My AI Toolkit: The Exact Tools I Use Every Day as a Knowledge Worker

TL;DR — This is not a list of "best AI tools" — it's what I actually open every morning as a knowledge worker: what I pay for, what I abandoned, and why the integration between tools matters more than any single tool. The real unlock is a deliberate daily system, not a stack of shiny apps.

I am not a tool collector. I have seen too many people who bookmark hundreds of tools on GitHub, try each one for a day, abandon it, and try the next one. Their productivity setup is elaborate and their output is mediocre.

My principle is simple: the number of tools should be small, the depth of use should be high.

My total AI tool spend: approximately $100 per month. That covers Claude ($20/month for Pro), Cursor ($20/month for Pro), Perplexity ($20/month for Pro), plus the bundled workspace-AI tier (~$10/month as part of the wiki SaaS subscription I rely on). I consider this a reasonable investment for what these tools contribute to my output.

Here is what I actually use every day, and why. Not what I have tried. Not what I bookmarked. Not what a YouTube video told me was the best setup. What I open every morning.

Morning: Information Input

Reeder plus RSS

Every morning, I read subscribed RSS feeds through Reeder. This is non-negotiable morning routine, like coffee.

Why RSS instead of Twitter or Xiaohongshu?

• Editor-controlled, not algorithm-controlled — I see what I chose to see, not what an algorithm decided I should see
• Asynchronous — I read when I want, not when the platform decides to surface content
• Focused — no comments, no likes, no notification badges competing for my attention

I have about 40 RSS feeds that I maintain carefully. Morning reading takes 30 minutes. This is my primary information diet and I am deliberate about keeping it curated.

The feeds I keep are a mix of:

• Indie hacker and bootstrapped business blogs (Paul Jarvis, Levels, Indie Hackers)
• DevOps and infrastructure articles (Monitoring Weekly, DevOps Weekly)
• Technology and science writing (Hacker News, Lobsters, selected Substack newsletters)
• Productivity and learning research
• My competitors — yes, I read what my competitors publish regularly

I drop feeds that start producing low-quality content. I add feeds when I find a writer whose thinking I respect. The curation of this list is itself a valuable activity.

Perplexity: Deep Research

For topics that need deeper understanding than RSS can provide — understanding a new technology before committing to it, finding authoritative sources on a contested topic, getting a structured overview before making a significant decision — Perplexity is better than Google.

Perplexity gives me a structured overview with cited sources, and I can go deeper into the original documents from there. It is particularly useful for research tasks where I need to synthesize information from multiple sources.

For example, before adding Stripe billing to my product, I spent 30 minutes in Perplexity reading about Stripe billing best practices, common pitfalls, and pricing model approaches. The structured output gave me a framework I did not have before. I walked away knowing what questions to ask and what mistakes to avoid.

Work: Deep Focus Time

Claude (Research and Thinking)

My primary AI tool is Claude accessed through its web interface. I use it for:

• Researching technical approaches before implementation
• Debugging errors by pasting logs and asking for analysis
• Thinking through product decisions out loud
• Writing and editing long-form content
• Reviewing code I do not fully understand

The key organizational principle: I keep research in Claude separate from implementation in Cursor. When I am in Claude, I am thinking. When I am in Cursor, I am building. Switching between these two contexts is a productivity feature, not a bug.

Specific Claude workflows I use daily:

Debugging: I paste an error message and full stack trace into Claude. I ask: what is the most likely cause, and what are the diagnostic steps to confirm? Claude is often faster than Google at diagnosing specific technical errors because it can reason about the specific context of my codebase.

Architecture thinking: Before building a new feature, I describe the feature to Claude and ask: what are the architectural trade-offs of these three approaches? This is not asking AI to make the decision — it is using AI to structure my thinking. The conversation that results often surfaces considerations I had not thought of.

Code review: After AI generates code in Cursor, I paste it into Claude and ask: what are the potential issues with this code? Claude finds things I miss on my first review — edge cases, potential bugs, style inconsistencies.

Cursor (Implementation)

Cursor is where I actually write code. I use:

• Cmd+K for inline code generation and editing — I highlight a block of code and ask AI to modify it in natural language. This is the feature I use most, probably 50 times per day.
• Composer for multi-file generation — when I need to build a feature that touches multiple files, Composer handles the multi-file generation.
• Agent mode for complex refactoring — when I need to make changes across many files, Agent mode handles the mechanical work while I review the result.

My Cursor usage pattern: small, frequent prompts throughout the day. I rarely ask Cursor to write an entire feature at once. I build features incrementally, prompting for 20 to 50 lines at a time and reviewing each piece before moving on.

Collaborative wiki workspace: second brain layer

That wiki-style SaaS stays my primary surface for everything that is not code:

• Meeting notes from my day job
• Product roadmaps for my indie project
• User feedback repository
• Weekly reviews and monthly retrospectives
• Content drafts

I keep using it partly because my wife and I share one workspace—household calendars, chores, budgeting, lightweight to-do boards. Shared context solves a tangible coordination problem rather than doubling as motivational advice.

Structured views genuinely help tracking: a minimalist Kanban for product priorities plus a thematic table rolling up user feedback.

The vendor’s bundled generative drafts help me scaffold meeting outlines or exploratory summaries inside the workspace; I treat them as provisional and still rework anything that touches nuance or tone.

Evening: Output and Recording

iA Writer: Writing Environment

For writing long content — blog posts, product documentation, this newsletter — I use iA Writer. It is the cleanest, most focused writing environment I have found.

The feature I use most: Focus Mode, which highlights only the current paragraph and dims everything else. This single feature has probably doubled my writing output because it eliminates the distraction of seeing a long document while working on one section.

