How We Work
How We Test AI Agents
An honest account of what's behind our rankings: what we do today, what we're building toward, and what we refuse to do.
Our Promise
The short version, up front
Our rankings are never for sale. No company can pay to be recommended, to rank higher, or to have a negative finding removed. We do earn affiliate commissions when readers buy through some of our links — at no extra cost to you — and our affiliate disclosure spells out exactly how that works. If that ever changed, you'd read about it here first.
How Our Research Works Today
Our current process, step by step
- Documentation review. We read each product's official docs, changelogs, and feature pages to understand what it claims to do — before we look at a single review or opinion.
- Pricing verification. We check plans and prices against vendor pricing pages wherever they load for us directly; where vendor pages block automated checks (many are JavaScript-heavy), we cross-reference independent pricing trackers from the same month instead. Every price is date-stamped, and we never present a cross-referenced price as a directly verified one.
- Feature comparison. We compare products side by side on the dimensions that matter for the category — capabilities, limits, integrations, and platform support — so rankings rest on differences, not vibes.
- Public consensus cross-check. We read reviews, developer forums, and community discussions to see where independent users agree and where they disagree with vendor claims, and we note the disagreement when it matters.
- Re-verification cadence. Rankings aren't write-once. We revisit them regularly as products ship features, change pricing, or get acquired — and we update the page when our conclusion changes.
What We Evaluate
Five dimensions, applied to every category
Capability
What the agent can actually do: the quality of its outputs, the complexity of tasks it handles reliably, and how it performs at the edges of its advertised use cases. We weight real-world task completion over feature-list length.
Ease of Use
How quickly a competent person can go from sign-up to a working result. We look at onboarding, documentation quality, sensible defaults, and how gracefully the product handles mistakes — yours and its own.
Pricing Value
What you get per dollar: whether the free tier is genuinely useful, where the paywalls sit, and how costs scale as usage grows. A great product with punishing overage fees loses points here.
Ecosystem & Integrations
How well the agent plays with the tools you already use. We count real, maintained integrations — not logo walls — and look at API quality, community templates, and third-party support for developers.
Trust & Privacy
Who sees your data and under what terms. We read the privacy policy and terms of service, note data retention and training-use policies, and give credit for self-hosting options and transparent security practices.
We deliberately don't reduce these dimensions to numeric scores. A single number hides tradeoffs that matter — an agent that's superb for developers and bewildering for beginners can't be fairly summed in a digit. Our rankings reflect judgment across all five dimensions, explained in prose.
Lab Testing Program
Where we're headed — stated plainly
We are rolling out a hands-on testing protocol: standardized tasks run against each agent in its category, with notes published on each review page as testing is completed. That work is in progress, not finished.
An honest line in the sand: we have not yet completed hands-on testing for every product we cover. Any review page that includes test notes says so explicitly; any page that doesn't include them is research-based, as described above. We will never fabricate test results, benchmark numbers, or quotes — if we haven't tested it, we say so.
Updates & Corrections
Rankings are revisited regularly. When a product ships a major update, changes its pricing, or a competitor leapfrogs it, we update the affected pages and the "last updated" date at the bottom of the page.
All prices on this site were last checked in October 2026 unless a page says otherwise — directly on vendor sites where accessible, otherwise cross-referenced against independent trackers from the same month. AI pricing changes constantly — always confirm the current price on the vendor's site before you buy.
Found a mistake — a wrong price, a shipped feature we missed, an outdated claim? We genuinely want to hear about it. Send details through our contact page and we'll investigate and correct the page, crediting the correction where appropriate.
Frequently Asked Questions
Can companies pay for a better ranking on SI Agent?
No. Rankings are never for sale, and sponsored placements — if we ever offered them — would be clearly labeled as such. Affiliate commissions never influence which products we recommend or the order they appear in, including in roundups like our best AI agents ranking.
Do you run hands-on tests of every AI agent you review?
Not yet. Today's reviews are research-based: documentation, verified pricing, feature comparisons, and public consensus. A hands-on testing protocol is being rolled out, and per-review test notes will be published as testing is completed. We never present research as lab testing.
How often do you update rankings and prices?
Rankings are revisited regularly as the market moves — and this market moves fast. Prices were checked in October 2026 and noted as such on each page. If a price has changed since we checked, the vendor's current pricing page is the authority.