What Does Agent-Ready
Actually Look Like?

Your IBM i system has no web presence agents can navigate by default. There is no llms.txt, no structured sitemap agents can read, no declared policy on what agents are permitted to query or invoke. From an agent's perspective, your system doesn't exist — until you build the discovery layer.

• Create llms.txt at your domain root declaring agent permissions and MCP endpoint
• Update robots.txt to explicitly allow or restrict specific agent crawlers
• Publish MCP server manifest — what tools are exposed, what schemas they expect
• Add Schema.org Organization markup with API contact and capabilities

For regulated lenders, discovery is also a fair lending consideration. If your products are discoverable by agents operating on behalf of some consumer segments but not others — due to missing structured data or inconsistent API coverage — you may have created a disparate access issue before a single loan decision was made.

• Declare product categories in llms.txt — LTO, installment, BNPL — with eligibility scope
• Ensure agent-accessible product catalog covers all products available to human applicants
• Publish fair lending policy note in machine-readable form
• Register with financial data aggregators (Plaid, Intuit) as a reachable institution

The business logic that governs your products lives in RPG programs and DB2 tables. No agent can read RPG. But Project Bob and Claude Code can surface that logic into human-readable documentation that becomes the foundation for machine-readable product specifications. This is the knowledge extraction step.

• Document DB2 schema — table names, field definitions, relationship maps — in structured markdown
• Use Claude Code / Project Bob to extract RPG program logic into readable specs
• Publish OpenAPI spec describing your product and eligibility endpoints
• Create machine-readable product terms — rates, fees, eligibility thresholds — in JSON-LD

Most consumer finance product disclosures are PDF documents designed for human readers. An agent evaluating your LTO or installment product against three competitors needs structured, parseable terms it can compare programmatically. Machine-readable product terms are the prerequisite for agent-driven recommendations.

• Implement Schema.org LoanOrCredit and FinancialProduct structured data on product pages
• Publish APR range, minimum/maximum terms, and payment structure in JSON-LD
• Structure eligibility criteria — minimum income, ID requirements, state availability — as queryable fields
• Deprecate PDF-only disclosures as primary product information surface

This is the layer IBM i shops are closest to — but haven't crossed yet. Mapepire opens DB2 to modern API consumers. MCP exposes RPG business logic as callable tools. The data is there. The interface isn't built yet. Two tools, one bridge.

• Mapepire confirmed on port 8076 — DB2 accessible via REST/WebSocket
• Expose key DB2 tables via Mapepire endpoints — applicant, account, eligibility
• Build MCP server exposing RPG decisioning programs as callable tools
• Document tool signatures — inputs, outputs, error states — in MCP manifest
• Version your API surface — /api/v1/ — so agents can pin to stable contracts

Consumer finance access has two tracks: read access (an agent checking product availability, rates, eligibility for a consumer) and write access (an agent initiating an application). Both require real-time APIs. Neither can be served from a batch file or a static product page.

• Publish a read-only eligibility check endpoint — pre-qualification without credit pull
• Implement rate and availability endpoints with real-time data — not cached daily updates
• Define read vs write scopes in your API authentication model
• Build agent-accessible application initiation endpoint — structured intake, not HTML form
• Document response time SLA — agents expect sub-second responses for product lookups

Agents executing transactions is the Stage 03 moment from Silver Lake Installment 2 — the event fires, the system responds, no human assembled the steps. For IBM i shops, this means wiring the MCP tool layer to event-driven workflows already running in your production environment. The logic exists. The invocation path doesn't yet.

• Define which RPG workflows can be invoked agentically — without human approval gate
• Implement non-human authentication — API keys, OAuth2 client credentials, not login forms
• Build webhook support — agents send instructions, your system responds asynchronously
• Implement idempotency keys — so duplicate agentic requests don't create duplicate transactions
• Define crawl/walk/run gate — which actions require human approval before going autonomous · MCP spec →

Action in regulated lending has a higher bar than action in unregulated contexts. An agent submitting a credit application on behalf of a consumer is a regulated event. The action layer in consumer finance must be built with adverse action infrastructure already wired in — not added after deployment.

• Adverse action reason codes must fire automatically at the point of agentic decisioning
• Consumer identity verification must occur before agentic action — agents cannot self-certify consumer identity
• ECOA notice delivery must be automated and timestamped at transaction time
• Define agentic action scope in your model risk governance framework before deployment
• Implement transaction reversal capability — agents make errors; your system must handle them

IBM i's object-level authority model and integrated audit journaling give IBM i shops a structural advantage here. The platform was built for auditability. The gap is connecting the agent's invocation trail to the IBM i audit log — so the record of what the agent called is reconstructable alongside the record of what the system did.

• Enable IBM i audit journaling for all MCP-accessible objects and programs
• Log every agentic tool invocation — timestamp, agent ID, inputs, outputs
• Implement structured log format parseable by downstream compliance tooling
• Define agent identity in your authority model — what profile does an agent run under?
• Publish your agent policy document — what agents can do, what requires escalation

Governance in consumer finance is not optional. SR 11-7 model risk management requirements, ECOA adverse action documentation, FCRA accuracy obligations, and UDAAP exposure mapping all apply to agentic decisioning. The governance layer is the layer that makes every other layer defensible.

• SR 11-7 model inventory must include agentic decisioning models
• Fair lending monitoring must run on agentic decisions — not just human-reviewed ones
• Adverse action documentation must be immutable — written at decision time, not reconstructed after
• UDAAP exposure mapping must cover agentic interaction patterns — tone, sequence, omission
• Examination readiness: can you produce a complete audit trail for any agentic consumer interaction in under 24 hours?