Reverse Engineering a Knowledge Worker

Many enterprise workflows require a knowledge worker to receive an unstructured request, consult multiple sources of documentation, reconstruct the relevant operational context, create an execution plan, interact with one or more software systems, observe the resulting state, collect evidence, and communicate the outcome to another person.

Structured knowledge work is often performed not as a single act of reasoning but as a stateful sequence of activities that span communication, documentation, task interpretation, planning, interaction with software, evidence collection, and reporting. Project REKNOW (Reverse Engineering a Knowledge Worker) is a reference architecture for decomposing structured knowledge work into modular, cooperating software components. The architecture combines work intake, knowledge retrieval, requirement analysis, planning, state-oriented orchestration, computer-using execution, evidence generation, and human communication. It documents a systems-engineering approach for composing existing technologies into an auditable, extensible, and stateful workflow architecture.

Project REKNOW began with a narrower engineering problem: repetitive graphical user-interface testing consumed engineering capacity and required a human tester to repeatedly perform this sequence. The resulting systems question was broader: Can the observable workflow of a structured knowledge worker be decomposed into reusable, cooperating software components? Project REKNOW explores this question through a modular reference architecture. The architecture does not treat the knowledge worker as a single general-purpose agent. Instead, it separates knowledge access, requirement analysis, planning, orchestration, execution, observation, evidence generation, and communication into independently replaceable components.

Modeling Structured Knowledge Work

Project REKNOW models structured knowledge work by decomposing it into eight functional stages:

1. Receive and register a work request.
2. Acquire relevant operational context.
3. Interpret the requirements.
4. Plan the work.
5. Execute the plan through software systems.
6. Observe the resulting state during execution.
7. Preserve evidence and execution status.
8. Communicate the outcome or request human intervention.

This decomposition was derived from the original GUI-based quality-assurance workflow. It includes cognitive functions such as context acquisition, requirement interpretation, and planning, together with operational functions such as interaction, observation, evidence capture, and communication. Project REKNOW reverse-engineers these functions at the workflow level. It does not attempt to reproduce the biological or psychological mechanisms through which a human performs them. Each function is instead represented as a bounded system responsibility that can be assigned to an AI model, deterministic software component, human participant, or a combination of these mechanisms.

Generalized Reference Architecture

Implementation: Autonomous GUI-Based QA

The reference implementation applied Project REKNOW to autonomous graphical user-interface quality assurance for a custom enterprise web application. The prototype processed a human test request and supporting requirement documents, reconstructed the existing business flows needed to exercise the requested change, produced an end-to-end business-level test plan, and executed part of that plan through visually grounded mouse and keyboard interactions.

The prototype instantiated the generalized architecture through three cooperating subsystems:

1. a cloud-based business-intelligence subsystem for work intake, requirement analysis, planning, and human communication;
2. a server-side state and artifact bridge responsible for document extraction, domain-knowledge retrieval, state observation, and cross-environment coordination;
3. and a desktop-based execution-intelligence subsystem responsible for observing and operating the target enterprise application.

The full technical report is available at: Reverse Engineering a Knowledge Worker

Implementation Artifacts: Source code, prompt templates, workflow configurations, and supporting evidence for the reference implementation are available at: Project-REKNOW on GitHub.