Terminal operation
Recognize the screen, map fields, follow the known navigation path, and check the expected response. Screen changes and session errors require explicit handling.
Example tool: lookup_request(reference)
Existing systems. A clearer front door.
AgenticModernization Factory
Make the systems you already depend on easier to use. Follow the work from a legacy screen to a controlled operation, a browser interface, and a release decision.
Explore the workflow ↓Discuss your process
Legacy → controlled operationsOperations → usable appsValidate → review → release
Research & technical references
Inspect the claim ledger ↗Source logos identify cited publications. They do not indicate endorsement, partnership, certification, or product validation. τ-bench is a paper hosted on arXiv.
What does it mean in practice?
An agentic modernization factory is a repeatable engineering process for understanding existing workflows, defining controlled operations, creating modern interfaces, and validating behavior before release. Agents assist the process; system owners retain deployment and business authority.
Anthropic distinguishes predefined workflows from agents that choose their own processes and tools. That is a useful design distinction: a stable rule can stay a rule, while an agent handles interpretation within an explicit set of operations.[BL-CLM-15]
GAO’s 2025 review selected 11 critical federal legacy systems for modernization. Plans for eight lacked complete documentation of key elements. The lesson for a modernization project is to define the work, milestones, and intended future of the existing system before promising a transformation.[BL-CLM-16]
Follow one piece of work
From green screen to browser.
Choose a stage to see the host view, the operation exposed to the agent, and the artifact an engineer or operator reviews. Try the mock lookup with SR-1042 or a reference that does not exist.
Interactive walkthrough / fictional data
TN3270 / EXAMPLE CASE INTAKE
PUBLIC SERVICES — REQUEST LOOKUP
REFERENCE: SR-1042
APPLICANT: ALEX MORGAN
STATUS: AWAITING REVIEW
PF3 = BACK · ENTER = LOOKUP
Simulated educational flow. No live document processing, sign-in, connector, or mainframe session.
Step 01 / Read screen
Read the example case intake screen and identify the request reference, applicant and status.
A read operation observes the current screen. No host record changes.
What comes out
A captured screen with source fields
How the host connection works
The agent calls a tool. The adapter talks to the host.
Microsoft documents capturing 3270 screens, defining navigation paths, and selecting method inputs and outputs. Its tooling stores metadata in an HIDX file for the connector. This is a concrete example of making a screen-driven transaction usable through a defined operation.[BL-CLM-17]
Application API
Where an approved API exists, expose a narrowly scoped business operation with typed inputs and results. Confirm authentication, business rules, and supported endpoints during discovery.
Example tool: get_request_status(reference)
Job or dataset interface
Where an authorized interface exists, define exactly which records or job results can be read. Keep submit and write operations separate from read operations.
Example tool: read_approved_job_result(jobId)
The API and job lanes above are illustrative integration patterns; the cited Microsoft documentation specifically supports the terminal-mapping lane. Tool names are examples. This page does not assert a tested Papyrus mainframe connector.
One transaction. A defined API.
Wrap a terminal workflow with OpenAPI.
Select a terminal field to see how it maps into an API contract. Then try a lookup to see the structured response. The adapter performs the screen navigation; OpenAPI describes the operation clients can call.
01 / Example legacy screen
CASE INTAKE / REQUEST LOOKUP
ENTER = LOOKUP · PF3 = BACK
Adapter responsibility
Recognize the screen, enter the reference, navigate the transaction, and read the result. Stop if the host returns an unexpected screen.
02 / OpenAPI 3.0.3 contract
GET /requests/{reference}
REFERENCE → reference
The lookup input becomes a required path parameter.
name: reference
in: path
required: true
schema:
type: string
pattern: ^SR-[0-9]{4}$A schema defines names, types, inputs, and outcomes. It does not create a connector or grant host access.
03 / Try the contract
A lookup, without the green screen.
SR-1042 returns a record. SR-9999 returns 404. An invalid format returns 400.
Response preview
// Run a lookup to inspect the response.
Interactive illustration with fictional records. All responses are generated locally in this browser; no mainframe or live API is contacted.
Questions worth asking.
What is an agentic modernization factory?
An agentic modernization factory is a repeatable engineering process for understanding existing workflows, defining controlled operations, creating modern interfaces, and validating behavior before release. Agents assist the process; system owners retain deployment and business authority.
Does modernization require rewriting COBOL?
No. Some workflows can use adapters or existing APIs while keeping their underlying business behavior. Code transformation or replacement may be necessary in other cases. The decision depends on maintainability, dependencies, and acceptance requirements.
How can agents interact with a mainframe?
Through a deliberately exposed connector or tool, such as a mapped terminal operation, an application API, or an authorized job interface. The adapter manages protocol details and access; the agent receives a limited operation with explicit inputs and results.
Is the Microsoft connector included in Papyrus?
The Microsoft source documents its own 3270 integration tooling. It is an example of an available approach, not confirmation that Papyrus includes or has tested that connector. Verify host access, licensing, and integration support during discovery.
Evidence, with boundaries
Read the sources. Inspect the claims.
Source review: October 9, 2026. Research findings, government guidance, and vendor documentation are identified separately.
Generative AI Profile · AI 600-1
Published / updated 2024-07-26
NIST AI 600-1 is a cross-sectoral generative AI profile supporting risk management throughout the AI lifecycle.
Voluntary risk-management guidance. It is not a certification or a finding about Papyrus.
Research benchmarkτ-bench · Yao, Shinn, Razavi & Narasimhan
Published / updated 2024-06-17
τ-bench evaluates tool-using agents against goal database states and tests consistency across repeated trials under domain policies.
The 2024 benchmark uses simulated users and domain environments. Its results are not Papyrus results or a measure of current models.
Vendor engineering guidanceBuilding Effective Agents
Published / updated 2024-12-19
Anthropic distinguishes workflows with predefined code paths from agents that dynamically direct their processes and tool use.
Engineering guidance published in 2024. The article notes that the tooling landscape has since changed.
Critical Legacy Systems · GAO-25-107795
Published / updated 2025-07-17
GAO identified 11 critical federal legacy systems in need of modernization; plans for eight did not fully document key elements.
These are GAO-selected federal systems, not all mainframes or all legacy applications. The report does not evaluate Papyrus.
Designing Metadata Artifacts for 3270 Applications
Published / updated 2023-11-08
Microsoft documents capture, navigation and method definitions for mapping 3270 screens and fields into HIDX metadata used by its connector.
This describes Microsoft Host Integration Server and Azure Logic Apps tooling, not a verified Papyrus connector. Requires host access and applicable software.
Bring one real back office workflow.
Start with the documents, systems, exceptions, and approval requirements. Decide what to automate after you understand how the work succeeds.