Signal Over Noise #14
August 8th, 2025
Dear Reader,
Back in April, I told you the AI tool landscape was a mess and gave you a simple framework: pick one core assistant, add research tools, maybe some workflow apps.
Some months later, I’m still using that framework, but my actual daily stack looks completely different than what I recommended, and it changes depending on context (at my desk, out and about etc.)
Here’s what I’ve learned after months of real-world testing, and what’s actually running my work and weekend projects right now.
The “One Tool” Fantasy Falls Apart
My daily AI stack has settled into three distinct roles—not the theoretical “one core assistant” I recommended, but three specialised tools that each handle something specific without driving me crazy.
Turns out the Swiss Army knife approach works better than the magic bullet.
ChatGPT: The Reliable Workhorse
ChatGPT and Perplexity have completely replaced Google on my phone. When I’m out and need to identify that weird plant, ask a quick question, or figure out what my kid found on the beach, it’s my go-to. The visual recognition alone makes it worthwhile.
But here’s where ChatGPT really shines: personal automation projects - or custom GPTs.
Me and a friend are co-DM’ing a D&D campaign for our kids, and I’ve built a campaign manager that tracks character stats, plot threads, spells, and world-building details. It’s nothing sophisticated by enterprise standards, but it handles personal RAG (retrieval-augmented generation) tasks really well.
ChatGPT excels at creative-but-structured projects. It’s reliable, the interface is familiar, and honestly, most people could replicate this kind of setup using Custom GPTs - still the lowest barrier to entry for this type of automation.
Claude Max: The Professional Orchestrator (With Quirks)
This is where I’ve made my biggest shift since April. I upgraded to Claude Max specifically for the MCP (Model Context Protocol) integrations, and it’s changed how I work—when it remembers it can do these things.
Claude now has direct connections to Notion, Obsidian, and Things 3. Instead of the endless copy-paste cycle between AI and my actual work tools, Claude can read my notes, update my task lists, and cross-reference projects across all my systems.
When I’m planning my day, Claude looks at my Obsidian daily notes, reviews my current task list, and helps me prioritise what actually matters. When it works, it’s genuinely transformational.
The problem? Claude keeps forgetting it has these capabilities. I’ve created TextExpander shortcuts that I run multiple times a day just to remind Claude it can access my files and what time zone I’m in. It’s annoying but worth it.
I’ve also had Claude analyse all my writing to create style guides for my book, blog, and newsletter work. Having an AI that understands my voice and can maintain consistency across different projects has been invaluable—even if I do have to remind it about this regularly.
Since upgrading to Max, I’ve only hit context limits once versus multiple times daily before. When your AI can actually execute in your workflow instead of just generating suggestions you have to manually implement, then it starts actually being of significant use.
Perplexity: The Research Engine
Perplexity is built into my browsers on both desktop and mobile, making it my frictionless go-to for research. When I need reliable information rather than creative assistance, Perplexity wins every time.
It’s not trying to be everything to everyone. It’s focused on finding and synthesising information from current sources. That clarity of purpose makes it incredibly reliable for the “I need to know this thing” moments that punctuate every workday.
The Comet Browser Experiment
Day-to-day I’m using Comet as my desktop browser, which has an AI assistant built-in that can take over and carry out tasks—reading blog articles and inserting related media, updating excerpts and tags, and even YouTube videos, for example.
Comet can take over some of the monotonous busy work while you wait to see if it screws up.
It’s genuinely helpful for content research and organisation, but the a mobile version is sorely lacking. So I’m back to switching between tools depending on whether I’m at my desk or on the go.
What’s Changed in My Thinking
The biggest shift since April is that I’ve stopped looking for the perfect all-in-one tool and started building a stack where each tool excels at something specific.
ChatGPT handles personal projects and mobile convenience. Claude orchestrates professional workflows (with regular reminders about its capabilities). Perplexity handles research. Comet manages desktop browsing and content tasks.
Each does what it’s best at, and I’m not trying to force any of them into roles they don’t fit.
I still need Obsidian as my backup, storing everything in plain text files that work regardless of which AI system I’m using. When Claude forgets it can access my files, or when I hit context limits, everything’s still there in a format I can access from anywhere.
The Reality Check
This approach requires more management than the “one tool” fantasy I originally recommended. Different tools for different contexts, regular maintenance of AI memory, and acceptance that mobile and desktop workflows remain separate.
But here’s the thing: the specialisation actually reduces friction once you stop fighting it. Instead of trying to make one tool do everything poorly, I have tools that do specific things really well.
Should You Follow This Approach?
If you’re still trying to do everything with one AI assistant, you’re probably fighting more friction than you need to. The tools have gotten good enough (and different enough) that specialization often beats the all-in-one approach.
But don’t just copy my stack. Pick tools based on what you actually do, not what sounds impressive in demos. Test the integrations that matter to your workflow. And be honest about the maintenance overhead—this isn’t set-and-forget automation.
The AI landscape is still a mess, updating at a ridiculous pace (ChatGPT-5 was just announced as I write this), but at least now we know which parts of the mess are worth navigating.
The question isn’t whether to use AI tools—it’s whether to build a thoughtful system or just chase whatever’s newest.
Until next time,
Jim
Signal Over Noise is written by Jim Christian with a spring in his step and a song in his heart. Subscribe at newsletter.jimchristian.net.
Made with ❤️ in Valencia by Jim Christian. For feedback, please reach out to [email protected].
