Using Claude Code for Everything Except Coding
Things AI Enablement Strategies Leave Behind

Katherine Cass • January 21 2026 • ~15 min read

I want to speedrun how I use generative CLI tools (aka Claude Code1) not just for coding - but (and at this point, mainly) for daily operations. I've been giving extra attention to how my ~system~ (the very specific-to-me way I use LLM CLI tools) works to help my brain work faster, especially while onboarding at a new company.

Every SVP-of-something/consulting company replacing the workforce with robots/professional LinkedIn poster is out here talking about how to codemaxx with Claude Code. And yeah, it's great for that2. But these tools can and should be used for way more.

Most recently, I used it to help onboard me to a new role, using a similar methodology as I do with personal projects to manage and build context.

It can be used in way more one-off tasks than I think we talk about as well - the random "I need to go through this giant 200-item ticket list and clean out all the stale stuff" tasks that somehow fill up our workweek. Have an LLM give a first pass, suggest changes, flag where new context is needed. It can speed up a lot of tedious work that isn't writing code.

Today I'm not going to focus on those one-off tasks (though they're VERY valuable). I want to talk about how I leverage it with the repetitive stuff: note taking, task management, learning, discovering and tracking patterns/connections between people and projects - to keep my own brain organized so I can operate at a higher level.

So let's talk about my ~system~3.

Foundational Assumption: LLMs are horrifyingly good at managing horrifyingly long todo notes

With baby's first multi-agent system (link to my retro) in December, I learned a lot. The first thing that really stuck with me was that LLMs are excellent at managing large, horrifying todo files (requirement files, Slack history, whatever) in ways I don't instinctively understand as possible (as a human).

Data formats we don't consider scaleable for humans work better for LLMs to think at4 and process, which changes some longstanding design/tradeoff decisions the industry has spent years designing around. And relates to this whole thing of "don't over-automate out the gate until you understand you AND the machine better" I'm building a hard and annoying opinion on, but fortunately for you this is not a post about that.

I figured if those agents I built were good at managing large horrifying requirements files and finding dependencies in code, they're probably good at updating/managing my own todos around personal constraints, and finding idea connections/task dependencies among initiatives I work on. So I prompted Claude CLI to set up a simple system, same way I did when I started working with my agentic system: "Hey, let's create a todos.md and that's going to be essentially our requirements doc. You're going to help me update it whenever I leave a meeting."

I kept this super simple because of my very strong opinion: don't automate everything out of the bat.5 I made that mistake a lot before. My current approach: overcorrect toward going through my regular tasks manually with Claude as an assistant, then tiptoe toward finding repeat actions codify based off statistics.

That said... I definitely feel I overcorrected and could have built some helpful ~system~ rules/scaffolding up sooner. But I'm confident that experiencing extremes firsthand is the best way to build those instincts around working with these systems. Transformative experience! Journey before destination!!

Pre Reqs

Download Claude Code CLI (or similar). If your company is more than 2k people you probably easily get a license (and meet a KPI for some AI enablement team out there). If your company is less than 2k people... you are already this CLI? Greetings my robot friend.

What I actually set up

I started with just a todos doc and a couple folders:8

While onboarding, the People Index came in handy multiple times in meetings9 where I saw names on the screen that I recognized. My onboarding buddy had given me all the context via multiple brain dumps, but of course I can't immediately remember all of it because I'm a human and my mind gets scattered when trying to learning everything in one week. So throughout onboarding, I just pull up the People Index whenever I see a name I recognize but don't fully remember. It's sort of like live flash cards and I think it helps me learn more quickly and navigate new meetings with more confidence on my assumptions.

More importantly: Claude kept it updated for me through all my onboarding sessions and meetings - via me flagging notes I took with <people index update> (I told claude that tag would indicate process/store/action needed + give a pointer on what). Claude occasionally suggested some updates I didn't notice itself. It was. I just... it's great y'all.

We all know org charts are kind of a nightmare and exist in three different places, and the best org chart is always that one Google Sheet that's horribly out of date but has the most context in it. Well, people-index is my CUSTOM fourth org chart. It's both better than any other org chart AND worse since not really relevent to anyone else! Hooray.

✨ NEXT BIG AI PREDICTION FOR 2026? ✨

There will NOW BE FOUR ORG CHARTS.

f o u r t h o r g c h a r t

For me, kept current, ~bespoke~, as a remote worker, this is my fourth place...

meet me for a coffee chat in the bespoke orgchart

How I interface with ~the system~

To double-check my notes.txt against a source of truth, I use Notion AI for transcripts (purely because that's the tool I figured out I had available day 1 to view transcripts of every meeting). If integrated with Zoom correctly, seems you can see who actually said which thing. Super helpful for context reference.

At the end of a meeting (or whenever I have a break), I can chat with Notion AI to review my notes and cross-reference the transcript, then update my notes with anything missing/corrections and then have Claude process my notes.txt. I also set Claude up (via skills/prompting) to look for any screenshots I took during the meeting or things I downloaded for additional context. It takes in everything and actions my notes/observations from notes.txt into todos/context updates/etc.

