Claude Code and OpenCode Are Not Just for Coders. They Are for Builders

S Ravi Kumar November 22, 2026 10 min read 23 views (23 unique)

Claude Code and OpenCode are called coding tools, and most people who are not programmers never open one. I used one to build my own Jarvis: it sorts my mail, brings me my numbers and plans my day. Here is how you automate your first repetitive task this week, and how to do it without your data leaving your laptop.

I built my own Jarvis

You know Jarvis from Iron Man. Tony Stark does not open an app and copy things into it. He says what he wants, and Jarvis goes and does it.

I have something like that now. It is not a product I bought. It is a folder on my laptop and a tool called Claude Code that works inside it.

Every morning I ask it one thing: what are my tasks today? Before it answers, it reads what my projects did overnight, sorts the new mail in my mailboxes, and writes me a reading list of the few mails worth opening. Then it gives me the day.

Last week I asked it how many people watched my last live stream. I did not open YouTube Studio. A minute later I had the answer: 53 views, about twelve minutes of watching in total. Not a happy number. But a true one, and I had it without touching a browser.

Then I asked a second question: which of my videos do people still watch? Nothing existed for that. So it built the automation itself, ran it, and gave me a table of my 25 most watched videos of the last 90 days. The automation is still there. Next week the same question costs me one line.

A friend of mine, who also builds things, told me the same story this week from his side: "I have made my own Jarvis with Claude Code. I give it a command and it does the job."

Neither of us was talking about writing software. We were talking about getting our own work done.

The label "coding tool" is wrong

Claude Code, OpenCode, Codex CLI, Gemini CLI. People call them coding tools, and most people who are not programmers never open one.

Here is what they really are. Each one is an AI model with hands. It sits on your machine, inside a folder you choose. It can read the files there, create and change files, run things, and talk to services such as your mail or your calendar. The usual name for this kind of tool is a harness: the model is the brain, and the harness gives it hands.

It was made for programmers first, because code is files and instructions to the computer. But your work is also files, and things to be done with them. Your reports, your invoices, your photos, your notes, your mailbox. So the same hands can do your work.

That is why I say these tools are for builders. A builder is anyone who wants a job done once and then done for them every time after. You do not need to read code for that. You need to know what you want.

In a chat window, you are the hands

Think about how most of us use ChatGPT, Claude or Gemini in the browser.

Your work lives in one place: a mailbox, a folder, a spreadsheet. The AI lives in another place: a chat window. So you become the courier. You copy the data out, paste it into the chat, wait, copy the answer, and carry it back. Paste. Save. Send.

The AI is the brain. You are the hands. And when the job comes back next week, you carry everything again.

Some chat apps now have "computer use": the AI looks at your screen and clicks for you. That helps when the only way into something is a screen made for people, like a government portal. But it is slow, because every step is a look and a click, and next week it has to look and click through all of it again.

A harness works where the work lives. And it can do what the other two cannot: it can turn the job into an automation that runs by itself afterwards.

Where the AI lives Who carries the work Next week
Chat window In a browser tab You, by copy and paste You carry it again
Computer use On your screen The AI, by looking and clicking The AI clicks through it all again
Harness In your folder, next to the work The AI, directly The automation runs; you only read the result

What my Jarvis does that is not code

I am a developer, and I do build software with these tools. But the things that changed my day are not software projects.

My mail is sorted before I read it. Five mailboxes. New mail is filed under labels by rules I agreed to, and I get one reading list for the day with the few things worth opening. There are 49 of those daily lists in my notes now. Before anything changes in my mailbox I can ask for a dry run that only shows what would happen, and every run can be undone.

My numbers come to me. The YouTube numbers from the start of this post. Once a day it also checks my public projects: downloads, new releases, whether the last automatic check passed. I used to open one tab after another for that.

My logos match. My projects share one logo family: the same letter shapes, the same tile, a different colour for each app. Asking an image model for "the same logo again, for the new app" never gave me the same logo twice. Now the logos are drawn from rules, and a new app gets a matching one in a minute.

My notes keep their own index. One command rebuilds two pages that show me the state of every project and every document I have.

