Look. If you’re an indie dev or running a small business, you’ve probably been bombarded with AI agent hype for the last 18 months. Every SaaS landing page says their agent will “10x your productivity.” Most of it’s noise. But some of it? Genuinely useful.
I’ve spent the last three months testing AI agent tools that an actual indie developer or small business owner could use. Not enterprise stuff. Not $500/month platforms with 47 integrations you’ll never touch. Tools that work when your team is you, maybe one other person, and a Slack channel that’s mostly tumbleweeds.
Here’s what’s actually useful. Not what’s trending on Twitter.
The Coding Side: What Ships Code Faster
If you write code, AI coding agents are where the rubber meets the road. And the numbers back this up. Stack Overflow’s 2025 survey found 84% of developers use or plan to use AI tools. JetBrains put it at 90% in January 2026. This isn’t a future thing anymore. It’s just how people work now.
Cursor is the obvious starting point. It’s an AI-first code editor built on VS Code’s bones — so the switch doesn’t feel like learning a new instrument. You describe what you want in plain English and it modifies your codebase. What makes it actually good, and I mean genuinely useful, is that it understands your project structure and makes changes that are consistent with how the rest of your code is written.
$2 billion in recurring revenue. That’s what Cursor hit. For a tool that didn’t exist three years ago.
But here’s the thing. Cursor’s pricing has crept up and if you’re an indie dev on a tight budget, $20/month might feel steep for something you’re still experimenting with. So what’s the alternative?
Claude Code. Runs in your terminal. No fancy UI. You type what you want, it edits files. Pair it with an MCP server — the Model Context Protocol thing Anthropic pushed that lets agents connect to external tools — and you’ve basically got a mini dev team in your terminal. A freelance developer I know in Barcelona cut his project turnaround time from two weeks to four days on a client dashboard build. Four days. He said the hardest part wasn’t the tool — it was articulating exactly what he wanted in enough detail.
And honestly? That’s the real skill now. Not coding. Describing what you want clearly enough that an AI can build it.
DeepSeek V4 Pro deserves a mention too. It’s roughly 4-5x cheaper than Claude Sonnet for API calls and it’s open-weight — meaning you can actually run it yourself if you have the hardware. For an indie dev who’s burning through tokens on side projects every night, that price difference matters. A lot. I switched my own agent backend from Claude to DeepSeek last month and my monthly API bill dropped from €87 to about €22. Same quality on 80% of tasks. The 20% where Claude is better are real — long context reasoning, complex multi-step logic — but for most coding work? DeepSeek handles it fine.
The Workflow Side: Automating the Boring Stuff
Here’s where things get interesting for small business owners who don’t code.
Make (formerly Integromat) lets you build visual workflows that connect your apps. Think: “When a new lead fills out my Typeform, create a contact in HubSpot, draft a welcome email in Gmail, and notify me on Slack.” That kind of thing. It’s not a conversational AI agent in the “talk to a chatbot” sense — it’s automation with AI steps bolted on. But for a small business owner drowning in manual busywork, it’s probably the single most useful tool on this list.
The free plan gives you 1,000 operations per month. Enough to automate a few core workflows without spending a cent. Paid plans start at $10.59/month. That’s less than a Netflix subscription.
Lindy is the other option I’d recommend if you want something more agent-like. It’s a no-code AI agent builder — pick a template, connect your email and CRM, add instructions in plain English, and you’ve got an agent handling lead qualification, follow-ups, or support triage. I set up a basic lead qualifier in about 20 minutes during a test run last week. Connected Gmail and Notion. Told it: “When someone emails about pricing, check if they mentioned their company size, if not ask, then log everything to this Notion database.” It just worked.
The catch? Lindy starts at $49.99/month. Which isn’t nothing. If you’re making money and the time savings are real, it pays for itself fast. If you’re pre-revenue? Maybe start with Make.
For the Indie Hacker: The Budget Stack
So what does a practical, affordable AI agent stack look like for someone building on their own?
Here’s what I’d actually recommend, not what looks good in a blog post:
For coding: Cursor free tier or Claude Code in terminal. If you’re paying for API access directly, use DeepSeek for most tasks and Claude only when you need the extra reasoning quality. You’ll save 70-80% on your API bill overnight. I did exactly this and the numbers don’t lie.
For automation: Make’s free plan handles the basics. When you outgrow it — and you will, once you taste what automation feels like — the $10.59/month plan covers a surprising amount.
For image generation and design: Ideogram or Midjourney’s basic plan. $10/month gets you a tool that replaces a designer for social media graphics, app store screenshots, and blog post headers. Every indie dev I know who ships regularly uses one of these now. Nobody’s commissioning custom illustrations anymore unless they’re funded.
For customer support: Honestly? A good FAQ page plus Crisp or Tawk.to’s free live chat still beats most AI support agents for small businesses. The AI support agents that work well — Intercom’s Fin, Zendesk’s bot — are priced for companies with actual revenue. If you’re solo, write a thorough FAQ and handle the 3 support tickets you get per week yourself.
And that last point is important. Nobody talks about it enough.
What You Don’t Need (Yet)
There’s a whole category of AI agent tools that are genuinely impressive but complete overkill for an indie dev or small business. CrewAI, AutoGen, LangGraph — these are frameworks for building multi-agent systems. They’re meant for teams orchestrating specialized agents across complex workflows. Unless you’re building a product around AI agents, you don’t need them.
I see indie devs spending weeks learning LangChain before they’ve shipped a single feature. That’s backwards. Ship something broken first. Add AI after you have users. The framework can wait.
Same goes for vector databases. Pinecone, Weaviate, Chroma — they’re cool technology. But if you’ve got 47 users and a side project that makes €200/month, a JSON file will serve your RAG needs just fine for a long time. I ran my first AI-powered search feature off a 14MB JSON file for six months before I needed anything fancier.
The Real Bottleneck
After three months of testing these tools — and this is the thing I keep coming back to — the bottleneck isn’t the AI. It’s almost never the AI.
It’s whether you can describe what you want clearly. Whether your data is organized enough for automation to work. Whether your processes are actually defined or just stuff you do from muscle memory. The best AI agent in the world can’t automate a workflow that exists entirely in your head.
A product manager at a fintech company told me something that stuck: three weeks of data preparation improved their agent’s performance more than all the model fine-tuning they tried. Three weeks of cleaning up how documents arrived in their system. That’s what moved the needle. Not better prompts. Not a newer model. Clean data.
So yeah. The tools are ready. They’re genuinely useful. But they’re not magic — they’re force multipliers. If you multiply zero, you get zero. If you multiply a messy but functional solo operation, you might actually get somewhere.
I don’t know what the right stack is for your specific situation. It depends on whether you’re writing code or sending emails, whether you’ve got revenue or you’re still building, whether your biggest problem is time or money or just figuring out what to build next. But the tools I’ve listed here — Cursor, Claude Code, DeepSeek, Make, Lindy — those are the ones that survived three months of actual use.
Not the ones with the best landing pages.