Hemingway Editor: Readability Check

After writing in iA Writer, I paste the content into Hemingway Editor to check readability. My target: Hemingway Grade 8 or lower. If it is Grade 10 or above, I simplify.

The reason: clear writing is clear thinking. If I cannot make something readable at a Grade 8 level, I probably do not understand it well enough myself. Hemingway catches my tendency to write long, complex sentences when I am uncertain about what I am trying to say.

My Actual Daily AI Workflow

6:00–6:30 AM: Morning routine — coffee, RSS feeds in Reeder, quick scan of product metrics from my analytics dashboard

6:30–7:00 AM: Review and plan — check yesterday notes, set today three priorities, quick review of any support emails from overnight

8:30 AM – 5:00 PM: Day job — full-time DevOps work at my Australian tech company, AI-assisted where it makes sense

7:00–8:00 PM: Family dinner and time with daughters — this is not negotiable

8:30–10:30 PM: Deep work — product development, content writing, user research

Saturday mornings: 2–3 hours — longer sessions for complex problems that need sustained focus

What I Have Abandoned and Why

ChatGPT free tier: Replaced by Claude for research and thinking. Claude longer context window and better reasoning for technical content made the switch obvious. ChatGPT is fine for casual use but not optimized for technical work.

Mem: Tried it for automatic note organization using AI. It was too magical — I never understood how it was organizing my notes, so I could not trust it. If I do not understand how a system works, I do not trust it with my information.

Roam Research: Too expensive for what it offered. The bidirectional linking was genuinely nice but Obsidian does the same thing for free. I use Obsidian for linked notes now.

Loom: Tried it for async video updates on my product. Recording, uploading, and watching Loom videos was slower than writing a short post. I switched back to text.

Evernote: Used it for 5 years. Migrated toward the collaborative wiki SaaS discussed above because relational blocks and relational tables fit how I collaborate. Evernote stays fine for quick capture—just not ideal for interconnected roadmaps tied to household + product context.

What AI Genuinely Cannot Replace (Yet)

I want to be honest about this, because a lot of AI tool lists are not.

AI cannot replace judgment calls. When I am deciding whether to build Feature A or Feature B, AI can give me frameworks for thinking about it. It can list the trade-offs. It can describe what other products have done. But the final judgment — what matters most right now, what my users actually need, what is worth my limited time — that is mine.

AI cannot replace relationship building. The trust I have built with my users over two years of genuine communication is not something AI can replicate. When a user emails me with a problem, AI can draft a response. But the response that actually resolves the problem and keeps the user happy requires understanding the human on the other end.

AI cannot replace deep reading. I notice that when I rely too heavily on AI summaries of articles and books, my thinking becomes shallower. The act of reading slowly, taking notes, letting ideas marinate — that is cognitive work that produces something AI summaries do not.

AI cannot replace the feeling of making something. Shipping a feature you built, holding a physical product, reading a letter from a user who genuinely got value from what you made — these are feelings. AI can help you make things faster. But the feeling of having made something is yours alone.

The Integration Principle

The most important lesson I have learned about AI tools: the integration between tools matters more than any individual tool.

What I mean by this: the value I get from Claude plus Cursor plus my wiki workspace together is greater than the sum of the value I would get from each individually. Claude anchors research and framing. Cursor materializes commits. That workspace persists decisions so I remember what shipped and why—still routed through manual judgment rather than brittle automation glue.

The best AI workflow is not a collection of powerful individual tools. It is a system where AI tools handle execution and you handle the coordination between tools and the judgment about what matters.

The Specific Prices I Pay and Why

Claude Pro ($20/month): I use Claude extensively — research, thinking, writing, code review. The Pro plan is worth it for me because I use it every day. If I used it occasionally, it would not be worth it.

Cursor Pro ($20/month): I use Cursor for all my coding. The Pro plan includes Claude integration and unlimited AI requests. Worth it for me because coding is how I earn my primary income.

Perplexity Pro ($20/month): I use Perplexity for research. The Pro plan gives better results and unlimited searches. Worth it for me because research is a core part of my work.

Workspace AI bundle (~$10/month): Priced alongside the SaaS wiki subscription. Worth keeping because coordinated documents are how my wife and I share household rhythm plus how I stash product decisions.

Total: $70/month for AI tools specifically. Plus $30/month for hosting, email, and other infrastructure for my product. Total monthly tech spend: approximately $100.

This is a business expense. I track it as such. It is worth it because the tools contribute more than $100/month in value.

The Exact Time I Spend on Each Tool

Claude: 45 minutes per day average. Morning research, afternoon thinking, evening review.

Cursor: 2 hours per day average on product development days. Less on newsletter days.

RSS reading: 30 minutes per morning. Non-negotiable.

Wiki workspace: 15 minutes per day for notes and tracking. More on weekly review days.

iA Writer: 1 hour per week for newsletter writing. More when writing blog posts.

Hemingway: 10 minutes per week for newsletter and blog post review.

The Specific Criteria I Use to Evaluate New Tools

When someone recommends a new tool, I ask five questions:

1. Does this solve a problem I actually have? Most tools solve problems I do not have.

2. What is the switching cost? If I start using this tool, how hard is it to leave? Low switching cost tools are safer to try.

3. Is there a free tier? I try free tiers before paying. Paying creates commitment bias — I keep paying even for tools I do not use.

4. Does it integrate with tools I already use? Tools that integrate with my existing stack are worth more than tools that do not.

5. What happens if the company disappears? Open source and self-hostable tools are preferable to proprietary SaaS for critical infrastructure.

Most tools fail these criteria. The ones that pass are worth trying.

Did this article help you? If you're working through career direction, or want to use AI to work smarter, let's talk — I'm happy to help you think it through.