Lending-finance website with AI orchestration

Process study · proposed method

Resolve the boundaries before connecting the systems.

Explore how synthetic prototypes, shared field contracts and independent review could reduce integration rework and expose visibility errors earlier.

Establish the baseline

Approved product descriptions
Reviewed jurisdiction-specific disclosures
Enquiry and data-handling requirements

Evidence to retain

Reviewed fields and role boundaries
Synthetic workflow and failure checks
Accepted handoff and unresolved decisions

Why work this way?

The process, and what it could improve.

Proposed method · no measured savings claim

Cost

Find the data boundary before connecting systems.

Prototype approved fields and role-specific views with synthetic records. Resolve disputed visibility and review rules before integration work.

What to compare: Implementation, model and specialist-review cost per accepted workflow, including security corrections.
The tradeoff: A synthetic prototype is cheaper to change, but it does not replace production security or legal review.

Compare the same scope and acceptance criteria. Include setup, coordination, failed attempts and human review alongside the successful model call.

Design the work

One problem. A coordinated agent team.

Proposed task plan

A model reasons. A harness runs its agent session and tool loop. Orchard coordinates the work, context and handoffs across them.

Orchard orchestratorScopes work · delegates · synthesizes evidence

CONTROL

Implementation lead

Example session

Model familyClaudeHarnessClaude Code

Create product pages and an accessible enquiry journey from approved lender content.

Bounded task assigned

Counterproposal agent

Example session

Model familyGPTHarnessCodex CLI

Challenge the proposed intake: minimize collected data and keep sensitive applications in an approved service.

Opposing assumption assigned

Independent reviewer

Example session

Model familyGeminiHarnessGemini CLI

Inspect disclosures, form behavior, data handling and secure handoff against lender-approved requirements.

Evidence criteria assigned

Shared project contract

Rules · skills · hooks · MCP configuration

Adapted to each harness

01 / 04 · Dispatch

Split the work. Preserve the question.

One objective, distinct responsibilities

Build a lender-reviewed website while challenging every transition from public information to sensitive intake.

No model invents rates, eligibility criteria, credit policy or approval decisions.

Implementation leadpage system + enquiry schema

Counterproposal agentdata-flow counterproposal + threat questions

Independent reviewerreview checklist + synthetic flow tests

Illustrative execution planAssignment Evidence Review

Model families and pairings are illustrative. Assignments here are manual examples, not measured model comparisons or automatic cost optimization.

Trellis / shared knowledge

Approved product knowledge

Lender-reviewed descriptions, disclosures and jurisdiction scope.

Data and review decisions

Field purposes, service boundaries, threat findings and accepted revisions.

BEAM / coordination

Bounded tasks, ownership notices, findings and handoffs between sessions. A queued message and a processed receipt are distinct states.

Recall → work → retain decisions

Tools / execution

Browser + accessibility toolsForm / application sandboxRepository + security checks

Access follows the configured harness, credentials and approved scope.

Inspect the example work orderJSON

An illustrative work-order format for this scenario. Changing an example session updates this document; no agents run from this page.

{
  "example": "lending-finance-website",
  "objective": "Build a lender-reviewed website while challenging every transition from public information to sensitive intake.",
  "boundary": "No model invents rates, eligibility criteria, credit policy or approval decisions.",
  "sessions": [
    {
      "role": "proposer",
      "model_family": "Claude",
      "harness": "Claude Code",
      "task": "Create product pages and an accessible enquiry journey from approved lender content.",
      "expected_artifact": "page system + enquiry schema"
    },
    {
      "role": "challenger",
      "model_family": "GPT",
      "harness": "Codex CLI",
      "task": "Challenge the proposed intake: minimize collected data and keep sensitive applications in an approved service.",
      "expected_artifact": "data-flow counterproposal + threat questions"
    },
    {
      "role": "reviewer",
      "model_family": "Gemini",
      "harness": "Gemini CLI",
      "task": "Inspect disclosures, form behavior, data handling and secure handoff against lender-approved requirements.",
      "expected_artifact": "review checklist + synthetic flow tests"
    }
  ],
  "project_contract": [
    "Rules",
    "Skills",
    "Hooks",
    "Tool configuration"
  ],
  "shared_context": [
    "Approved product knowledge",
    "Data and review decisions"
  ],
  "evidence_required": [
    {
      "label": "Copy stays within approved claims",
      "evidence": "Trace product statements and disclosures back to their reviewed source."
    },
    {
      "label": "Data collection is justified",
      "evidence": "Map every field to purpose, destination, consent and retention requirements."
    },
    {
      "label": "The boundary works",
      "evidence": "Test synthetic enquiries, error handling and the secure application handoff."
    }
  ],
  "release_owner_gate": "Lender product, legal and security owners review their respective requirements before publication."
}

The process principle

Move uncertainty into a reviewable prototype.

Scope of this exampleNo rates, eligibility promises, credit decisions, or approvals are generated here. Real applications need approved systems and legal, security, and lending review.

A fictional site concept, not a lender, financial offer, underwriting service, or customer result. Application security reference

Lending-finance website

Different tools.
Shared project context.

Where the work changes.
How to judge the difference.

Outlook listener

Modernize an Android app

Modernize a COBOL system

Legacy OS to the cloud

Compliance auditing

Travel agency website