Methodology
How Hypervision rates AI: three tiers for the big picture, and a six-step autonomy scale (L0–L5) inspired by the SAE levels of self-driving cars.
Tiers
Every entry sits in one tier: Tool (you drive), Agent (you delegate) or Frontier (labs pushing toward AGI and superintelligence).
Classic generative AI. You give precise instructions, it returns an answer, an image or a sound. Powerful, but it waits for you.
Systems that plan, use tools, browse, write and run code across many steps — then report back. The shift happening right now.
The labs and research pushing toward AGI and superintelligence. Not products you use today — the forces shaping what comes next.
The autonomy scale
| L | Autonomy | Human role | Example |
|---|---|---|---|
| L0 | No AI Classic software. Rules written by people, no learning, no generation. | Does everything | A spreadsheet formula |
| L1 | Assist Responds to a single instruction with one output: an answer, image, translation or sound. No memory of goals, no actions. | Prompts and decides every step | Image generator, translator |
| L2 | Copilot Works inside your workflow, keeps context and uses tools on request (search, code, files) — but you drive and apply every change. | Stays in the loop, applies results | Chat assistants, code completion |
| L3 | Agent Plans and executes multi-step tasks with tools — browsing, editing files, calling apps — and asks for approval at key points. | Sets the goal, approves checkpoints | Browser agents, app builders, workflow agents |
| L4 | Autonomous agent Works unattended for long stretches within a domain — takes a ticket, does the work, tests it and hands back a result for review. | Reviews the finished result | Asynchronous coding agents |
| L5 | General autonomy Any task, any domain, at or beyond expert level, without supervision. This is AGI — and beyond it, superintelligence. Not reached. | Optional | — |
What we ask
- Does it only respond, or does it plan?
- Can it use tools — browser, files, code, apps?
- How long can it work without a human?
- Who checks the result, and when?
Slovenian support
“Good” means the tool understands and writes natural Slovenian. “Basic” means it works, but with weaker quality or limited features. Ratings are editorial and reviewed regularly.
Editorial assessment, last reviewed: 2026-09.