Why not send transcripts directly to Claude Code? You could! But check your company's policies first. There are very likely rules about what data you can and cannot put into AI-assisted systems. Keep your job and protect others' data by being aware of this.
How I build ~the system~: In this post I say things like "I have it do X and it does Y." Please note this is just prompting Claude Code CLI to set up things up for me. I don't even create the files/folders. It's like 20-30 minutes of initial work, plus some continued maintenance6 as you learn yourself or find any improvements you want to make. I'm NOT here peddling my repo or product because for me, this is a methodology... a way of learning how to build ai tooling custom to you. I feel that if I share my structure, it's kind of missing the point.

On improving the system: pretty much out the gate on my first day, I started flagging things to Claude in my running notes.txt, so I started using descriptive tags in my notes to help with that (basically prompted Claude to know to pay extra attention/action things with < a tag >... formatting was a random choice).

Sometimes my notes are just for me to build my mental framework live while I'm in the meeting. But a lot of times it's something that needs to be actioned (tags aren't always exactly the same but examples below):

That last one is huge. I can stay present in the meeting because I know the idea is written down and will get processed later (into todos/relevent docs/etc.), even if I get busy.

Sidebar on voice-to-text tooling: Voice-to-text options are way better now. I've been using Wispr Flow for my personal work and love it. The macOS default voice-to-text is horrible. Don't let this or past experiences turn you off. Try a newer tool (there's free/open options too, or just vibecode your own?) - it's going to make you work 4x faster. I never thought I'd switch over, but tried it for literally 2 minutes, and it changed everything about how I work.

Tip: if you do use voice-to-text, update your tool's main context (e.g. CLAUDE.md) to know that, and to infer intent behind any weird/entertaining mistranscriptions.

Daily logs

So I have these daily logs (I call them dev logs) that I barely even look at daily. But I have Claude write them for me after sessions/meetings, every day. It's a habit I picked up from my personal projects for a variety of reasons, but mainly for self-managing growth and noticing personal anti-patterns. Sometimes it helps ground myself in what was actually accomplished if I hit a wall or have a bad day (my work is not measured in jira tickets alone).

It also works very well as an index for Claude to find more context on something (e.g. "what was that incident we talked about last tuesday at some meeting?", "generate an summary for X person on how I used AI in random day-to-day tasks this week", "what concrete feedback can I put in the developer experience survey this quarter?").

It's weird and challenging to stay on top of all this paradigm-shifting (and a lil bit existential-pressure-inducing) AI stuff that changes week over week, especially as a leader. Sometimes it can feel like I'm not learning anything, or I feel behind or confused. There's not a book or a class on this (and I like books and classes :c).

Fortunately I'm definitely a stream-of-consciousness person, so in my notes (and Claude sessions) I'm always flagging how my perspective is changing about something, something I want to work on or grow a skill set around, something I think I did really well that I want to mark down and figure about how to do more of (or brag about myself in a blog post on). And now, all that gets neatly organized by Claude for me into daily entries, pretty much automatically.

This underlying habit allows a tool (like Claude CLI) to speed read weeks of work and reflections at any time, and surface mental changes and learnings7 with a timeline.

Another enablement example: onboarding curriculum

Something I normally do when I start a new role is come up with my onboarding self-curriculum. What's everything I need to know after the first few days? What are my known unknowns? How do I build a study plan to learn all this stuff? This becomes especially important if team-specific onboarding documentation is sparse or stale.

I have many spreadsheets going back many jobs. But for the first time, instead of doing all that manual labor myself, Claude could just go through all my files after a few days and put together a suggestion for me.

Of course I seeded the list with a rambling voice-to-text braindump of things on my mind, double-checked/edited output a bunch as a human (we have value! please spare us!). But it was a really good starting point. I'm looking forward to having this context machine at my fingertips once my day-to-day settles too.

Other ways I've used this (beyond onboarding)

I've experimented with AI to help pull together an entire postmortem for an incident. I uploaded all the resources/docs I could (error logs, related tickets, Slack history, etc.). It reviewed stuff in record time, put together a timeline, flagged issues and red herrings. In a way that I couldn't quite do as a manager with limited hours. Yes, still review it as a human. Verify. But a hand with copy/paste/formatting/mental math stuff + initial observations from your robot intern can help.

Other things I've used it for:

I've operated in the CI/CD + quality + release management space my entire career, so I'm excited to more regularly use GenAI CLI tooling like Claude Code to quickly process and analyze historical information sources, pull out repeat high-level patterns across multiple systems/tools, and summarize findings for my review to help inform a roadmap that gets the whole team (and company) into a better state.

