Knowledge workers have more AI tools available than ever and somehow aren't dramatically more productive. The average developer or PM now has an AI chat assistant, an AI writing tool, an AI meeting summarizer, and an AI code completion engine — and still spends 40% of their day on repetitive work.
The problem isn't the AI. It's the approach.
TL;DR
AI productivity tools for knowledge workers split into two categories: general-purpose assistants (ChatGPT, Claude) that help with any task but forget context between sessions, and workflow-specific tools that learn your patterns and compound over time. The highest-leverage tools are the ones that build persistent memory of your work — not the ones with the most features.
Why most AI tools don't actually save time
Here's the typical workflow with a general-purpose AI assistant:
- Open chat
- Explain your context (project, constraints, what you've tried)
- Get a useful response
- Close chat
- Next day: repeat from step 1
Steps 2 and 5 are the tax. For complex work, explaining context takes 5-10 minutes. If you use an AI assistant 8 times a day, that's an hour just on context-setting. The AI did useful work, but the overhead ate most of the time savings.
This is why the second generation of AI productivity tools focuses on persistent context. They don't start from zero each session.
The productivity tool landscape (honest assessment)
Tier 1: Context-free assistants
Examples: ChatGPT, Claude (standalone), Gemini
Good at: Answering questions, writing first drafts, explaining code, brainstorming. Bad at: Remembering your project structure, learning your preferences, compounding knowledge over time.
Productivity gain: 10-30% on tasks where you use them. Net gain drops when you factor in context-setting overhead.
Tier 2: Context-aware assistants
Examples: Cursor, GitHub Copilot (in-IDE), Claude Code, AI coding agents
Good at: Working within your codebase, suggesting completions that match your patterns, understanding your project. Bad at: Anything outside the IDE, cross-application workflows, non-coding tasks.
Productivity gain: 30-50% on coding tasks. Near-zero on everything else.
Tier 3: Workflow-specific tools
Examples: Screen-aware recorders, workflow automation platforms with AI, personal knowledge management tools
Good at: Learning your actual work patterns, automating multi-step processes, compounding over time. Bad at: General-purpose tasks, anything they haven't observed you doing.
Productivity gain: Variable — starts low (learning period) but compounds. After a month of use, high-frequency workflows can be 80-90% automated.
The gap between Tier 2 and Tier 3
Most knowledge workers are stuck at Tier 2. Their AI tools are excellent within one application (the IDE, the writing tool, the design tool) but blind to everything else.
The real productivity gains live in the connections between tools: the export-from-Figma-upload-to-S3-paste-URL-in-Notion workflow, the review-PR-run-tests-merge-deploy-notify pipeline, the research-synthesize-draft-review loop.
These cross-application workflows are where you lose the most time, and where current AI tools help the least.
What to actually look for
Persistent memory
Does the tool remember what you did yesterday? Last week? If you have to re-explain context, it's not a productivity tool — it's a fancy autocomplete.
Multi-application awareness
Does it work across your entire workflow, or only within one app? If it only helps inside VS Code, it's ignoring the 60% of your day that happens outside VS Code.
Compounding returns
Does it get better the more you use it? A tool with compounding returns is worth the learning-curve investment. A tool that's equally useful on day 1 and day 100 has a ceiling.
Local-first data
Your workflow data includes credentials, internal docs, client information, and personal communication. Any tool that processes this data should keep it local by default, not upload it to a cloud inference endpoint.
Common mistakes
Tool hoarding. Adding a seventh AI tool doesn't help if the first six don't talk to each other. Consolidate before you add.
Evaluating on day one. The best productivity tools have a learning curve. If it's useful immediately, it's probably a simple automation. If it gets dramatically better after a week, it's learning your patterns.
Ignoring the integration tax. Every new tool in your stack has a context-switching cost. A slightly less capable tool that covers three workflows is often better than three best-in-class tools that each cover one.
Confusing activity tracking with productivity. Tools that tell you how many hours you spent in each app aren't making you productive. They're making you aware. Awareness without automation is just guilt.
How Distill fits in
Distill sits in Tier 3. It's a macOS app that watches your screen workflows, builds a library of skills from repeated patterns, and compounds its understanding over time.
It doesn't try to be a general-purpose assistant. It focuses on one thing: learning the workflows you actually do, across all the applications you use, and making those workflows faster and more reliable over time.
The first week, it mostly watches and learns. By the second week, it starts recognizing patterns. By the fourth week, your most common workflows are partially automated — not because you configured anything, but because the tool observed enough repetitions to understand the structure.
FAQ
How many AI productivity tools should I be using?
As few as possible that cover your key workflows. For most developers: one in-IDE assistant, one general-purpose chat, and one workflow-specific tool. Three is the sweet spot. Above five, the context-switching overhead starts erasing gains.
Should I invest time in prompt engineering for better AI productivity?
For general assistants, yes — better prompts yield better results. But prompt engineering is a workaround for tools that don't remember context. The long-term bet is on tools that learn your context automatically.
What's the ROI timeline for workflow-specific AI tools?
Expect 1-2 weeks of break-even (learning curve offsets any gains). Weeks 3-4 show clear time savings on repeated workflows. By month 2, high-frequency workflows should be significantly faster. If a tool hasn't shown value by month 1, it's not learning your patterns well enough.
Do these tools replace my existing automation (scripts, CI/CD, macros)?
No — they complement it. Your CI/CD pipeline handles the well-defined, fully automatable steps. Workflow-specific AI handles the semi-structured, partially-manual steps that live between your automated pipelines.