Craig Aucutt, Designer

Introduction

Productivity tools tell a deeper story than just features and UI. They are reflections of how we think, remember, organize, and collaborate. Over the years I've had the privilege of working on memory-first tools (Evernote's Collect), vertical systems (for education and construction), and consumer products (Out of Milk). Through those experiences, the recurring tensions — between flexibility and structure, automation and control, interoperability and lock-in have always been present. Now with AI entering the mix, those tensions are rebalancing again.

The Arc of Productivity Tools

Before personal computers, productivity was analog: paper planners, ledgers, index cards, Rolodexes, file cabinets. The "tools" were pens, notebooks, filing systems, and human routines. The leap into software began when offices moved from typewriters and paper ledgers to word processors, spreadsheets, and early database systems. Essentially digitizing what was already familiar.

The PC and Microsoft Office dominance

Microsoft's dominance shaped the mental model of productivity: document → share → revise. Users got comfortable with "files, folders, Word, Excel, PowerPoint." But these tools were still mostly document-centric, not workflow- or task-centric.

Over time, Microsft Office evolved adding shared editing, cloud syncing, collaborative tools (Office 365). Microsoft's Copilot ambitions now push AI deeper into the Office ecosystem: summarizing meetings, drafting emails, generating insights inside Excel and Word. In that sense, productivity software is circling back to intelligence built in. (See for example "Microsoft's Copilot built into M365" commentary)

The rise of mobile and distributed productivity

Once smartphones and tablets matured, productivity tools had to adapt to always-on, always-with-you contexts. That shift forced new tradeoffs: simpler UIs, intermittent connectivity, sync constraints, and gesture-based interaction.

Task managers, note apps, reminders, calendar tools flourished on mobile because they matched micro-interaction patterns. Tools like Any.do became central to daily task capture.

Mobile also ushered in context-aware features: location-based reminders, voice capture (dictation), camera-based scans (of receipts, documents). AI now layers on top: image understanding, voice transcription, document summarization. A recent empirical study of AI usage in mobile apps shows common patterns of using on-device ML for vision, language, and context prediction, but also highlights privacy risks and mixed user sentiment. (“An Empirical Study of AI Techniques in Mobile Applications” )

Indeed, a more recent work, What Users Value and Critique: Large-Scale Analysis of User Feedback on AI-Powered Mobile Apps, found that users praise reliability and personalized assistance in AI apps, but frequently complain about technical failures, pricing, or lack of language support. That tension highlights that AI isn’t magic; execution quality matters. (arXiv )

Another forward-looking proposal, VisionTasker: Mobile Task Automation, shows how AI (via vision + LLMs) can automate UI workflows on mobile. Imagine telling your phone, “Open last message, save attachment, create calendar event” — and the tool executes it. That kind of automation could reduce friction in cross-app workflows. (VisionTasker, arXiv)

The rise of AI and influence with productivity tools

Now, productivity tools are not simply hosting tasks or notes, they are becoming intelligent agents. The earlier model was: user does → tool records. The new model is: tool suggests, drafts, assists → user reviews.

The diversity of mental models & why tools proliferate

From the beginning, we externalized mental load: paper, notebooks, index cards. David Allen's Getting Things Done popularized that idea further: if you capture everything out of your head, you free it to think. But the catch was always: once captured, how do you organize it? Outlines? Tags? Boards? Maps? Each person's internal model is unique, and the more expressive a tool is, the better chance it can adapt.

When I was part of the Evernote team, one of the most illuminating lessons was seeing how wildly different users structured their notebooks and tags. Some built fine-grained taxonomies with nested tag nets and saved searches; others ignored tagging completely and used notebooks as time buckets or projects. Our job was to support that diversity: to allow fluid structure that could evolve, not lock people into rigid templates.

That kind of flexibility is exactly what tools like Notion strive for. They promise a canvas where your mental model can grow and shift — relational databases, wikis, blocks, pages all working together.

Notion in the ecosystem: promise and pushback

Notion feels like a natural next step in this trajectory. It is not just a document tool, or a wiki, or a task manager — it's a building-block system where pages, databases, embedded blocks, and relations coalesce. You can prototype a knowledge base, build a project tracker, host a site, or embed docs. It's an ambitious vision for knowledge work.

Their AI layer leans into that ambition. According to Notion's own guide, you can use Notion AI to "tap into your workspace and the wider world to help you create high-quality content, brainstorm ideas, draft documents, and even reformat pages based on templates." Their blog Introducing Notion AI for Work announces built-in meeting notes, enterprise search, research mode, and access to top models — "AI Meeting Notes: Never take notes again!" they claim.

One telling design essay, Speed, Structure, and Smarts: The Notion AI Way, describes AI as "part of the core product architecture, designed to support the way you work, not force you to adapt to the way it works." That phrasing is critical: the ideal isn't to wrap AI on top, but weave it into the fabric of structure and user intent.

