Frequently asked questions

AgenticReview reads the git history of your repositories and turns 34 signals into eight meters, four composite indices - Capacity, Composable, Craftsmanship and Cooperation - and one total from 0 to 10. Steering shows up in what the humans do around the agents: guardrails and instruction files set before an agent starts, specs and plans written before it builds, and review, correction and reverts after it delivers.

The model scores only what no competent architect would argue against, such as duplicated rules, missing tests or leaked secrets. Strategic choices where both sides have defenders are mapped, never scored.

The review runs on GitHub's own runners with read-only access. Your code is analysed in memory and discarded when the run ends. Only derived metrics leave GitHub, and they are stored in the EU. No source code, file contents or commit messages are stored, and none are used to write the analysis.

An agentic review is the next step of code review, not another tool next to it. Traditional code review checks whether each change is correct, and traditional tools measure output and rule compliance - throughput, bugs, vulnerabilities, test coverage.

An agentic review keeps that ground and adds what decides in the AI era: whether the architecture stays composable, whether the work can be handed over, and whether the team steers the agents that now write much of the code - or just accepts what they produce. You do not add a review; you upgrade the one you have.

AgenticReview does not certify compliance with the EU AI Act or any other regulation. It gives an independent, versioned record of how human oversight of AI-generated code works in practice, over time and against the market, which boards, auditors and acquirers can use as part of their own assessment.

Each of the four pillars is shown on a 0-10 scale, and the total is their average, weighted by how many of each pillar's meters could be measured. Levels are set against a reference cohort of real engineering organisations, not against a fixed rulebook, so a score says where you stand relative to organisations building software today.

There is no universal pass mark. Every report shows your score next to the global, regional and size-band reference, so you can see whether you are above or below comparable organisations - and which pillar moves the total most. The most useful comparisons are often internal: over time, to see where you improve and where the potential lies, and between your own products, teams or divisions, to surface the success stories and role models others can learn from. The external reference adds a first indication of where you stand in the market.

Against a reference cohort of public engineering organisations, analysed with the same method as every customer review, balanced across seven regions and compared within size bands. You choose the comparison in the report: global, your region, and organisations of your size. Accounts can choose to share anonymised, aggregated metrics with the reference, so the benchmark can extend to organisations that build in private repositories.

No. Results are reported per product, per division and for the company as a whole. Contributors are counted to normalise volume, so a small team is not marked down for being small, but no per-person score is calculated, stored or shown.

Most teams run a review once every week or month (or 2/3 weeks depending on sprint design), or after a larger change in how they work with agents. Every report includes a written analysis of what drives the score and which changes would move it most, so improvement starts with the one or two pillars that hold the total down.

Yes. Your first report already includes about six months of history, reconstructed from your git history, and every new review adds to the trend. You can follow the total and each pillar over time, per product or for the whole company. Every report states its model version, and a change of version is never mixed into the same trend line.

Yes. You group repositories into "products" (or projects), and products into "divisions" (or companies). The dashboard ranks them side by side on the total and on each pillar, and each one has its own report and trend over time.

Usually a few minutes from start to report. When GitHub's runners are busy it can take up to around half an hour. You do not need to wait on the page - the report appears in your dashboard when it is ready.

Yes. The report gives an independent, versioned view of how a software asset is built, steered and handed over - including risks that only show late, such as duplicated logic or knowledge that sits with one person. Because no code leaves GitHub, a target company can run it without exposing its code to the buyer.

The first review every month is free, up to 10 repositories. Premium Bronze is $20 per month for 5 reviews of up to 20 repositories each; Silver and Gold cover more reviews and larger products. Six-month and yearly plans are discounted. Prices exclude VAT. Every paid plan comes with a 14-day money-back guarantee, and you can cancel at any time - the plan then runs to the end of the period you have paid for, including discounted six-month and yearly plans.

The benchmark is built from public GitHub organisations, analysed continuously with the same method as customer reviews. Calibration anchors are recalibrated as the benchmark grows, and every report states its model and threshold version so results stay comparable over time.

AgenticReview is built by DataRBL AB, a Swedish data company founded by Rufus Lidman, a licentiate in statistics with 30 years in digital and data strategy, four published books, some 300 keynotes, around 50,000 followers and three AI forums he moderates. It is developed in cooperation with academia and a broad international network of established industry experts. Derived metrics are stored in the EU. The full list of sub-processors is in our Privacy Policy, and a Data Processing Agreement is available for every customer.