Start with the problem. The right setup is the one that solves it - not the most powerful one.
The short version
Pick the setup that fits the problem. A simple problem does not need a complex setup. A complex problem will break a simple one. Three things determine the right level: what you are trying to do, how much you want to spend, and how much experience you have.
Start with the problem
Before picking a tool, ask: what does this problem actually need?
Writing, research, quick answers
APL 1 is enough. ChatGPT, Claude.ai, or Gemini Chat. Free or around €20/month. No setup needed.
Building a workflow in n8n or automating a SaaS tool
APL 2 is the right fit. An AI assistant inside your tool - with access matched to that tool only. Going higher adds cost and complexity without benefit.
Building software, editing code across many files
APL 2 or 3 depending on how much you can review. Claude Code, Cursor Agent, or Aider give the agent direct file access. Speed increases - and so does risk.
Running multiple agents that work together on complex tasks
APL 4 - but only if you understand what each agent is doing. Multi-agent systems are powerful and expensive. They require experience, monitoring, and clear boundaries.
Budget
APL 1Free, or around €20/month for better models. No infrastructure cost.
APL 2€20-40/month for the tool, plus API costs if you use your own keys.
APL 3API costs grow with usage. Heavy use can reach €50-200/month depending on project size.
APL 4Multi-agent systems multiply costs. Makes sense when the output justifies the spend.
Use the lowest level that solves the problem well. Do not pay for power you do not need.
Experience
Just startingAPL 1. Learn what the AI can and cannot do. Build trust in the output before giving it more access.
Some experienceAPL 2. An assistant in your editor or tool. It can see your work and suggest changes. You still approve everything.
Comfortable with code or toolsAPL 3. The agent can write, edit, and run things directly. You need to review what it produces and understand the changes.
Experienced builderAPL 4. You understand the risks, know how to monitor pipelines, and can stop something that goes wrong.
Moving up is not just about learning new tools. It also means understanding what the agent can now reach on your system - and making sure your control keeps pace with that access.
When access and control do not match
Too little access
Your tools are more capable than what you let them do
Example: running Claude Code only for conversation
The AI Setup Snapshot flags this as STAGNANT
Too much access
The agent can reach systems you cannot fully govern
Example: APL 2 tool with write access to your entire production system
The AI Setup Snapshot flags this as CRITICAL
Not sure where you stand? The AI Setup Snapshot finds your APL and AAL in 3 minutes and tells you exactly what to fix.
