Introduction
When you start developing software with AI coding tools, you quickly notice that your AI agents keep suggesting platforms you’ve never heard of, like Modal, Inngest, Neon, Upstash, Fly.io and Cloudflare R2.
You’re a small business owner or a no-code person who knows WordPress and maybe some Airtable. Suddenly your agent is telling you to “spin up a Cloudflare Worker” and asks for your permission to bash, grep or curl. It’s easy to get overwhelmed.
So here’s a quick field guide to the platforms I actually use across my Zero2Webmaster projects, grouped by what they do. Written with a goal of giving you just enough information so that the next time your AI agent suggests one, you’ll know the basics of what it’s talking about. And as for bash, grep and curl…I’ll save those for a future article 🙂
One thing before you read on: the time to figure out which of these tools your project needs is before you start coding. I always develop a project idea in conversation with AI first — usually in Perplexity, sometimes in Claude desktop or mobile — and write up a Project Instantiation Prompt (think of it as a lightweight PRD, or Product Requirements Document — the spec a product manager writes before development starts) before I open my IDE. That conversation is where most of these tool choices get made. The article you’re reading exists so that when the AI suggests Inngest or Neon or Cloud Run during that planning conversation, you have a fighting chance of understanding why — and maybe pushing back if you think you know better.
Envato Elements search: “developer workstation modern technology platforms” — or Z2W AI Suite prompt: “Wide hero image of a small business owner at a clean desk with a laptop, surrounded by softly glowing abstract platform logos and infrastructure diagrams floating in the air around them, modern flat illustration style, green and gold accents on white background, conveying curiosity and overwhelm in equal measure.”
Backups & Version Control
Why this comes first: your code lives in a folder on your computer, but virtually every AI coding project also keeps a synced copy in a Git repository — almost always on GitHub. Git is a version-control system (it tracks every change you’ve ever made and lets you roll back to any point), and GitHub is the most popular cloud service for hosting those Git repositories. The two copies stay in sync, so GitHub doubles as a continuous off-site backup. AI coding tools rely heavily on Git under the hood, which is why this category sits at the top of the article.
GitHub
The cloud home for your code. Free for public and private repositories, integrates with everything, and serves as a continuous off-site backup of every change you’ve ever made. You don’t strictly need GitHub to write code — it lives on your computer either way — but AI coding tools (Cursor, Claude Code, Codex, all of them) assume your project is in a Git repository, and GitHub is the path of least resistance for that. If you only adopt one tool from this entire article, it’s GitHub. We’ll come back to it in a moment when we talk about GitHub Actions.
Database & File Backups
Code backups are easy (push to GitHub). Database backups are the part people forget. Most modern database providers include automated backups and point-in-time recovery in their free or paid tiers, which covers the basics. For extra protection — and for backing up files, exports, and assets that don’t live in a database — Cloudflare R2 is the standard target: cheap, S3-compatible, and you almost certainly have a Cloudflare account already. Set up a scheduled GitHub Action or cron job to dump and push, and you’re done.
Compute & Serverless Functions
What is compute? Anywhere your code actually runs. Serverless means you don’t manage the server. You write a function, hand it to the platform, and it runs when something triggers it (a schedule, a webhook, an event) and shuts off when it’s done. You pay for the seconds you use, not a server sitting idle 24/7. A big factor in choosing between these tools: how often will the thing get pinged? Constant traffic favors always-on; bursty or scheduled work favors scale-to-zero.
GitHub Actions
Since your code already lives in GitHub (see above), Actions is the natural place to run scripts and workflows for free, within generous limits. Originally built for testing and deploying code, but I use it constantly as a free cron job runner — scheduled tasks, scrapers, automated content jobs, database backups. The killer feature: it’s already wired into your repo, so deployment is just a configuration file (specifically a YAML file, which is a human-readable text format for configuration — you don’t really need to know more than that).
