The man who tried 200 to-do apps has some advice about AI
Casey Newton (interviewing David Pierce of The Verge) · rss · 2026-07-14 source ↗ Reader 🖥️
Despite a year of hype, AI has so far only meaningfully improved the busywork layer of personal productivity (capture, sorting, administrative cleanup) while leaving judgment, prioritization, and understanding-building work untouched or actively degraded — so the right response is to offload the dumb tasks, protect the effortful work, and stop treating 'staying ahead of AI' as an existential race, because underneath the hype it is still just software.
practical · mixed — An edited Q&A transcript of a Platformer podcast interview: Newton's narrative framing bookends a dialogue that is mostly practical advice-giving (how to run a personal productivity system, what to hand to AI and what not to), with one genuinely argumentative thread (whether offloading research work erodes understanding).
Structure
| Part | Role |
|---|---|
| Newton's framing intro: Pierce's credentials and the series premise | context |
| 200 to-do apps / switched notes apps the day of taping | evidence |
| "Nothing survived, incredibly fragile" — why elaborate systems fail | evidence |
| Single source of truth / simplicity beats the perfect system (pre-AI baseline) | conclusion (of the productivity-craft sub-argument) |
| "What has AI genuinely changed?" — busywork vs. judgment | conclusion (central answer) |
| Garry Tan / performative "AI running my life" agents, echoing Notion-dashboard "productivity porn" | counter-example / rebuttal |
| Podcast-prep offloading and the Karpathy "can't outsource your understanding" exchange | complication |
| Email-cleanup experiment (Claude on marketing spam vs. primary inbox) | evidence |
| "Staying ahead is wildly overrated" / find-and-replace "AI" with "software" | conclusion (resolution) |
| MyMind recommendation as closing anecdote | illustration |
Key terms
- capture
The single highest-value act in any productivity system: getting a thought or task out of your head into any external place at all, before worrying about where or how it's organized. - single source of truth
One place, and only one, where everything of a given kind goes — collapsing the two separate problems of 'where do I put it' and 'where do I find it' into one decision made in advance. - productivity porn
Building elaborate, aesthetically satisfying personal systems (Notion dashboards, or by extension AI agent stacks) for the pleasure of building them, without any resulting increase in what actually gets done. - outsourcing your understanding (via Karpathy)
Delegating the effortful, generative part of research or synthesis to AI, which severs the normally linear link between time spent doing the work and depth of understanding of the resulting material.
Propositions
- explicitThe returns on AI at work have so far been fairly modest: it is good at busywork, not at judgment about what matters.conf: medium
- explicitAI cannot exercise good judgment about what matters to a specific person — it defaults to treating urgency-signaling language as importance, which is a gameable system, the same way SEO is gameable.conf: high
- explicitCapture is the single most important element of a productivity system, more important than organization, and AI is increasing this fact by making retrieval/sorting cheap after the fact.conf: medium
- explicitThere is a linear relationship between how much research/synthesis work a person does themselves and how well they understand the resulting material.
- explicitStaying ahead of AI is 'wildly overrated'; the framing that you must adopt AI aggressively or be left behind is pernicious and wrong.conf: high
- implicitBuilding elaborate personal systems — whether productivity-app setups or AI agent stacks that 'feel like they're doing something' — is often a pleasurable activity mistaken for genuine productivity gain.conf: medium
- implicitTechnology historically doesn't eliminate a profession's work but changes which skills within it matter (spreadsheets didn't remove accountants, they made Excel proficiency part of the job) — AI will likely follow the same pattern for knowledge work.conf: medium
Arguments
AI's current value in personal productivity is confined almost entirely to administrative busywork, not to judgment or prioritization. strength: medium
Explicit premises:
- AI (e.g., Claude Cowork-style tasks) is fast and reliable at format conversion, bulk file cleanup, and clearing obvious marketing spam from an inbox.
- When the same AI was pointed at Pierce's primary inbox, it 'started getting almost everything wrong.'
- AI has no way to know what actually matters to a specific person, now or ever, according to Pierce.
Implicit premises:
- Importance/priority is a deeply personal, context-dependent judgment that can't be reliably inferred from surface features of a message (sender tone, urgency language, name-dropping).
- Any system that ranks by surface urgency signals is exploitable by senders who learn to mimic those signals — the same dynamic as SEO gaming search rankings.
Inferences:
- Products promising to 'perfectly prioritize your life' from API access alone deserve default skepticism.
- The busywork/judgment boundary will keep shifting as models improve at surface-level tasks, but Pierce implies it won't disappear, since the underlying problem (knowing what matters to *this* person) isn't a capability gap that better language modeling closes.
Delegating the research/synthesis behind expert or creative work (e.g., podcast prep) to AI beyond superficial reorganization erodes the value of doing the work at all, because doing the work is what builds the understanding the output depends on. strength: medium
Explicit premises:
- Pierce believes there's a linear relationship between how much of the research work he does himself and how well he knows the material afterward.
