← All writing

What people here are actually building with

The AI tools that come up again and again in real builds. Not a ranked list, just what working people reach for and what they use each one for.

By Branislav · · 2 min read

Every week there is a new "top 50 AI tools" post, usually written by someone who has used none of them. This is a different thing: the tools that keep appearing in real builds people describe, and the specific jobs they get used for.

For writing and reasoning

Claude and ChatGPT do the bulk of the work in most projects, and the choice between them is usually decided by one specific task rather than a general preference. Long documents and code review lean one way; image understanding and ecosystem plugins lean another. Most people building seriously end up with accounts for both and route by task.

Gemini shows up most often for anything involving very long inputs or video.

For building the thing

Cursor and Claude Code dominate when the person writing the post is a developer. The pattern people describe is the same: the AI writes the first draft of a file, the human edits it, and the value is in never starting from a blank page.

Lovable, v0 and Bolt show up when the person is not a developer, or wants a working interface in an afternoon. They come up especially in posts where someone validated an idea before writing real code.

Supabase and Vercel are the default backend and hosting in a striking number of builds, mostly because both have a free tier that survives a first launch.

For agents and automation

n8n, Make and Zapier are how most non-developers get an agent into production. The honest tradeoff people report: fast to build, harder to debug when it silently stops working at 3am.

LangChain and LangGraph come up in more complex agent posts, and so do the complaints about abstraction overhead. A recurring lesson on this site: people often start with a framework and end up with direct API calls.

Vapi is the one that keeps appearing for voice agents specifically.

For media

ElevenLabs for voice, Runway and Veo for video, Midjourney and Flux for images, ComfyUI when someone needs real control over the pipeline.

The consistent lesson in creator posts is that the model is rarely the bottleneck. Consistency between shots is. The technique that comes up repeatedly is reusing one character reference across every prompt.

What this list does not tell you

Which of these is right for you. That depends on what you are building, what you already know, and how much time you have.

Which is the actual argument for a place like this: a list of tools is nearly useless, but a hundred specific accounts of people using them on real problems, including where they hit a wall, is worth reading.

Keep reading

Join the conversation.

Indie hackers, solo founders and AI enthusiasts, all in one feed. Follow them, ask them anything, share what you find. Free, two minutes.

Make your profile