Choosing Your AI Coding Assistant
What it does
An AI coding assistant writes the code while you describe what you want in plain English: "a drying log where my techs enter moisture readings by room," or "a list of subs whose COIs expire this month." You're the one in charge: you say what the tool should do, check the work, and decide what ships. The options fall into four groups:
- Agents with their own cloud computer (Grok Bot): the agent works on a computer in the cloud with its own browser and command line. It runs the commands and opens the web pages itself, stops for your approval on big steps, and puts your code in your own GitHub and hosting accounts. This is what John uses for his own builds.
- App builders in the browser (Lovable, Replit): you type, it builds, and the app lives on their platform. Easiest start.
- Agents that work in a folder on your computer (Claude Code, OpenAI Codex, Cursor): they change real files you own. More setup on your own machine, more control.
- Helpers inside a code editor (GitHub Copilot): cheapest, but assume you're comfortable in an editor.
What it costs
| Option | Free plan | Paid plans (monthly) |
|---|---|---|
| Grok Bot (Cursor) | Free trial: a usage credit with a 7-day window | Included with Cursor Pro $20, Pro+ $60, Ultra $200, and Teams $40 per user. Weekly usage allowance; extra only if you turn on on-demand |
| Claude (Anthropic) | Yes, no Claude Code | Pro $20 ($17 billed yearly), includes Claude Code. Max $100 or $200 |
| ChatGPT / Codex (OpenAI) | Yes | Go $8, Plus $20, Pro from $100 |
| Cursor | Hobby, limited | Pro $20, Pro+ $60, Ultra $200 |
| GitHub Copilot | 2,000 completions, 50 chat requests | Pro $10, Pro+ $39, Max $100 |
| Replit | Starter: a daily allowance and one published app | Core $20 ($18 yearly), Pro $100 ($90 yearly) |
| Lovable | 5 build credits a day, up to 30 a month | Pro from $25 ($21 yearly), Business from $50 ($42 yearly) |
Most plans limit use within a time window, even when paid. Free plans run out fast on a full app.
A 10-minute setup (Grok Bot)
- Get Cursor Pro ($20 a month) at cursor.com. Grok Bot is included; there's no separate subscription.
- Download the Grok Bot desktop app from the Grok Bot page of your Cursor dashboard, and sign in with your Cursor account. It sets up your first Bot and opens a chat.
- In Bot settings, give it a name, one job, and a short description of how it should work, including: "Ask before you deploy, change an account setting, send a message, or delete anything."
- First job: "I run a water, fire, and mold restoration company and I'm not a programmer. Make a simple page that lists three made-up jobs with a claim number, and show it to me in your browser." Click the computer preview to watch.
- Ask it to explain anything you don't follow.
Using a different tool? Pick one, not three, install it from the official site, and start with the same small first job.
Common mistakes and gotchas
- Pasting secret keys into chat. In Grok Bot, use the secure secret card or the Bot's Secrets section; the value is never shown to the Bot. With other tools, put keys in a settings file that's kept out of GitHub.
- Letting it change everything at once. One change, test it, save it, next.
- Believing "done." Click through it yourself.
- Real customer data in prompts. Use made-up homeowners and claim numbers.
- Prices change often. Check the official page before paying.
Where it fits in your build
Pick the assistant first. It's the tool you'll talk to for every other step: setting up GitHub, putting the app on Vercel, creating the database tables, and wiring up texts, email, or payments. Whichever tool you pick, give it a short written description of your shop and the one job you want the app to do, and keep that description up to date.
