Enterprises with legacy mainframe estates face a dilemma: maintain aging systems or attempt risky ‘big bang’ migrations. Google Cloud proposes an AI-accelerated modernization strategy that treats mainframe modernization as more than code-to-code conversion. The real challenge is handling tightly fused application logic, legacy non-relational data formats (VSAM, flat files, IMS), transaction monitors (CICS, IMS TM), intricate sequential workflows, proprietary protocols, and deep operational lock-in. Modernization must also update data models, resolve obscured dependencies, and validate code with production traffic before go-live.

The approach combines Gemini models for code understanding with mainframe-specific products across four pillars:

1. Assessment – The Mainframe Assessment Tool (MAT) reverse-engineers legacy codebases at scale, providing dependency visualization, automated business rule extraction (BRE) into plain-language requirements and decision trees, automated documentation, and domain/business function discovery. Outputs feed into agentic workflows via MCP for high-accuracy code transformation.

2. Modernization – Two paths are offered: Rewrite/Reimagine for high-ROI business logic innovation, using MAT and Mainframe Modernization Agents to extract business rules and forward-engineer with Antigravity as the agentic harness, optionally with human-in-the-loop governance. Deterministic modernization (like-to-like) preserves exact behavior for workloads requiring structural modernization without changing external interfaces, using AI for direct code-to-code conversion with strict contract fidelity. A financial services example shows mixing both: like-for-like for stable batch jobs, deterministic AI for core general ledger, and rewrite for a customer-facing loan origination platform.

3. De-risking – Dual Run processes live production traffic simultaneously on the mainframe and Google Cloud, comparing outputs (protocols, messages, data changes) until equivalence is achieved, eliminating go-live risk. It is already used by global enterprises for regulatory approvals.

4. Data migration – The Mainframe Connector copies data off the mainframe into BigQuery, Spanner, Cloud SQL, Cloud Storage, etc., handling codebase and data-type conversions, and integrating into existing ETL processes to offload processing and unlock siloed data.

The article concludes with a call to action: start with an automated codebase assessment using MAT, then participate in an agentic modernization workshop and pilot with Google Cloud experts to build a business case for one application.