Modal.com
Python-focused serverless compute, built for working with AI models and large datasets. The difference from GitHub Actions: Modal gives you on-demand access to GPUs (graphics processing units — the specialized chips that AI models need for fast computation), sub-second cold starts, and you define infrastructure inside your Python code instead of separate config files. When I need to run something heavy — batch image generation, processing large datasets, running an AI model over thousands of inputs — Modal is the move. Free tier includes monthly compute credits.
Inngest
Inngest is a workflow orchestrator. Where GitHub Actions runs a script top to bottom, Inngest runs durable step functions — meaning if step 3 of 7 fails, it retries just that step and picks up where it left off. Built for event-driven workflows, background jobs, and the kind of multi-step AI workflows that have to be reliable. Think: “when a user signs up, send email, wait 3 days, check if they did X, then do Y.”
Cloudflare Workers
Tiny pieces of code that run on Cloudflare’s edge network — meaning physically close to your visitors, anywhere in the world. Ideal for low-latency API endpoints (an API, or Application Programming Interface, is just a way for one piece of software to talk to another), request rewriting, geolocation logic, AI inference at the edge. Different from Modal in that Workers are designed to be small, fast, and responsive (milliseconds), not long-running compute jobs.
Google Cloud Run
Container-based serverless. (A container, usually built with a tool called Docker, is your app bundled together with everything it needs to run — code, libraries, operating system bits, all in one portable package. The point is that the same container runs identically on your laptop, on Cloud Run, on Fly.io, anywhere.) You hand Cloud Run a Docker image and it runs your container, scales it up under load, and scales it back down to zero when traffic stops. I use it in the Z2W Multi-Lingual translation queue: requests come in bursts, the container spins up, processes the queue, and shuts back down. This is the right tool when you have an existing app (anything that runs in a container) but don’t want to pay for it to sit idle. You can also configure minimum instances if you need it always warm.
Application Hosting
What is hosting? The home for your whole app or website. Unlike compute (which runs short jobs), hosting keeps your application available 24/7. The platforms below differ mainly in what they host best — static sites, full apps, or traditional servers — and how much of the infrastructure they manage for you.
Vercel
The default home for Next.js apps and modern JavaScript frontends. Push to GitHub, Vercel builds and deploys automatically with a live URL. Excellent for marketing sites, web apps, and anything React-based. Generous free tier; gets expensive at scale.
Cloudflare Pages
Vercel’s main competitor for static and JAMstack sites. Pages is part of Cloudflare’s broader ecosystem, so it pairs naturally with Workers, R2 storage, and D1 databases. Generally cheaper than Vercel at scale and benefits from Cloudflare’s global network.
“But I’m a WordPress person — why would I ever use Vercel or Pages?” Fair question. The answer: not every project is a WordPress project. If a client needs a marketing microsite, an interactive calculator, a lightweight web app, or a tool that just doesn’t belong in a WP admin, Vercel or Cloudflare Pages is often the cleaner home. Pages also makes sense for the static side of a hybrid setup — a fast frontend on Pages talking to a WordPress REST API on the back end. Knowing both worlds means you can pick the right tool instead of forcing every project into the WP-shaped hole.
Fly.io
For when you need a real application — Docker container, full backend, persistent connections, custom runtime — running close to your users globally. Fly runs your container in multiple regions and routes users to the nearest one. Use it when Vercel or Pages isn’t enough because you need a real server process, not just a function.
Hetzner
A traditional German hosting company with absurdly cheap virtual machines. A Hetzner VPS (Virtual Private Server — basically a slice of a real computer that’s all yours) is a Linux box you connect to remotely via SSH (Secure Shell, a command-line connection method) and manage yourself — the opposite of serverless. I use one for projects where I want full control and predictable monthly pricing instead of usage-based billing. Trade-off: you manage updates, security, backups.
Databases
What’s a database here? If you’ve used Airtable, you already know what a database is — a structured place to store data. The difference: Airtable is a friendly interface bundled with a database. The tools below are just the database, which your code talks to directly. More power, no built-in UI.
Neon
Serverless PostgreSQL. Same Postgres database that’s powered serious apps for 30 years, but reimagined so it scales to zero when idle (you stop paying) and spins up in milliseconds when needed. The killer feature is database branching — like Git, but for your data. Want to test a migration? Branch the database, try it, throw the branch away. Free tier is genuinely usable.
