How I Set Up Claude Code as a Daily Pair Programmer (Without Burning the Budget)

Luna · AI Tinkerer — Fri, 12 Jun 2026 06:28:50 +0000

I went from "I'll just try it once" to "I have a whole process around it" in about three weeks. Here's the short version of what survived contact with real codebases.

1. One project, one `CLAUDE.md`, not a global bible

The default advice is "put a giant CLAUDE.md in your home directory." That sounds great until Claude Code starts apologizing for things in your side project while you're working on a Node script for work. A per-project CLAUDE.md is more honest. Mine is short:

Stack + Node version, package manager, formatter, test command.
Two or three "don't do this" lines (don't change lockfiles, don't add dependencies without asking, don't write tests for files I didn't ask about).
A short "things I always forget" section. Mine says "this repo uses the new auth flow, not the legacy one."

That's it. ~30 lines. The shorter the better — long files become a tax on every single turn.

2. Treat it like a junior with a very long memory

Claude Code is good at pattern-matching across a whole repo. That means it will happily refactor a function and ten other functions that "look similar" if you don't fence the scope. Two habits saved me:

"Look at this file. Do not edit any other file." Sounds restrictive, but for a 200-line patch it works.
"Before editing, list the files you would touch." If the list is longer than I expected, I cut the request.

3. The real cost isn't the API — it's the context

Most people I know hit the budget limit not because Claude Code is slow, but because they paste the whole 800-line file into the prompt "just in case." Three tricks cut my token use by roughly half:

Paste the failing test, not the implementation. The error message is usually enough.
Tell it to read the file first, then propose a diff, then apply. The pause lets me catch a wrong direction early.
/clear between unrelated tasks. The longest "I lost an hour" stories start with "it kept bringing up stuff from two requests ago."

4. Don't ask for tests you won't read

Auto-generated tests are seductive. They look productive, they run, the bar goes green. But tests Claude Code writes for me tend to test the implementation, not the contract — so they pass when the code is wrong. The pattern that works for me: I write the test first (one or two cases), then ask Claude Code to either implement against it or extend it. The test is the source of truth, the model fills in the body.

5. The one thing nobody mentions

The thing that actually moved the needle was using /rewind to recover from a bad turn. I expected to use git for that, but /rewind is faster for a single chat turn. It also teaches me what kind of prompt tends to go off the rails — I now know that "refactor this" is dangerous, and "rename this function and update callers" is safe.

Verdict

It's not magic. It won't write your app. But for the boring 80% — the "wire this up to that", the "explain this weird error", the "write a tiny script to massage this CSV" — it's the first tool in a while that's actually changed my daily loop.

What I'd Actually Tell a Friend About Free AI Tools in 2026

Luna · AI Tinkerer — Fri, 12 Jun 2026 01:58:21 +0000

What I'd Actually Tell a Friend About "Free" AI Tools in 2026

A practical, not-cheerful guide to the four or five actually-useful no-cost AI offers right now, and the four or five that sound free but quietly aren't.

I keep getting asked which AI tools are "really free" right now, and the honest answer is: most of them are free the way a sample at Costco is free — you get a taste, then you stand in line to buy the whole rotisserie chicken. This post is the short version of what I'd tell a friend over coffee.

The genuinely useful free tiers in 2026

Local models via Ollama / LM Studio + a 7B/8B model. This is the only "free" that doesn't have a meter running. A decent quantized Qwen or Llama 3 model on a laptop with 16GB of RAM is more than good enough for: summarizing text, drafting emails, rewriting code comments, generating test data, asking "what does this error mean." It's not as good as a frontier model for hard reasoning, and you can't browse, but it costs $0 forever and the privacy story is a feature, not a footnote.
GitHub Copilot free for VS Code. The free tier is genuinely useful for autocomplete and small in-editor refactors, not for whole-file generation. If your expectation is "tab to complete this line, occasionally expand to a function," you'll be happy. If you expect "write my whole service for me," you'll burn through it in an afternoon.
Google AI Studio (Gemini flash) and OpenAI's playground "free" tier. Both let you hit a frontier model with a generous-enough rate limit that you can do real work — drafting, summarizing, code review — without paying. The catch: the limits move. What was free in March might be 50% smaller in April, and there's no warning. Treat these as "free for this month's project," not "free forever."
Hugging Face Inference API free tier. A handful of models per day, useful for embeddings, small classification, the occasional OCR-style task. Best for backend devs who want a quick embedding endpoint without standing up a vector DB.
Cursor's free trial and the various "free credits" programs. Worth using once, not worth building a workflow around.

What "free" actually means in 2026

Free tiers come in three flavors, and the marketing copy rarely tells you which one you're getting:

Hard quota, soft cap. Gemini, OpenAI, Copilot. You get a daily or monthly limit, the page tells you when you're close, and going over is a soft "wait until tomorrow." Most useful for ongoing work.
Time-limited trial, hard cap. Cursor, certain Pro features, "free for the first 30 days." Useful for evaluating, not for daily use.
Bring-your-own-key with a markup. Some "free" tools route your requests through their own API key, which means you're paying (through them) for the model. The free part is the chat UI. If you can't see the token usage and the cost per request, you're not really free — you're on someone else's tab.

The four traps I keep seeing people fall into

Auto-renewal after a free trial. Two tools I won't name had this in 2025; one of them still has it. Set a calendar reminder the day you sign up. If the cancellation flow is more than three clicks, that's a signal.
"Free for students / non-profits / OSS maintainers" — until verification expires. Some programs do this well (GitHub Student Pack is the gold standard). Others quietly let your access lapse and don't tell you.
"Free" in-app purchases that aren't free. The model is free, but the storage, the team seats, the export-as-PDF feature is $9/month. Read the pricing page, not the landing page.
Free model, paid context. Some "free" chats quietly upload your conversation to train the next model. This is a feature for the vendor, not for you. If your work touches anything sensitive — internal docs, customer data, code under NDA — local is the only honest free.

My actual stack right now

Daily workhorse: local Qwen 8B through Ollama, for code review and quick questions. Cost: the electricity to keep the laptop plugged in.
Hard problems: Gemini flash via AI Studio, free tier. Cost: occasionally hitting the rate limit and waiting 10 minutes.
In-editor: Copilot free, mostly autocomplete.
Backups: a bookmark to the local Ollama install, because nothing on this list is "free" in the sense of "I'm confident it will be here in two years."

If you're starting from scratch today, I'd install Ollama first, fight with the model download for ten minutes, and then try everything else. The cost curve of "free AI in 2026" goes from "free if you run it yourself" to "free if you're careful" to "free if you read the fine print" — in that order.

DEV Community: Luna · AI Tinkerer