Yet with power comes risk. The more freedom a user has, the more likely they'll spend time refining schema, building dashboards, reconfiguring linkages — sometimes more than doing actual work. In community forums, you'll encounter frustrations like:

"Almost entirely useless. Could be decent if you're a writer or need simple things summarized like meeting notes or something."

That kind of comment captures a tension: the promise of AI is seductive, but if it doesn't ground itself properly or let users validate it, it becomes noise.

Another concern is that when Notion automates structure (auto-tagging, schema suggestions, or AI properties), it risks pulling users out of the cognitive loop. If the tool is making structural decisions invisibly, mistakes or misalignments propagate silently.

It's worth noting Notion's ambition is growing: their upcoming Mail app will let users compose, organize, and archive email inside Notion, using AI to schedule, archive, and draft messages. This positions Notion not just as a workspace but a hub across workflows.

AI, "workslop," and productivity illusions

Embedding AI into productivity tools feels like magic — but beneath the surface is a growing hazard: workslop. As defined by a recent Harvard Business Review article, workslop is "AI-generated work content that masquerades as good work, but lacks the substance to meaningfully advance a given task." Axios reported on this too, calling workslop a major drain on efficiency.

One survey estimated that a company with 10,000 employees could lose $9 million annually to the overhead of fixing or cleaning up AI-generated workslop.

When your tool auto-generates a summary, a slide deck, or a draft memo, the illusion of output is tempting. But colleagues downstream must often review, correct, or reinterpret — and that work falls silently on them. A TechCrunch warning puts it succinctly:

"Beware coworkers who produce AI-generated 'workslop' … content that masquerades as good work, but lacks the substance to meaningfully advance a given task."

A design fallacy is to treat AI as a "zero-effort shortcut." In fact, it often front-loads work into the review stage. As some legal teams are noting, AI can accelerate first drafts, but quality control still demands human input and revision.

In one vivid expression of the risk, a Hacker News thread discussed how teams using AI heavily have not, in fact, seen delivery speed or margin gains — despite claims. Some developers described needing to rewrite or revalidate AI work manually, erasing extraordinary claims of productivity gains.

So the question becomes: how do you gain leverage without amplifying noise? AI should be a thinking partner, not an amnesiac ghost writer.

Interoperability, portability & avoiding silos

Even the best AI or workflow tool fails if your content is locked in. Interoperability remains a recurring pain point in productivity software. Tools often export flat text files, CSVs, or markdown — but lose relational structure, metadata, tags, version history, and permissions in the process.

Today there are some promising efforts: the Data Transfer Project (by Google, Microsoft, Facebook, etc.) aims to standardize service-to-service data transfers. But it was designed for bulk data exports (photos, contacts), not real-time two-way productivity workflows. ActivityPub, the decentralized social protocol, proves federation is possible — but it's not built for nested relationships, tasks, or versioned documents.

Through my lens: memory tools, domains, and design philosophy

When I think back to building Collect in Evernote, the heart of the feature was this: capture with zero friction. Take a photo, a voice memo, a clipped article — drop it in. Let structure adapt. If you can't capture quickly, users tune out trust. That belief shaped how we handled tags, notebooks, search, and saved views. The variance in how people structure that content was always astonishing.

Later, working with shopping-lists (Out of Milk) reminded me how small UX details matter — sorting by aisle, sharing, persistent templates. The domain data in those lists can suggest future structure (frequent items, categories). Tools that harness that signal (with consent) can evolve smarter.

In education (Edmodo) and construction (PlanGrid), constraints sharpen focus: offline reliability, role-based views, version sync, document alignment across teams. There's no room for messy schema drift when a structural mistake onsite costs thousands.

Toward guardrails & opportunity space

The future productivity tool will likely be hybrid: a robust core + domain-specific layers + AI that remains visible, reviewable, and user-directed.

Here's how I think we can do better:

AI should suggest, not impose structure. Always visible, always editable.
Use provenance: show which block, doc, or conversation fed an AI suggestion.
Tiered trust: some suggestions commit automatically only after user review.
Resist full automation shortcuts. Shortcuts are helpful; autopilot is dangerous.
Interoperability should be baked in: semantic exports, APIs, connectors.
Interfaces should scaffold structure gradually (start simple, evolve).
Measure quality, not just volume (rework, corrections, trust metrics).
Provide audit trails and versioning for AI edits so you can revert or inspect.

Closing thoughts

Productivity is a deeply human problem: how we think, collaborate, perceive time, and structure context. Tools have come and gone to mold better yet none been more simple and useful as AI tools such as chatGPT. But as AI grows its role, we must guard against opacity, drift, and broken expectations.

The Long Arc of Productivity Tools