My day is planned from what really happened. It reads what was finished yesterday and what is still open, and writes today's list. I correct it, and it remembers the correction.

None of this is impressive as software. That is the point. Each one is a small chore I would never have sat down to automate myself, because building the automation would have cost more than the chore. With a harness, the cost of building it dropped to a conversation.

Build the automation once, then let it run

Here is the lesson I did not expect.

At first I asked the AI to do the chore itself, every day: "sort today's mail". It worked, slowly, a little differently each day, and it used up the monthly allowance of my subscription.

The better instruction was: "automate this for me." The AI is used once, to build the automation. After that the automation runs by itself. It is fast, it costs nothing to run, it does the same thing every time, and when it is wrong I can point at the exact rule that was wrong.

I keep the AI for the parts that need judgement, such as choosing which five mails out of two hundred deserve my morning. Everything that can be counted, matched or moved is done by the automation.

You never have to read what it built. You only have to say what you want, look at the dry run, and say yes or no.

Your data can stay on your laptop

Many people tell me they do not want to give their mail, their documents or their family's files to OpenAI, Anthropic or Google. Some worry that the models learn from what we type. Whether they do depends on the plan you are on and the settings you chose, and most people have never looked.

A harness gives you a way out of that question. The harness and the model are two separate things. OpenCode, for example, can be connected to a model that runs on your own machine, through free programs such as Ollama or LM Studio. Free models that anyone may download, such as Qwen and Gemma, are now good enough for many everyday automations. I run models like these on my own Mac.

You need a strong machine for it: a laptop with a good graphics card, or a recent Mac with plenty of memory. A local model is slower and less clever than the best paid ones. But for sorting, renaming, collecting and summarising, it is often enough, and nothing you give it ever leaves your laptop.

So you can choose. The strongest paid model for hard work on things that are not private. A local model for anything you would not want on someone else's server.

What you should know before you try

It looks like a programmer's tool. A harness opens in a terminal, a plain window with text. You do not need to know any commands. You type what you want in ordinary language. The first ten minutes feel strange. Stay for the eleventh.

It costs little to try. Claude Code needs a paid Claude plan; the free plan does not include it. OpenCode is open source, comes with a set of free models, and can also use a model that runs on your own machine.

Hands can break things. An AI that can move your files can also move the wrong ones. Every harness asks before it does something risky, and you should let it ask. Start in one folder, with copies of your files, not your only originals. Ask for a dry run first. Ask for a way to undo.

It is still wrong sometimes. I wrote in my last post about who pays when the model is wrong. That has not changed. The difference is that an automation is wrong in the same place every time, so you find the mistake once and fix it once.

The chat window still has its job. Thinking through a decision, learning something new, drafting a hard message: I still do those in a chat. I am not asking you to throw the chat apps away. I am asking you to stop being the courier.

Automate one repetitive task this week

Do not start with a big project. Start with one chore.

  1. Pick one thing you do every week that means carrying things from one place to another. A report you put together. Files you rename and sort. Numbers you collect from three websites.
  2. Install one harness. Claude Code and OpenCode are the two I have used most; each name opens its install page. Codex CLI and Gemini CLI work the same way.
  3. Make a folder with a copy of the material, and open the harness in that folder.
  4. Tell it, in your own words, what you do every week and what you want at the end. Then say: "Automate this for me. Show me a dry run first, and tell me in plain words what you will touch."
  5. Next week, ask it to run the automation. Notice that you did not carry anything.

The day that happens, you have the first piece of your own Jarvis. The second piece is easier, and by the fifth you will not remember why you thought this was a tool for coders.

It never was. It is a tool for builders, and if you have a chore you are tired of, you are one.

Rate this article
0.0 · 0 ratings One rating per email — no sign-in needed
SR
S Ravi Kumar Author of this post.

More Posts

Comments (0)

No comments yet

Be the first to share your thoughts on this post.

Leave a comment

No account needed — just your name and email.

Your email is never published — it is used only for confirmation and moderation.
Generated and checked by this site — no third-party service.
Comments appear after email confirmation and moderation.
An unhandled error has occurred. Reload ×