Sources
- Claude pricing (Free $0; Pro $17 a month billed yearly, $200 up front, or $20 monthly, includes Claude Code; Max from $100, 5x or 20x Pro usage; five-hour usage windows plus weekly limits on paid plans): https://claude.com/pricing (checked October 7, 2026)
- Claude Max plan (Max 5x $100 a month, Max 20x $200 a month, monthly only): https://support.claude.com/en/articles/11049741-what-is-the-max-plan (checked October 7, 2026)
- OpenAI Codex pricing (Free $0, Go $8, Plus $20, Pro $100, $200, or $500; Business $20 per user a month billed yearly for 2 or more users, $25 billed monthly): https://developers.openai.com/codex/pricing (checked October 7, 2026)
- Cursor pricing (Hobby free; Pro $20, Pro+ $60, Ultra $200, Teams $40 per user; Grok Bot access on Individual and Teams. The visible plan card shows $20 for Pro; the Pro+ and Ultra prices come from the price list built into the same page): https://cursor.com/pricing (checked October 7, 2026)
- GitHub Copilot plans (Free 2,000 completions and 50 chat requests; Pro $10, Pro+ $39, Max $100; AI credit totals $15, $70, $200; individual-plan data used for training unless you opt out): https://github.com/features/copilot/plans (checked October 7, 2026)
- Replit pricing ($20, or $18 billed yearly; $100, or $90 billed yearly): https://replit.com/pricing (checked October 7, 2026)
- Replit plans (Starter free with a daily Agent allowance and one published app; Core; Pro with up to 15 builders and a 28-day database restore window): https://docs.replit.com/help/pricing-and-plans and https://docs.replit.com/billing/plans/starter-plan (checked October 7, 2026)
- Lovable pricing (free plan: 5 build credits a day, up to 30 a month, plus 20 Cloud credits a month): https://lovable.dev/pricing (checked October 7, 2026)
- Lovable subscription plans (Pro from $25 a month for 100 credits, $250 a year; Business from $50 a month, $500 a year): https://docs.lovable.dev/introduction/subscription-plans (checked October 7, 2026)
- Grok Bot plans, usage, and free trial (included on Pro, Pro+, Ultra, and Teams; weekly usage; on-demand monthly limit; trial is a usage credit with a 7-day window; SuperGrok or X Premium+ link): https://cursor.com/help/grok-bot/plans (checked October 7, 2026)
- Grok Bot overview, setup, and working with it: https://cursor.com/docs/grok-bot, https://cursor.com/docs/grok-bot/get-started, https://cursor.com/docs/grok-bot/work (checked October 3, 2026)
- Grok Bot approvals, model selection, data location: https://cursor.com/docs/grok-bot/security (checked October 3, 2026)
- Grok Bot secrets (secret card, Secrets section, saved values can't be read back): https://cursor.com/help/grok-bot/secrets (checked October 7, 2026)
Last checked: October 7, 2026
Full article
What kind of tools restoration and remodeling owners build
The ones that tend to earn their keep are small and specific (judgment, from John's own builds):
- A subcontractor list with COI expiration alerts.
- A job board that shows each loss from mitigation through rebuild, with claim number and adjuster.
- A drying log techs fill in from their phones.
- A change-order log that records who approved what and when.
- A collections list: which deposits, deductibles, and final payments are still out.
Any of the assistants below can help build these. The difference is how much they do for you, and how much of the result you own.
Why John recommends Grok Bot
I started building with AI agents in February 2026, on a platform called OpenClaw. Then I switched to Hermes Desktop and built 18 different applications on it. Later I started working with Grok Bot, and I feel it's a much easier platform for this purpose. My most recent applications were built with Grok Bot.
John H. Hodsdon
What it is, per Cursor's own pages (checked October 3, 2026): each Bot works on a persistent cloud computer with a browser, files, and a command line, and work keeps going when your laptop is closed. It runs on Mac, Windows, and Linux, with an iPhone app. When an action needs your OK, it shows you the action with Allow once, Always allow, and Deny. For sign-ins, it hands you the computer, and it doesn't see your password. API keys go in a secure secret card that keeps the value out of the chat and away from the Bot. Plugins connect it to services such as GitHub, Vercel, and Neon, and skills let it reuse a saved set of instructions, like a written description of your shop and your project.
The trade-offs:
- It's a subscription. Grok Bot comes with Cursor's paid plans. Pro includes a weekly usage allowance. When it runs out, Grok Bot stops until the weekly reset unless you turn on on-demand usage, which is billed up to a monthly limit you set.
- You don't pick the AI model. Cursor manages that. Simpler, but less control.
- Your project lives on a cloud computer. It keeps your files, but Cursor's help pages say a recover or reset can lose very recent changes. Push to GitHub often.
- Plugins can have bad days. Cursor acknowledged a Vercel plugin connection problem in September 2026. Everything can also be done on the service's own website.
- It runs in Cursor's cloud, on computers in the United States today. Read its data settings before you put real customer information anywhere near it.
The four groups, in more detail
Agents with their own cloud computer (Grok Bot). The agent has its own browser and command line, so it does the clicking and typing you'd otherwise do yourself: creating the project, running it, pushing to GitHub, and setting up hosting, with your approval on the big steps. The code still ends up in your own GitHub and Vercel accounts, and the app keeps running if you stop paying for Grok Bot.
App builders in the browser (Lovable, Replit). You describe the app in a chat box and watch it appear. Hosting is usually included. This is the fastest way to see a working screen. The trade-off is that your app lives on their platform. Lovable says you own your code, but moving it somewhere else later is still work.
Agents in your own folder (Claude Code, Codex, Cursor). These read and write real files on your computer, run commands, and fix their own errors. You then put the code on GitHub and host it on Vercel, under your own accounts. You own every piece, but the setup on your own machine is yours to do.
- Claude Code comes with Claude Pro and Max.
- Codex comes with ChatGPT plans.
- Cursor is a code editor with an agent built in.
Editor helpers (GitHub Copilot). Suggests code as you type and answers questions inside an editor. Cheapest paid option, but it assumes you're already comfortable in a code editor.
Prices, in more detail
- Grok Bot: included with every paid individual Cursor plan (Pro $20 a month, Pro+ $60, Ultra $200; Pro+ and Ultra give more weekly usage) and with Cursor Teams ($40 per user a month). An individual SuperGrok plan (SuperGrok, Plus, or Heavy) or X Premium+ can be linked for usage instead; linking doesn't add usage on top of a Cursor plan. The free trial is a usage credit with a 7-day window, and one large task can use most of it.
