skips the ideological debates.
• Design principle for the book: every chapter should contain enough ideas to fill a standalone book — maximum compression, not padding.
• Chapter on time management reframed around why people abandon systems even when they want them to work.

Q&A highlights

• Big projects vs. small tasks: David Allen's next-action approach works for simple tasks; complex projects require sustained cognitive blocks, not interleaved micro-tasks. Use multi-scale planning (quarterly → weekly → daily) to protect that time.
• AI and career capital: for most knowledge workers, AI in its current chatbot form is not yet disrupting jobs. Prompt engineering is likely a temporary skill. Wait until the disruption vector is visible in your specific field before adapting.
• Saying no: face your "productivity dragon" — list everything you're responsible for, recognise it's too much, then reduce deliberately. Valued, reliable people have more latitude to say no than they think.
• Kanban across an organisation: works well at team scale (≤ ~6 people); keep it simple — centralised task visibility, WIP limits, brief check-ins. Avoid over-engineering with complex rules.
• Time blocking for nurses: not applicable — nursing shifts are externally structured, not autonomously managed. Sustainable healthcare work requires reducing system friction (e.g., EMR overhead, realistic patient loads), not personal scheduling tools.
• Task boards and projects: task boards work best at single-session granularity. Bigger projects live in quarterly plans; only put project sub-tasks on the board when you need a reminder of what to tackle that week.

AGI vs. superintelligence — clarifying the distinction

• AGI (artificial general intelligence): a subjective threshold where AI does what it already does — text, images, research — at roughly human level. Not a binary event; systems are already near-human at some tasks.
• Economic and security impact of AGI: real but incremental — comparable to the powered loom displacing textile workers, not a sci-fi rupture.
• Superintelligence: a fundamentally different concept requiring artificial consciousness, autonomous self-improvement, and recursive capability gains. Still speculative; the core assumptions (consciousness is achievable, recursive improvement is unbounded) have no empirical basis.
• Conflating the two creates unnecessary panic. Crossing the AGI threshold does not unlock new capabilities — it means existing capabilities get measurably better.