Upstash
Serverless Redis (and a few other data stores). Redis is a fast in-memory database, ideal for caching, rate limiting, session storage, and queues. Upstash gives you Redis with pay-per-request pricing and an HTTP API, so it works from serverless functions where traditional Redis connections are awkward.
Baserow
The Airtable alternative I reach for when row counts get high or API call limits become a problem. Baserow is open source and self-hostable, with no row caps and no metered API calls — both real ceilings on Airtable’s paid plans. Honest admission: setting Baserow up used to be a slog before AI tools made the install and config tractable. Now it’s a viable replacement for Airtable in any project where you’ve outgrown Airtable’s limits or want full ownership of the data layer.
“Wait — Neon and Upstash aren’t open source. Aren’t I just creating vendor lock-in?” Worth thinking about, but in practice the risk is small. Neon is Postgres under the hood; Upstash is Redis. If either company ever made the platform unworkable, you could export your data and stand up a self-hosted Postgres or Redis instance on Hetzner in a weekend. That’s a very different lock-in profile from, say, building everything on a proprietary platform with no escape hatch. Combine that with genuinely generous free tiers, and the trade is usually worth it.
Storage
What’s object storage? A place to dump files — images, videos, PDFs, backups — that your application or website serves. Cheap, durable, and S3-compatible (meaning most tools written for Amazon S3 will work without modification).
Cloudflare R2 & Bunny.net Storage
These two are close cousins for object storage. Both are S3-compatible at a fraction of what you’d pay AWS, both have generous free tiers, both are battle-tested. Cloudflare R2 is the natural choice when you’re already in the Cloudflare ecosystem (Pages, Workers, Images) — zero egress fees and tight integration. Bunny.net Storage (worth noting: this is different from Bunny Stream, which is their video hosting product — you may already know Bunny for video) is the right pick if you’re already using Bunny for video, want to keep your storage in one vendor, or simply don’t want to set up another Cloudflare account. For pure file storage either works; for video specifically, Bunny Stream is purpose-built and hard to beat.
Monitoring
What’s monitoring? Knowing when your stuff breaks before your customers tell you. The internet is full of paid services for this; the one below is the open-source option I run myself.
Uptime Kuma
Self-hosted, open-source uptime monitoring — the obvious alternative to UptimeRobot, which recently changed its terms to disallow business use on the free plan. (A nice illustration of why open source matters: proprietary vendors can change terms, pricing, and feature availability whenever they want. Open-source tools can’t pull that rug out from under you.) Uptime Kuma points at your sites and APIs, pings them on a schedule, and notifies you (Slack, email, push, whatever) when something goes down. Free, looks great, runs in a Docker container. Mine runs on Fly.io and monitors all my projects and client sites.
Envato Elements search: “server monitoring dashboard green status uptime” — or Z2W AI Suite prompt: “Clean modern dashboard mockup showing rows of green status indicators for various monitored web services, soft drop shadows, professional UI design, white background with Z2W green (#82C92B) accents, conveying reliability and peace of mind.”
Old Players Still Doing Real Work
One of the misconceptions about AI-era development is that it requires abandoning the tools you already know. It doesn’t. Some of the most reliable infrastructure in my projects is software that’s been around for years.
WordPress
Still the right answer for most content-driven sites, especially when the client needs to edit their own pages, run a blog, sell products, or manage a membership. AI tools haven’t replaced WordPress — they’ve made it dramatically easier to extend it with custom plugins and integrations, which is exactly what we do at Zero2Webmaster. One related tool worth knowing: Local by Flywheel (or just “Local”) is a free desktop app that runs a real WordPress site on your own computer, which lets your AI’s plugin updates appear instantly without uploading new plugin zip files to a live site every time. I keep a single Local site where all my plugin development gets tested.
Airtable
Still excellent as an operational database — for tracking projects, content calendars, client work, anything where you need a fast UI on top of structured data. I use it for business operations daily. The reason to reach for Baserow instead is purely scale: high row counts or unlimited API calls.
