People own every decision. AI agents do bounded, reviewable work inside written rules.
Finray uses AI agents in software development. This page describes the mechanism: the path a change to product code takes, the rules that govern the use of AI tools, and the certification the company holds.
Agents write and check code. People decide what is built, and a person approves the exact version before it is merged.
Eight steps between a task and a merge.
-
Task and plan recorded
The task is recorded in Pync, Finray's project system, together with its acceptance criteria and the plan for meeting them.
-
A written brief
The brief sets out the scope, the files involved, what must not change, and how completion will be proven.
-
A failing test first, then the implementation
A test that fails is written before the code that makes it pass. An AI agent then implements the change in an isolated workspace.
-
Automated checks on every change to product code
Every change to product code is run through secret scanning, dependency and vulnerability scanning, static analysis and the test suite.
-
Independent review
A different AI model family from the one that wrote the change reviews it. Findings go back to the same workspace and are re-checked in the next review.
-
Live test with synthetic data
The change is tested live on a dedicated test environment that uses synthetic data.
-
Human approval of the exact version
A person approves the exact version that was reviewed and tested. Any later change to it restarts review.
-
Merge with a complete audit trail
The change is merged with a record of who did what, and at which version. "Who" is either a person or a model.
What Finray's internal AI guideline requires.
-
Approved tools only
Finray keeps an allow-list of approved AI tools, plugins and extensions. Anything not on the list is not approved.
-
One reviewed source for shared skills
Shared AI skills come from a single, internally reviewed source. Local, unreviewed variants are not used.
-
No browser-driving agents against production
Agents that drive a browser run only against local, development or test systems. They are never pointed at production.
-
Re-approval for upgrades that change behaviour
A tool upgrade that adds hooks, network calls or telemetry needs approval again before it is used.
-
Synthetic data only
Agents work with synthetic data. Customer data, personal data and secrets are never given to an AI tool.
-
Every action is attributable
Each action stays attributable to the real person or the model that performed it.
Certified information security management.
Finray Technologies Ltd holds ISO/IEC 27001:2022 certification (NQA, UKAS-accredited, certificate No. 215646, valid 21 October 2025 – 21 October 2028). The 2026 surveillance audit was passed.
This is a company-level certification of the information security management system. It is not a product certification.
View the certificate, or read About Finray.
Questions about our development controls.
We will walk a compliance, risk or procurement team through any step above.