Editorial Policy
How we research, write, and rank everything on this site. We publish this policy openly so you can hold us to it.
Independence
Agentbrisk does not accept payment for placement, paid reviews, or sponsored rankings. Editorial decisions, what we cover, how we rank it, what we recommend, are made independently of any commercial relationship. No vendor, advertiser, or affiliate partner has any influence over the content of a review or where a product appears in a ranking.
We will refuse and turn away any offer of money, free hardware, paid trips, or other compensation in exchange for a more favorable review. If a vendor asks us to soften criticism, change a ranking, or remove a comparison, we say no. If a vendor pulls advertising in retaliation, we cover the story.
What we cover
We focus on three categories: AI agents (autonomous and semi-autonomous tools), agent frameworks (open-source toolkits for building agents), and Model Context Protocol (MCP) servers. Within each category, we cover products that meet at least one of these criteria:
- Real users and active development.
- Notable funding, acquisition, or shutdown that's relevant to readers evaluating the space.
- A genuinely novel approach worth understanding even if adoption is limited.
We don't cover vaporware, abandoned projects without historical importance, or products that exist only as a marketing landing page.
How we test products
Our testing approach varies by category. For coding agents, we run them against representative real-world tasks: multi-file refactors, bug fixes, feature implementation, test generation. For browser and computer-use agents, we run them through common workflows: form submission, scheduling, research synthesis. For frameworks, we build a small but non-trivial agent and observe both developer experience and runtime behavior.
Some products require enterprise contracts, invite-only access, or hardware we don't have. When we can't test something firsthand, we say so explicitly and lean more heavily on documented capability, customer interviews, and public benchmarks. We never claim hands-on experience we don't have.
Sourcing standards
Every claim about features, pricing, or product behavior is checked against at least one of these sources:
- The vendor's official documentation.
- Product release notes or changelog.
- First-hand testing by the editorial team.
- On-the-record statements from the vendor.
When we cite specific numbers (pricing, search volume, GitHub stars, market position), we link to the source. When a number changes between our verification and your read, the page's "last updated" date will reflect the most recent review of that fact.
Fact-checking and review
Every review goes through at least two reviewers before publication: the writer and a second editor who checks claims against sources. For technical reviews, the second reviewer is someone with hands-on experience in the relevant area. We don't publish reviews where the second editor flags significant unresolved questions.
For ratings and rankings, the second editor's job is to argue the opposite position. If they can find a plausible case the rating is wrong, the writer addresses it before publication. The goal is to produce judgments that hold up to honest pushback.
Updates and corrections
Every page is reviewed at least quarterly. Pricing data, version numbers, and live signals (GitHub stars, release dates) refresh daily through automated sources, with editorial review of any changes flagged as significant.
If we make a factual error, we correct it as soon as we can verify the correct information. We add a dated note at the top of the page describing what was changed and why. We don't quietly update pages without acknowledging changes.
Spotted an error yourself? Email [email protected] with a link to the page and the source we should check. We aim to verify and correct within five business days.
Conflict of interest
Editorial team members disclose any personal financial interest in companies we cover (equity, advisory roles, employment, family ties). When a conflict exists, that team member doesn't participate in writing or reviewing coverage of the affected company.
We do not own equity in vendors we review, accept stock options, or take board or advisory positions with companies we cover.
Affiliate and advertising relationships
Agentbrisk earns money from display advertising and from a subset of outbound links that are affiliate links. These commercial relationships do not influence what we write or how we rank products. The full breakdown of how we make money is in our advertising and affiliate disclosure.
We refuse advertising from products we wouldn't recommend or that violate our standards. We also refuse advertising arrangements that come with strings attached on coverage.
AI-assisted writing
We use AI tools (including Claude and similar models) to help draft research notes, outlines, and first drafts. Every published article is reviewed and edited by a human editor before going live. The editorial team takes full responsibility for the final published content, including its accuracy, tone, and judgments.
We do not publish AI-generated content as-is. We don't use AI to fabricate quotes, invent benchmarks, or simulate user reviews. We write in a deliberately specific, opinionated voice that reflects real testing and judgment.
Reader feedback
We take reader feedback seriously, especially from people with hands-on experience using the products we cover. If you think we got a review wrong, we want to know. Email [email protected] and we'll engage with the substance of your argument, not just acknowledge receipt.
Anonymous bylines
Most reviews on Agentbrisk are published under the editorial team byline rather than individual writer names. We made this choice for two reasons: we want the work to be judged on its merits, and the small team often collaborates closely on each review. For specific pieces where individual authorship matters (long-form essays, opinion pieces), we credit the writer by name.
Standards we hold ourselves to
Beyond the specifics above, we aim for these editorial standards:
- Specific over general. Concrete examples beat vague claims.
- Honest over balanced-for-balance. We name winners and losers when the evidence supports it.
- Useful over comprehensive. We trim sections that don't help the reader make a decision.
- Clear over clever. Plain language beats jargon.
- Updated over evergreen. We'd rather rewrite a stale page than leave it sitting.
Contact
Questions about our editorial standards: [email protected].