Amazon SES
Transactional email — receipts, password resets, notifications, automated workflow emails. SES is dirt cheap (fractions of a cent per email) and reliable. Most “modern” email services are just nicer interfaces on top of underlying infrastructure like SES. For raw delivery at scale, it’s hard to beat.
The Terminal
You’re going to hear several words that all refer to roughly the same thing: terminal, shell, command line, sometimes SSH (which is specifically a way to connect a terminal on your computer to a server somewhere else). It’s the black-or-white text window where you type commands instead of clicking buttons. The terminal is older than the personal computer itself — it’s how people used computers before graphical interfaces existed, and it’s still how serious infrastructure work gets done. Cursor has one built in; your computer has one too (Terminal on Mac, PowerShell or Command Prompt on Windows). You do not need to understand much about it. The pattern in AI-powered work is simple: the AI tells you what to paste in, you paste it, you copy the output back into your AI chat so it can read what happened. That’s 90% of terminal use in practice. The other 10% is muscle memory you’ll pick up without trying.
Cloudflare
Cloudflare has shown up several times in this article already (Workers, Pages, R2), but it deserves its own mention as the place I park every domain I touch. Nobody makes DNS management easier or cheaper. (DNS, or Domain Name System, is the internet’s phone book — it’s how typing yourdomain.com actually lands a visitor on your server.) Cloudflare’s registrar pricing is at-cost, and the dashboard is the most usable in the industry. If you only adopt one new tool from this article, make it Cloudflare for your domains and DNS — the rest of the ecosystem becomes much easier to use once you’re already there.
The IDE Layer: Cursor + Claude Code
Cursor is the IDE (Integrated Development Environment — basically the editor where you write and run code) I use to do the actual work. If you’re new to AI-powered coding, my overview of Cursor covers the basics. Cursor gives you direct access to the major AI models out of the box, which is plenty for most people.
What I do that’s a little different: I installed the Claude Code extension inside Cursor and connected it to my Claude subscription. As a nonprofit (SAVE THE FROGS!), I get 6.25x the Claude Pro usage on the nonprofit plan — enough that I can work essentially all day on Opus 4.7 (currently Anthropic’s most advanced and most expensive model) without hitting limits. That setup saves me hundreds, sometimes thousands of dollars a month compared to using Cursor’s Ultra plan with Opus.
For non-nonprofit users, Anthropic’s Claude Max plan at $100/month gives you 5x Pro usage, which is enough for most full-time developers. The trick: route Claude Code through your Claude subscription, then turn off Cursor’s overage option so only your Claude subscription usage flows through. That prevents the surprise charges that catch people off guard when Cursor’s metered models burn through credits in the background.
The Point
None of these platforms are required to build with AI. But the more you build, the more you’ll see them in your agent’s suggestions, in tutorial code, and in the AI-generated architecture diagrams that show up in your projects. Knowing roughly what each one does — and which neighbors it overlaps with — is the difference between accepting your agent’s suggestions blindly and making a real decision.
If you’re new to building this way, the move isn’t to learn all of these tools. It’s to understand the categories: backups, compute, hosting, database, storage, monitoring. Then when an AI suggests a specific tool during your planning conversation, you’ll know which question to ask: “Is this the right category for what I need, and is there one I already use that does the job?” Sometimes you’ll even know enough to say, “Actually, let’s use X instead — here’s why.” That moment, when you start steering the AI instead of being steered by it, is when AI-powered development really starts to pay off.
Envato Elements search: “person confident using AI laptop modern technology” — or Z2W AI Suite prompt: “Wide image of a confident person at a laptop, hands gesturing as if directing a softly glowing AI assistant figure emerging from the screen, modern flat illustration, Z2W green (#82C92B) and gold (#FFD700) accents on white background, conveying mastery and partnership rather than confusion.”
Want help building or untangling your own stack? Book a consultation at zero2webmaster.com/booking.
Dr. Kerry Kriger, Founder — Zero2Webmaster