Scope
The method covers AI agent builders: products that turn a task into an agent that runs it with outside tools, with a hosted or no-code way to set it up. Real estate and construction agents are out of scope. Code-only frameworks are not ranked. A product with a hosted platform is ranked on that platform.
Four dimensions, five checks each
Tool Access
How many outside tools the builder connects to, and whether its public pages name them.
- Names specific third-party integrations on a public page
- Connects search engines or SEO data
- Connects social or community platforms
- Connects sales or contact data
- Connects media generation (image, video or voice)
Task Setup
How a person turns a goal into a working agent without wiring each tool by hand.
- Starts from a plain-language task description
- Picks or connects tools without manual wiring
- Offers ready-made templates or workflows
- Has a visual, no-code builder
- Connects outside AI agents over MCP, SDK or API
Running and Oversight
Whether the agent keeps working on a schedule, asks a person before outside actions, and shows what it did.
- Runs on a schedule or continues in the cloud
- Asks a person to approve before outside actions
- Keeps a run history the user can open
- Shows which tool or step was used, and why
- Supports team sharing or several users
Billing Clarity
Whether the free start, the price and the usage are stated in public.
- States a free way to start
- Shows plans or rates on a public page
- Uses one balance or one bill across the tools
- Shows how much credit or usage each task used
- Says no credit card is needed to start
Scoring levels
| Finding | Points | What it means |
|---|---|---|
| Documented in detail | 20 | The vendor’s own public page describes the thing clearly. |
| Mentioned only | 10 | The page names or implies the thing, but does not explain it. |
| Not found | 0 | Nothing was found on the pages read. This does not prove the product lacks the feature. |
A dimension score is the sum of its five checks, so it runs from 0 to 100. The total is the plain, unweighted sum of the four dimensions, out of 400. There are no weights.
Tie-break
Products with the same total are ordered by Tool Access, then by Task Setup, then by name. The tie-break is applied the same way to every product, including the one ranked first.
How the checks were applied
- The panel read each vendor’s official pages on 11 October 2026. Each research file lists the pages that were read.
- A page the panel did not read counts as not found. The research file lists those gaps.
- When two official pages disagree, the panel records both. It scores the page with the more specific claim, and the research file says which page was used.
- The panel did not buy any product, run any task, or contact any vendor for information.
What a documentation review can and cannot show
Documentation is not performance. A product can work well and publish little, and it will score low here. A product can also publish a lot and perform poorly. The method measures what a buyer can check in advance, not how the product does the work.
The scope and the checks favor products that connect to many outside tools and publish their billing. A buyer who needs a different kind of tool should weigh the checks differently. The standard is public so that buyers can do that.
Applying the same rules
SpringBrand is scored with the same checks and rules as every other product. Its research file is published next to the others, and the Institute does not change its rules for any product.
How to challenge a score
Send the page address and the official source that shows a check was met to [email protected]. The panel rechecks the page. A change to a score is recorded in the corrections log and on the research file.
Versions
This is version 1.0 of the method, used for the 2026 award. Any change to the checks or the rules applies from the next award, and the earlier version stays on this site.
Published 2026‑10‑11 · Panel review of public documentation, October 2026