How does Opus turn knowledge of work into reusable workflows?
Two business workflows can contain much of the same work and still be built as if neither existed before.
A supplier-onboarding process checks submitted information. A process for changing supplier details checks some of that information again. Each may be documented separately, with its own description of the task, its own evidence and its own assumptions about what happens next. When teams come to automate them, they have to establish which parts can be shared and which differences matter.
The Opus Work Knowledge Graph gives that knowledge a common structure. It connects reusable tasks with the context and evidence that explain how they fit together. Each workflow becomes a connected part of the graph, allowing the work of understanding one process to contribute to the next.
The same task, more than one workflow
The Work Knowledge Graph contains tens of millions of nodes. At its centre are tasks with a single responsibility: units of work that can be assigned, executed and reused independently. Directed connections record the flow of execution between tasks, including where one task’s output becomes another’s input or trigger.
For a simplified example, supplier onboarding and supplier-detail changes might both require validation of submitted banking details. Where the task and its requirements match, both workflows can reuse it. Where an operating policy requires a different check, the graph preserves that distinction.
This gives Opus a structured starting point for workflow generation. Existing tasks and supported connections can be recovered and reused, with attention directed to the requirements of the new process.
What justifies the connection?
Knowing that two tasks exist does not establish that one should follow the other. Evidence for checking banking details, and evidence for authorising payment, would not be enough to conclude that payment can immediately follow the check. A separate approval may be needed.
Opus therefore requires generated additions to carry evidence for the whole workflow, each task and each execution connection. The handoff has to be supported as well as the work on either side of it.
The graph also uses observed frequency to help recover established process paths. Commonly observed handoffs have shorter edge lengths; uncommon handoffs have longer ones. Frequency helps with discovery. Suitability still depends on the business context and the evidence for using that path.
Knowledge has to fit the context
Processes with similar names can differ by industry, geography, language and level of detail. Those differences affect what can usefully be reused.
Opus classifies its coverage across industry, sub-industry, geography, BPMN level and language. A combination identifies a precise area of knowledge, such as a particular level of retail-banking work in a particular country and language.
This lets Opus identify specific gaps. A region of the graph may have extensive coverage of a broad process while lacking the detail needed for a particular context. Improving that coverage requires targeted evidence and validation.
How Graph Gardening keeps it current
Opus uses Graph Gardening to maintain and extend the Work Knowledge Graph. Autonomous gardeners inspect assigned regions, identify missing workflows or handoffs, and rank possible improvements by user value, evidence availability, duplication risk and cost.
A gardener might propose a workflow for an underserved context, enrich a sparse description, or investigate whether two tasks represent the same work. It retrieves evidence before asking a model to produce a candidate.
That candidate then goes through factual grounding, structural checks, citation checks and entity resolution. High-confidence duplicates can be merged into a shared task. Similar tasks remain separate where the evidence is insufficient to justify treating them as identical.
Routine additions can be accepted automatically after validation. New connections between workflow outputs and inputs require human review because they change how processes compose. Gardeners work within a shared budget, and approved changes are published in versioned snapshots with a record of their evidence and approvals.
The distinction matters in practice. A model can help investigate and describe the work. Acceptance into the graph is a separate decision, with a history that can be inspected and changes that can be reversed.
Search the graph and add your own work
Users can search in natural language. Opus combines keyword and semantic matching with recognised classifications and contextual signals to find relevant graph content. Search workers examine different regions in parallel and return results as they find them.
Existing Opus workflows can also be imported. Their task structure and connections are preserved, while enrichment can fill missing descriptions. The import retains its source, workspace and version, so an organisation’s own workflow remains distinguishable from a generated best-practice template.
Enterprises can start with Opus’ master graph and extend it in a private deployment. Their workflows, evidence and subsequent graph growth remain private. Contributions return to the master graph only with explicit client authorisation.
Why this matters for building automation
The immediate value is a better starting point. Teams can draw on existing tasks, supported process paths and relevant context when building another workflow. They can inspect the evidence behind a proposed handoff and see where their requirements call for a different approach.
Over time, the graph also provides a place to retain the work involved in resolving those questions. Shared tasks can remain identifiable across workflows. Differences can be recorded explicitly. Corrections can be traced to the knowledge they affect.
These properties support workflow design and review; execution still depends on the implementation, systems and conditions of the particular workflow. The Work Knowledge Graph makes the underlying process knowledge easier to find, examine and reuse, so each new automation can build on work already done.