Skip to content
Partners

Take governed AI delivery to the mainframe core.

Integrators now show banks and insurers how to run AI as a governed engineering capability. The hardest place to apply that is the mainframe core: policy administration, billing and accounting systems that are thirty or more years old. Cobra is the specification library, AI harness and execution environment built for those systems. It works beside the conversion and transformation tooling you already use.

Why the core is the hard case

A governed model needs four things the core does not have.

A governed model needs

A specification library

There is no specification. The rules are in the COBOL.
A governed model needs

Agents with managed context

Public models have never seen the client’s copybooks, records or field layouts. They guess.
A governed model needs

A harness that validates AI output

Validation needs the real program to run against real data. A mocked test passes while the business breaks.
A governed model needs

Knowledge that outlives individuals

The people who understand the system are retiring now.
What Cobra provides

The missing pieces, for core systems.

Specification library
Business rules written as decision tables. Each row is one rule, linked to its source line, with an automated test that proves it.
Managed context
The Dictionary: 90+ read-only tools that give any agent authoritative facts about programs, records, fields, call chains and data lineage.
Skill registry
A library of version-controlled agent skills that carry hard-won mainframe lessons from one program to the next.
AI harness
Gates in the delivery pipeline that measure the result, not the artefact. A test must check a value the real program computed. A test that cannot fail is rejected.
Human on the loop
Graded autonomy. At the lowest level agents take work to a human gate. At the highest, a rule-based controller merges work only when every check passes. Changes to test data always stop for a person.
Evidence
Every finding links to the line of code it came from. Results are labelled as measured or inferred. Data access is labelled by sensitivity and logged.
Where Cobra fits

Three steps in a joint engagement.

Step 1

Assessment

Cobra catalogues the core estate and sizes it from the code: what exists, what depends on what, and where the risk is. Days, not months of workshops.
Step 2

Pilot on one domain

Choose one business process, such as premium billing or policy accounting. Cobra extracts the rules, proves them against the real programs, and runs the harness in the client’s pipeline. Success criteria are agreed first.
Step 3

Scale

You lead the operating model, the roles and the programme. Cobra supplies the core-system specification library, harness and runtime. Programs move in small, reversible steps.
Division of roles between the partner and Filix
YouFilix and Cobra
Client relationship, operating model, role transformationCore-system specification library and AI harness
Your delivery framework and governanceMainframe execution environment and the Dictionary
Programme delivery at scaleMainframe and insurance domain depth
Regulator engagementEvidence that goes back to each line of code
Maturity, stated plainly

What is proven, and what is not yet.

A supplier that tells you where the edge is can be checked. One that does not, cannot.

Maturity of the Cobra platform
Built and workingYesEverything described on this site runs today.
Proven against real client production code and dataYesThree client estates run under test every week.
Running a client’s live operationNot yetThat is the next step, and we do not claim otherwise.

Who stands behind it →

See it worked through, step by step

Each compares the default approach with the Cobra approach.

Talk to the engineer

Partner conversations come direct to Michael.

No sales queue. Tell us which client situation you have in mind and we will show you what Cobra would contribute. partners@filix-it.com