The tool map: find AI tools that fit your role
Claude Code, Claude Design, Cursor, Lovable, Figma Make and a handful of others. We've tested quite a few, and here's how we think about which ones are worth your time.
You don’t need every tool
New AI tools launch every week. It’s easy to drown in the options and never get going. What we’ve learned after trying a bunch is that most people get by with two or three tools, not ten.
Exactly which ones depends on what you do and how you want to work. Here’s how we think about the map. If you haven’t read the first module of the AI guide yet, it’s a good place to start.
Three categories of AI tools
1. Cloud services (ChatGPT, Claude, Gemini)
This is what most people think of when they hear “AI”. You open a web page, type a question, and get an answer.
ChatGPT (OpenAI)
- Good at: everyday questions, text generation, broad tasks, and with the GPT-5 family, reasoning too
- Weaker at: consistency when you’re working deep and long in the same project
- Best for: someone who wants a quick assistant for broad tasks
Claude (Anthropic)
- Good at: long texts, reasoning, coding, following instructions carefully
- Weaker at: image generation (it has no support for it)
- Best for: professional work where quality and consistency matter
Gemini (Google)
- Good at: multimodal (text, image, video), integration with Google’s ecosystem
- Weaker at: coding, consistent style
- Best for: someone who already lives in Google’s world
How we use them ourselves: Claude when we want considered answers, ChatGPT as a complement for quicker questions.
2. Agent tools (Claude Code, Cursor, Lovable and others)
This is AI that doesn’t just answer but also does things in your projects — for example reads files, writes code, or creates documents for you.
- What: an AI agent with full access to your project, available as a desktop app, web, IDE extension and in the terminal
- Strengths: agents, skills, plugins, parallel work, sandboxing, deep project understanding, and a 1M context on Opus for really large projects
- Our experience: the tool we’ve come to use most. The one that has come closest to a work partner for us. We’ve written about how Claude Code fits the whole team, from designer to developer.
- What: Anthropic’s newer product for generating designs, prototypes and slides together with Claude
- Our experience: we use it in short, targeted bursts, usually when we need a visual starting point or have hit a design wall. It’s expensive to run for long, so we take the result over to Claude Code and continue there.
- What: AI-powered code editor (based on VS Code)
- Strengths: smooth integration in the editor, good for developers who want AI “alongside” them
- Our experience: solid tool. For how we work, Claude Code fits better, but Cursor is a strong alternative, especially for someone who wants to work more in a traditional editor.
- What: web platform that builds whole apps from descriptions
- Strengths: fast for going from idea to working prototype
- Our experience: useful for prototyping. A bit less flexible as the project grows. We made the move from Lovable to Claude Code ourselves as the site grew.
- What: AI features integrated into Figma
- Strengths: generate design directly in Figma, good for visual work
- Our experience: useful as a complement. It doesn’t replace an agent-based way of working for us, but it might for others depending on how they work.
- What: AI platform for building web apps visually
- Our experience: interesting concept. We don’t quite see it as ready for our daily work yet, but it’s developing fast.
- What: AI agent for web research and automation
- Our experience: works well for research. Less good for creative work, at least for what we’ve tried to do.
Codex (OpenAI)
- What: OpenAI’s CLI agent (similar to Claude Code)
- Our experience: we’ve tested it. In our tests Claude Code has worked better for our kind of tasks, but it’s worth trying yourself.
- What: AI coding assistant focused on large codebases
- Our experience: an interesting approach for large codebases. For our work, Claude Code is enough, but it may well be different for someone working in larger systems.
3. Local models (run on your own machine)
AI that runs entirely locally, with no cloud service. We go deeper in module 8. Short summary:
- Why: privacy (data never leaves your computer), cost (free after the hardware), offline work, no dependence on external providers
- When: experimentation, sensitive data, or if you simply want to understand how AI works under the hood
- Tools: LM Studio, Hugging Face, Unsloth, MLX (Mac)
- Current models: GLM 5.2 benchmarks close to Claude’s top models, and Qwen 3.6 sits at roughly Sonnet level and works great fully locally, and Gemma 4 is also worth a try
What does it cost?
Pricing affects the choice, so let’s be concrete.
| Tool | Cost/month | What you get |
|---|---|---|
| ChatGPT Plus | ~200 SEK | GPT-5 family, limited use |
| Claude Pro | ~200 SEK | Claude Sonnet + some Opus access, limited |
| Claude Max | ~1000 SEK | Generous but not infinite Claude Code use |
| Cursor Pro | ~200 SEK | AI-assisted code editor |
| Lovable | 0–500 SEK | Prototyping with limits |
| Local models | 0 SEK | Requires decent hardware |
Claude Max is what we’ve landed on for our daily work. A thousand kronor a month can sound like a lot, but it’s heavily subsidised compared to per-token API pricing. For us it has paid for itself quickly in saved time. Exactly how that looks for you depends, of course, on how much you use it.
Suggestions per role
Designer / UX:
- Claude Code (Max plan), your main tool
- Figma MCP integration, connect Claude Code with Figma
- Complement: Lovable for quick prototyping of new ideas
Project manager:
- Claude Code (Max plan), meetings, communication, documentation
- ChatGPT Plus, quick questions, brainstorming
- Complement: Claude skills with mail integration
Manager / decision maker:
- Claude Pro or ChatGPT Plus, to understand the possibilities
- Focus on giving the team the right tools (see module 9)
The curious one:
- Start with the free versions of Claude and ChatGPT
- Try LM Studio with a local model, it costs nothing
- Upgrade once you’ve found your use case
Our setup
We run:
- Claude Code on the Max plan as our primary work tool
- Figma MCP to connect design with Claude Code
- Claude Design for ideation and when we’ve hit a design wall, then back to Claude Code for implementation
- Local models via LM Studio for experimentation and sensitive tasks
- Our own framework with UX/UI agents, brainstorming skills and design plugins
It took a while to land here, and the way we work keeps changing as the tools do.
Next step
In the next module we’ll go through how to build your first AI workflow, step by step. You don’t need the same setup as us to get going, one tool and a rough plan is enough. If you’d rather have help along the way, read about how we help companies get started or get in touch.
Yes, we wrote this together with Claude. It would be a bit odd otherwise given the subject. But we know what we want to say, and we’ve read, poked and discarded until the phrasing, opinions and choices are ours.