jared hebb ~ %

Work

Things I’ve built with AI

Real software and products, shipped and in use. Each one is a project I ran end to end without writing the code myself.

Live web app

electricaltoolbox.com

A growing library of NEC calculators. Runs solo, no code.

Visit →

Book · Amazon

NEC Code Quickstart

The twelve exam calculations worked start to finish, each with the exact code reference. Kindle and paperback. The interior layout and print cover came out of a publishing pipeline I built with AI, not a designer.

On Amazon →

Free download

The Systems Audit Kit

The systems audit a consultant charges $500 for, run yourself in one morning. 27 questions about your clients, your tools and your time, a way to sort the answers into fix now, fix soon and skip on purpose, and a two-page plan you fill in and pin up. 25 pages, built to print at home.

Get it →

Free tool

First message

Most of what goes wrong with Claude goes wrong in the first message. You type a sentence and a half of what you want and Claude guesses the rest, confidently. Answer a few plain questions and get a first message worth sending, with the parts people always leave out already written in.

Open →

Free tool

Claude house rules

Claude reads one file before it touches your project, and if nobody wrote it, Claude makes up its own rules. Answer a few plain questions and download that file. No code knowledge needed, no signup, and nothing you type leaves your browser.

Open →

Free tool

Claude settings

Claude asks permission before it does almost anything, and saying yes forty times a day trains you to stop reading the one that mattered. Tick what it can get on with, what it has to ask about, and what it must never touch, then download the settings file. Nothing you type leaves your browser.

Open →

Free tool

Cost estimate

Everyone asks what Claude costs to run before they build the thing, and every answer online is written in units nobody bills their own time in. Answer four questions in plain words and see the monthly bill on four current models side by side, plus the two things that halve it and whether they would actually work for you.

Open →

Free tool

Claude helper

Claude can hand a whole kind of work to a specialist you set up once, called a subagent. A reviewer that only reads, a helper that writes your release notes, a researcher that goes and looks things up: each one works in its own space and hands back just the result. Answer a few plain questions and get that helper as a file.

Open →

Free tool

One-word command

If you find yourself typing the same instructions to Claude over and over, save them once. A command is a shortcut: you type one word, like /recap or /standup, and Claude runs the whole set of instructions you saved behind it. This builds you that command as a file.

Open →

Free tool

Answer style

If you keep asking Claude to be shorter, or to lead with the answer, or to stop being so chatty, set that once instead of every turn. Pick a tone and a format and this builds you that saved style as a file.

Open →

Free tool

Status line

Claude Code can show a line of live information at the bottom of the screen: which model you are on, the folder you are in, how full the context window is, what the session has cost. Setting that up means writing a little script. Tick what you want to see and this writes the script for you.

Open →

Free tool

Connect a tool

MCP is how Claude Code reaches an outside tool: your issue tracker, a database, a hosted service. Setting one up means adding a small block of settings, and there is one easy slip that quietly stops it from loading. Answer a few plain questions and get a block that is right by construction.

Open →

Built with Claude, not for sale

The projects above are finished products with a price or a listing. Most of what I actually build is neither: personal tools I run every day, with no storefront and often no public URL. Here is the rest of it, honestly, with what each one proves rather than a number I do not have.

Personal app · not public

Meal planner

An Astro app that plans a week of dinners around what is actually in the fridge, scores recipes by rating and protein rotation, and builds the shopping list. Approving a menu pushes a full email (menu, shopping list, prep plan) and a calendar entry through my own Google account. Built to solve a real weekly chore for my own household, not as a demo.

Proves: a full app with real Google OAuth and a scoring algorithm, with no LLM call at menu-generation time.

Desktop app · not public

Watson

A voice-driven Electron companion I talk to instead of typing. It runs on the Claude Agent SDK, holds a persistent session, and has its own rule for what it is allowed to do without asking first, separate from what it has to bring back to me before acting.

Proves: a production desktop app built around an explicit tool-permission boundary, not a chat window bolted onto an API call.

Internal tool · not public

Engage

The dashboard behind my own social and video work: drafts land here for me to approve line by line before anything posts, metrics and follower counts are tracked over time, and a video pipeline turns raw footage into captioned, scheduled posts. Nothing publishes without a decision recorded here first.

Proves: a real internal tool with its own test suite, an authentication layer, and a nightly database backup, built and kept running rather than left as a one-off script.

Node CLI · not public

Ebook pipeline

The production layer behind NEC Code Quickstart, above: a command-line tool that turns a Markdown manuscript into an EPUB, a print-ready PDF with correct spine width, and both covers, using the Chrome already installed on the machine instead of a paid rendering service.

Proves: the book was a pipeline, not a one-time project. The same tool is what the next book in the series runs through.