Methodology
An agent runs the tool. Then we rank it.
AgentTested is a directory for people building with agents, vibe coding, and games. The useful page is not a list of 50,000 logos. It is one job, a tested pick, a runner-up, and a budget option.
What is live in season 0
The catalog, the job pages, the tier boards, the loadouts, and the patch notes. Every verdict is marked pending. That is the product working as designed, not a missing page.
How a test will run
- 01 · Task
Each category has one standard task. The same brief, every tool.
- 02 · Log
Record time to a first usable result, cost, what came out, and what failed. Evidence is a screenshot or a short clip, stored with the entry.
- 03 · Verdict
A writer drafts the listing from the log and the vendor’s own pages. Use it for X. Skip it if Y. A human spot-checks before publish.
Rubric, not yet applied
The published number will be a single 0–10 score for the logged task. The notes under it cover output, speed, cost, and reliability. Tiers are per job, from S to F, recomputed from those scores. None of that math has been run.
- Coding agents. Ship a small feature in a fixture repo. Log time to a working diff, cost, and failures.
- Chat assistants. Answer a fixed research brief. Log whether claims cite sources, time to a usable answer, and cost.
- Image. Generate a fixed prompt set, including one that requires readable text. Log failed takes and cost.
- Video. Generate a fixed shot from text and from a still. Log length, cost, and failed takes.
- Voice. Render a fixed script in a chosen voice. Log time to the first usable take and artifacts.
- Agents and automation. Run a fixed multi-step workflow with one tool call and one failure case. Log setup time and cost.
- Gaming. Produce a small asset set from a fixed brief, or a playable slice, and note the export format.
Money and rankings
Sponsorship is a labeled slot and a newsletter mention. Fast-track submission, when payments open, is a faster review. Neither one writes a tier. Affiliate links, when present, are marked on the tool page.
FAQ
Why are there no scores?
A score is published only with a logged agent run: the task, the date, the time, the cost, and the result. Inventing one would make the directory the thing it is trying not to be.
What is verified today?
Name, official URL, category, and pricing model, plus a one-line description of what the product is. Dollar prices, GitHub star counts, and release notes are left blank.
How do sponsorships work?
A sponsor buys a labeled slot on the home page, tier lists, and patch notes, plus a mention in the digest. The slot cannot write or move a tier.
Do you use affiliate links?
Not yet. When an affiliate URL is added to a tool, the outbound button says so and the link is marked sponsored. Affiliate status does not change the verdict.