I have to admit, I slept on Claude at first, and that was a mistake. I was already using several other AI tools and just didn't see the need to further complicate my stack. Perplexity for research tied to a local LLM for summarization, and I didn't see the need for more chefs in the kitchen. Except, I should have fired them all and moved over to Claude Code, because it's far more versatile and powerful for the workflow I use for ideation and research.

Okay, but why Claude Code?

The toolchains are easier for me to understand

I might have taken a while to dive into Claude Code, but that's mostly because I'm no developer. Yes, I took some computer science modules at university (in pre-LLM AI, no less!), but those were focused on C+, and in one case, Assembly, and are so far divorced from the current meta for LLM work that I might as well forget most of it.

But after working on a couple of projects that use Claude Code, I realized I don't need any of it. I can just... talk to the computer, as if it was someone on Discord. No code needed, I don't even really need any coding knowledge because I'm not using Claude Code for, well, code.

I know, that sounds odd, but hear me out. Once I've attached Claude to a couple of MCP servers to search for up-to-date information and social media posts, it turns into an ideation machine for future content.

I'm on the Max plan, but maybe I don't need to be

I'm no developer, but my wife is as part of her work, and that's why we've got the Max plan (although it's the x5 plan and not the x20). The amount of Claude usage I do daily barely uses any tokens, while her work needs most of the context window for front-end development use.

Claude Code helps with the friction of the creative process

I've never been much good at ideation

I love being a writer, but there's one aspect of it that isn't my favorite. Ideation is hard work, partly remembering what I've already covered, but also looking for relevant, timely, and genuinely useful topics can feel like an uphill struggle at times. Okay, most of the time, especially in this shifting media landscape.

But Claude Code can take the annoyances out of the process for me. AI is much faster at browsing the internet and finding relevant social media posts or new GitHub repos with cool tools to highlight. And with a quick CLAUDE.md file, it can find me relevant technical sources and topics that suit the types of software and service journalism that we try to live up to.

Instead of diving into traffic numbers, Claude can find me the latest topics, before they hit virality. It doesn't always work, but it's close enough that I can generally feel which options are best to move forward with.

With a little role-play, Claude will put some options together

I can tell Claude Code to set things up for research, and it'll create a new folder and a new markdown file in it to write in. Then it's time for a little role-play, because prompting always works better for ideation with a defined role:

You are my technical writing partner.
Using CLAUDE.md, propose 5 specific article ideas that would be valuable
for $userType using $tool1 and $tool2 at home.
For each idea, include:
- Working title
- One-sentence “why this matters”
- 3–5 bullet points of key sections

CLAUDE.md already set up the average user for the site, but I'm going to add more archetypes based on traffic patterns, and see what that does to ideation quality. They're already similar to what I was writing, but without the hours spent browsing Reddit, Google Discover, and Feedly to find interesting topics.

Which can be fleshed out into sections

Once a shortlist is together, I'll get Claude Code to expand on a few ideas and let it loose on a folder full of relevant material. Sometimes this'll be a call to Context7 to do some actual code document reading, especially if I'm working with Proxmox or anything else with good source documentation. I could probably use the desktop app for all these things, but Claude Code is set up to create files by default, and that keeps me more organized, and I don't lose anything as I go along.

I don't use agents yet, but I should

The next step for me is to split the workflow up into sections and create dedicated subagents for each section. That way, I can keep the ideation thread free of the search and expansion that is normally part of the linear process. One that searches and finds relevant information to download, while another scans that folder to glean insights and create summaries. Perhaps an automated agent that orchestrates the others, so I can wake up every day with new projects and ideas to flesh out. My imagination is really the limit here, and because agents are described in plain language, I don't need to write a single line of code.

You don't have to be a developer to use Claude Code and that's most of its power

I've always been a big-picture thinker, and pseudocode makes sense to me. It's when you get into the nitty-gritty that my brain refuses to cooperate, but I don't need to worry about that when setting up Claude Code workflows. Everything is defined in Markdown filesin plain English, and letting Claude find things with minimal supervision is the best way to use this generation of LLMs. They can see interlinked ideas better than I can because that's at the core of their design, which means I can generate ideas to pursue further at breakneck speed, helping me get on with the important task of writing.