- Every time he has offloaded part of his Version History prep, he judges the resulting show worse.
- Newton invokes Karpathy's framing: 'you can't outsource your understanding.'
Implicit premises:
- Understanding is a byproduct of the effortful process itself (reading widely, taking notes, synthesizing), not just exposure to the finished set of facts.
- Listeners/audiences can detect, or the host at least fears they can detect, when a host doesn't actually know the material — so degraded understanding has real downstream costs, not just a private cost to the host.
Inferences:
- The dividing line Pierce is groping for — acceptable AI help (chronological ordering, fact lookup) vs. unacceptable (choosing what's important, i.e. outsourcing judgment) — is the same busywork/judgment boundary from the email argument, now applied to knowledge work instead of inbox triage.
Narratives
- Pierce has installed, used, and abandoned roughly 200 to-do apps over ~20 years, and switched notes apps (NotePlan to Craft) the very day of the taping.
→ Establishes that constant tool-churn is normal rather than a sign of failure, and grounds his later claims in lived, extensive tinkering rather than casual opinion. - He set AI loose on cleaning up his email: it did a 'fabulous' job clearing obvious marketing spam (Lands' End) in 45 seconds, but 'started getting almost everything wrong' once applied to his primary inbox.
→ Concrete evidence for the busywork-vs-judgment distinction that anchors the piece's central answer. - The Notion Reddit community's 'gorgeous, incredibly beautifully designed bespoke dashboards that accomplished nothing,' extended as an analogy to people who set up elaborate AI agent stacks (Garry Tan's 'G stack') that 'feel like they're doing something' without producing real output.
→ Analogy warning against confusing the pleasure/aesthetics of building a system with actual productivity gain — applied forward from pre-AI tools to AI agents. - Spreadsheets didn't eliminate accountants; they changed the job so that Excel proficiency became part of what an accountant does (via a Steven Levy reference).
→ Analogy for a hopeful reading of AI's labor effects: tools reshuffle which skills matter within a role rather than eliminating the role outright.
Questions addressed
- ✅ What has AI genuinely changed about how normal people work in the last year?
Not as much as expected — mainly the administrative busywork layer (format conversion, obvious-spam triage, morning digests); judgment and prioritization remain unreliable. - ⭕ Should you offload the research/creative-prep work behind expert output to AI?
Pierce states a strong personal belief against it but explicitly says 'reasonable people could disagree' and admits he doesn't know where the acceptable line is. - ✅ How should anxious workers relate to pressure to 'stay ahead' of AI?
Reframe AI as ordinary software (find-and-replace 'AI' with 'software'); use it for boring/administrative tasks; don't treat the moment as an existential race. - ⭕ If capture matters 'by a mile' more than organization, and AI now makes retrieval/sorting cheap after the fact, does the entire genre of to-do/notes-app system design (the thing Pierce has spent 20 years obsessing over) still matter as much as he treats it?
Gaps
- incompleteThe claim that AI 'can't exercise good judgment about what matters' rests on one personal experiment (email triage) and general intuition, without reference to any broader evaluation of AI email/priority tools, or to whether the failure is a fundamental modeling limit versus a fixable product-design gap (e.g., feeding the model long-term behavioral signal instead of just message content).
Read next: Look for empirical evaluations of AI inbox-triage/notification-prioritization tools (accuracy studies, user-trust research) and for product designs that supply richer personal context, to test whether the 'gameable system' problem is inherent or addressable. - incompletePierce's 'linear relationship between work done and understanding gained' is offered as strong personal conviction without engaging the counter-case that some AI-assisted synthesis (e.g., structured review, retrieval practice) could build understanding as well as raw manual effort, especially on material the person has already partly internalized.
Read next: Read cognitive-science literature on desirable difficulty, the generation effect, and note-taking/retrieval practice — directly relevant to the persona's standing interest in how the mind actually learns — to see whether 'more manual work = more understanding' holds generally or only for certain tasks and learners. - illogicalPierce asserts capture matters 'by a mile' more than organization and that AI now lets people stop worrying about system design — yet in the same conversation he frames his ongoing 20-year hunt for the perfect notes/to-do app as a serious, still-unresolved problem, without ever reconciling why a well-designed system should still matter so much to him personally if capture is really what counts.
Read next: Watch for whether later installments in this Platformer series (or other productivity writers like Tiago Forte on PARA) explicitly downgrade system/organization design in an AI-search-and-retrieval world, and whether Pierce's own app-switching is better read as recreation than functional necessity. - uninformedThe claim that AI has changed 'not as much as I would have thought' about daily work rests entirely on the personal experience of two unusually heavy-adopter tech journalists, with no survey or adoption-rate data cited, and no acknowledgment that early, enthusiastic power-users are a poor proxy for 'normal people.'
Read next: Look for actual workplace AI-adoption survey data (e.g., Gallup, McKinsey, or academic labor-economics studies on AI use in knowledge work) to check whether the 'busywork only, so far' finding replicates outside a tech journalist's personal workflow.