And I think there's a lot of potential areas for enablement (beyond code generation) for non-engineering positions (marketing! tpm! sales!), where anyone can benefit from developing even the absolute minimum amount of computer terminal literacy.10

The flexibility of AI cli tooling allows you to have something that can easily pull ALL the data (even accepting screenshots/copy-pastes), do a more accurate analysis (often via writing actual scripts), and even action/change things itself if needed, if granted permission (be careful). Even if you're not a software engineer, consider what you could do if you had a personal analyst and an executive assistant at your disposal, customizable even for every odd, one-off task.

Note on setting up GenAI CLI tools like Claude Code for use cases beyond coding: It's not about configuring any special tooling, integrations, etc. Just download + open a GenAI CLI tool and talk to it.

On pulling information/data in via GenAI CLI tooling: It can be easy but comes with risks. Do your research. And when you're ready, don't get hung up on "I don't have a way to pull X info from Y." Yes you do. Be creative. Think outside the box. Have it write a javascript snippet to pull something live from a webpage. Old school ctrl+c. Take a screenshot.

And of course, if your company provides access to one of these tools, check guidelines on what you can do and share with these models.

TL;DR: use cases that aren't coding

maintain a people-index
Create a doc tracking everyone you meet: who they are, what they own, how they connect to your work. Claude updates it from meeting transcripts + your notes, using context you've provided it about your role and needs.
10-second lookup mid-meeting when you blank on someone's role. Your fourth org chart.
meeting processing
Though I recommend only using this specific approach for personal hobby stuff (virtual meetups/workshops/etc.) to avoid getting into corporate policy gray areas, shoutout to: Download transcript → Claude processes it → archives + updates relevant docs/todo lists. Point it at your downloads folder and notes.txt if more context there.
Actually listen instead of frantically note-taking. Context captured, processed later.
daily logs
Running log of what you worked on, can be entirely written by Claude (who can summarize work from session logs, todo/doc changes, etc.). Use for self-management, pattern recognition, indexing, etc.
"I wasted all day on this bug" → check the data → "oh I was also in 4 hours of very productive meetings"
building curriculum
Your self-guided study plan. Have Claude comb through daily logs + meetings + docs + whatever other resources, ask you about concerns and gaps, and suggest what to learn.
Scattered onboarding chaos → structured learning plan. You can still seed/edit/verify as human.
incident postmortems
Upload any and all data you can: error logs, complaints, slack history, related tickets. Claude builds a timeline, suggests red herrings & root cause, as well as fixes into one doc. Again - check company policies here.
Fast first pass on the tedious scaffolding work. You review/edit/judge.
status reports
End-of-week updates, metrics views your dashboards aren't currently configured to provide. Once I even used this to auto-bucket hundreds of support questions to estimate improvemnts in certain issue areas. Claude crunches through data sources for YOUR role.
Skip the "ask engineer, wait 3 days" loop. Explore systems yourself first. Measure things that have been typically hard to quickly measure.
a rant about slop

With great tools comes great responsibility. Do NOT generate slop and make your reports read it for you.

Take ownership of what you generate. Review it yourself first. Make sure it's something you, as a human being, would present and use another human being's time to read.

See this if you need help with your moral code.

/ end rant /

Final Note: it's a methodology, not a product

I don't think you need to build a tool or framework for everything. You just need to get familiar with how to use these tools more easily (and securely), and incorporate that into your day-to-day.

Note: the second you start figuring out how to use tools like Claude Code CLI, you may want to immediately build a tool or framework for everything.

The VERY BEST tool you can use for "AI enablement" is your own ability to learn and develop a custom, balanced work methodology that provides a good symbiosis with LLMs. It's not a specific product or repo or skill set or tool you download.

I'm very pro on learning how to build with it as you work, and figuring out yourself the line between:

Build your bespoke tool and methodology based on what you need. Learn how to work with yourself at a meta level to understand how you fall into traps with AI tooling and get better at that.

I highly recommend daily logs for that too - it will help point out those meta patterns for you. :)

[1] "Great" is nuanced. Some things I read this week that handle nuance: agent psychosis and where I'm at with AI. Also, maybe do a author bias/source check on these, don't listen to a random person on the internet.

[2] Locking into one ecosystem is bad for so many reasons but here I am and I feel weird about it but also, I'm busy

[3] ~My system~ is a methodology. Not a product. It's a vibe.

[4] Run linear algebra on? Statistic at?

[5] yes I am going to just continue saying this

[6] Beeee carefulllll don't go overboard with tweaking the system more than you're actually using it!! Don't over automate!! Don't over automate!!!!

[7] Surely there's nothing wrong with this approach right

[8] "Katherine, you're just building a memory/context management system." Yes, congratulations! This was easier than reading a bunch of papers and integrating stuff. I don't know what I'm doing, it's a baby.

[9] I'm glad my work is largely remote/over Zoom. Most of my ~system~ here might not be relevant if you're not remote-first.

[10] Or just wait a few weeks for things like Cowork to roll out.

doodles hand drawn & painted, ai formatted <3