- Claude: Free $0 (no Claude Code). Pro $20 a month, or $17 a month billed yearly ($200 up front), includes Claude Code. Max is $100 a month for 5x Pro usage or $200 a month for 20x, billed monthly only. Usage limits reset on a rolling five-hour window, and paid plans add weekly limits.
- ChatGPT / Codex: Free $0, Go $8, Plus $20, Pro from $100 a month (tiers at $100, $200, and $500). Business, for 2 or more users, is $20 per user a month billed yearly or $25 billed monthly. Codex is included in the plans.
- Cursor: Hobby free with limited agent requests, Pro $20 a month, Pro+ $60, Ultra $200, Teams $40 per user a month.
- GitHub Copilot: Free (2,000 completions and 50 chat requests a month), Pro $10, Pro+ $39, Max $100 a month. The plans page also lists credit totals of $15, $70, and $200; those are included usage amounts, not prices.
- Replit: Starter is free, with a daily Agent allowance, a small monthly cloud credit, and one published app that goes down after 30 days. Core is $20 a month ($18 billed yearly) and Pro is $100 a month ($90 billed yearly).
- Lovable: free plan with 5 build credits a day (up to 30 a month) plus 20 Cloud credits a month. Pro starts at $25 a month for 100 credits ($250 a year), and Business starts at $50 a month ($500 a year). Both come in bigger credit tiers.
Which fits which owner (judgment, not a ranking)
- "I want to see something work this weekend." Lovable or Replit. Know that moving off their platform later takes work.
- "I want to own the code and have the agent do the clicking and typing." Grok Bot.
- "I want to own the code and do the setup on my own computer." Claude Code, Codex, or Cursor. Pick one and stick with it through the whole build.
- "I already pay for ChatGPT or Claude." Try the coding tool that comes with it before buying another subscription.
- "I'm comfortable in a code editor and want the lowest bill." GitHub Copilot.
- "My coordinator will be the one building." Start with whoever's going to do the work. A tool they're comfortable in beats a better tool they avoid.
The full setup, step by step (Grok Bot)
- Get Cursor Pro at cursor.com ($20 a month). Grok Bot is included.
- Install the desktop app from the Grok Bot page of your Cursor dashboard. Download it from there only.
- Sign in with your Cursor account. There's no separate Grok Bot login. Your first Bot opens in a chat.
- Set up the Bot. Click its name, choose Bot settings, and give it a name, one job, and a description: "I run a water, fire, and mold restoration company. I'm not a programmer. Explain things in plain language. Show me each command before you run anything that installs, deletes, or changes an account. Ask before you deploy, change an account setting, send a message, or delete anything."
- Give it a first small job. "Make a simple page that lists three made-up jobs with a claim number and status, and show it to me in your browser." Click the computer preview and watch what it does. Ask why.
- Save keys the safe way. When it needs an API key, it shows a secure secret card. Paste the value there, never in chat, and keep your own copy in a password manager; saved values can't be read back.
- Leave your own computer out of it unless you need it: set Settings, Computer, Execution on this computer to Never.
- Watch usage. Check the plan screen after the first week. Leave on-demand off, or set a low monthly limit, until you know what a week of building uses.
The full setup, step by step (folder-based agent)
- Choose one tool and sign up.
- Install it. For a desktop app, download from the official site only. For a command-line tool, follow the official install page.
- Make a project folder, for example
Documents/restoration-job-tracker. No spaces or capital letters in the folder name saves trouble later. - Open the folder in the tool.
- Tell it who you are. "I run a water, fire, and mold restoration company. I'm not a programmer. Explain things in plain language and tell me exactly what to type when I need to type something."
- Give it a first small job. "Make a simple page that lists three made-up jobs with a claim number and status." Watch what it does. Ask why.
- Set a spending limit on your plan, if it offers one, and check usage after the first week.
More gotchas worth knowing
- Data settings. GitHub says individual Copilot plans may use your data for training unless you opt out. Read each tool's data settings before you paste in anything about real customers.
- Customer information doesn't belong in prompts. Homeowner names, addresses, claim numbers, and adjuster emails should be made up while you build.
- "Done" isn't tested. Assistants sometimes say something works when it doesn't. Open the page and try it like your tech would.
- Big changes hide mistakes. Ask for one change at a time and save it to GitHub before the next.
- Switching tools mid-build costs time. Each tool has its own habits. If you switch, give the new one a written summary of the project first.
- Usage limits hit at bad times. A storm week isn't the time to discover your plan's weekly limit. Know where your usage meter is.
How to use it on a first project
Start with one small, specific tool, like a subcontractor list with COI expiration dates. Set up the assistant's rules once (ask before deploying, changing an account setting, sending a message, or deleting anything), then work one change at a time and save each working version to GitHub. Every prompt in the other guides works in any capable assistant. A plain chat assistant can't run commands for you, so you'd do that part yourself.
