AI tools for your business, work, and life

This is mainly for people who are newer to AI or to technology. It covers the vocabulary, how to choose a tool, how to get better results, the details worth your attention, and what you could build.

I'm an ultra-generalist. I've spent ten years building software products, but I'm not an engineer by training. I was a Founder Fellow at South Park Commons, and before that I worked on new technology products for riders at Lyft, including as a founding member of the bikes and scooters team.

You don't need a specific degree or pedigree to build software with AI. You need to be curious and have an appetite to learn. As a product manager I could describe a product really well, but I could never build it myself. I still work with my cofounder and CTO and our product designer, and now I contribute code directly too. I've built websites, mobile apps, a Chrome extension, Claude skills, and an AI running coach for my iPhone.

A large language model, or LLM, is trained on lots of data and predicts what should plausibly come next. It doesn't reach into the internet unless you give it access, and it can only see what you connect it to, like files on your computer or pages online. Even people in the industry use model, LLM, and AI interchangeably, so if you aren't sure what someone means, ask.

An agent has your permission to do things, like open a file, run a command, or go through this quarter's invoices and flag anything odd. It isn't deterministic, so you get slightly different results each time. That brings some creativity, which can be good or bad, so double-check the facts and sources in something like a research report. An automation follows fixed rules. An invoice arrives, it gets filed, and a template email goes out the same way every time. You can also combine the two.

Companies like Anthropic, OpenAI, and Google make models, and they also build tools on top of them. Claude, Claude Code, and Claude Cowork come from Anthropic. ChatGPT, Codex, and Sora come from OpenAI, which makes the GPT-5 series of models. Google has Gemini, NotebookLM, and many more, plus models like Nano Banana for visuals. There are independent tools too, like Lovable, Vercel's v0, and CodeYam, and some of them let you work with models from more than one company.

A chat in your web browser mostly sees what you paste or upload into it. A desktop app, like ChatGPT's or Claude's, can reach files on your own computer if you give it permission. A browser extension can browse and test web pages for you. The terminal looks more technical at first, but it's a really powerful place to direct and manage agents using command-line interface (CLI) tools. In every case, you control the permissions.

Lovable and v0 are great for things like static websites. You describe what you want in plain English and get a working site back. Claude Cowork, which today you'd probably run from the Claude desktop app, can create automations and skills for business use. Cursor lets you run several agents at once in an editor, which suits people who have used editors like VS Code. Claude Code and Codex run in editors, in the terminal, and in desktop apps, and both have mobile apps too. Things change quickly, so do your own research.

Set a time limit on your research, then pick one tool and give yourself time to go deep with trial and error. I use the Claude desktop app heavily, Cowork for presentations and automations, CodeYam with Claude Code to build and ship software, CodeYam and sometimes v0 for design, Gemini in search results, ChatGPT for quick copy edits, and Riverside to record podcasts. The biggest gap is between using no tools and trying your first one.

Your screen won't look exactly like mine, and that's normal. Written guides go stale quickly because tools change and setups differ. Everyone works by trial and error, professional engineers included. If something looks different, paste what you saw into the chat (the exact words if you can) along with what you expected and your question. Still stuck? Email me or ask in the CodeYam Discord.

There's a false belief that one perfect, detailed prompt will get it right in one shot. You have context the agent doesn't, and it has context you don't, so expect to iterate quickly. Brief it like a capable new colleague, somewhere between too vague and hundreds of pages. Give key context up front, say what you don't want (like a generic purple website), and ask for the simplest version that works before building from there.

Compare "Build me a booking site" with "I'm a coach with about fifteen clients. I want one page where somebody picks a 30- or 60-minute session from my free slots and pays a deposit to hold it. Start with the simplest version that works, and tell me what you're assuming." The first leaves the agent guessing what a booking site is, so you may get features you never wanted. The second gives it the context you already have.

A model can give a confident answer that's wrong. Check any number or fact you can and find out where it came from. It can't pull data that doesn't live somewhere it can connect to, and it can't read what's in your head or in a spreadsheet you haven't shared. It also tends to follow the path you give it, which may not be the best one, so ask questions that get it to rethink the approach.

Everyone gets stuck, engineers included. Before asking for a fix, ask the AI what happened and where it went wrong, so you fix the root cause instead of adding a Band-Aid. Paste errors back in with what you expected to happen and ask it to explain them until they make sense. You can also go back to the last version that worked and find where things changed.

What matters depends on you and your business. Check whether your chats are used to train models, since defaults vary by tool and plan. Check whether memory is on between chats. Think about how broad file access is and whether the tool can only read your files or also change them. And know whether the work runs on your own machine or on someone else's server.

If you handle other people's information, check your privacy settings before you start. If you work at a bigger company, learn its AI policy. If you work with others, share context with your human teammates, not just your agents. I use Claude artifacts to show our team what I've built. For any of these, you can ask the AI to help.

You can now do work that used to need a specialist or a bigger team, like a landing page, a booking flow, a simple dashboard, or a script that runs every week without you. Bringing an idea to life, even one you end up dropping, shows you what the real problem is. It also helps you brief an expert better if you decide to work with one.

Apps are great, but they aren't always the right place to start. Begin with the problem and the smallest, leanest thing that solves it. An automation is a rule, like emailing you when certain data comes in. A skill is a set of instructions for your agents that you can reuse. A connection lets an agent like Claude reach a data source or another app, which is what MCP, MCPB, and MCP servers are about.

Cost shouldn't stop you from learning to build. A static site, where everyone sees the same page, is often free or very cheap to host. A dynamic site or web app with logins, a database, and more features can cost more because you may pay for a database or a login service. A mobile app adds app store fees, review times, and a separate process to build and submit iPhone and Android versions.

In most cases a mobile app has to pass app store review before people can download it. Your first review can take a while, and the app may be rejected for reasons you don't expect. A website goes live when you decide, with no outside reviewer, so changes ship quickly. If you build a mobile app, start small, ship often, keep it easy to review, and use TestFlight or Google's equivalent to test with friends and family.

The more I build, the more ideas I have, so write yours down. Whether your vision is small or big, pick a bite-sized piece you can picture someone using. Who is the user? What do they see first, and what do they do? What problem does it solve for them? If you don't know yet, describe it more loosely. A vague first idea may come back generic, and that's normal. Use it to add detail and try again.

Join the free CodeYam Discord, where people are already building with CodeYam and Claude. You can also email us, follow the blog, keep going here in CodeYam Academy, or reach out to me directly.