Skip to content
DWS Specification

Why DWS?

The governance gap

Every functional institution in human history has solved the same problem: how do you deploy human intelligence at organisational scale while maintaining coherence, accountability, and the ability to improve over time?

The answer has always been the same. You wrap non-deterministic intelligence in deterministic governance structures. Roles. Authority boundaries. Institutional memory. Performance contracts. Escalation paths. Audit trails.

We spent thousands of years building that scaffolding. It is the foundation of every functioning institution.

AI arrived, and the industry threw all of it away.

The current state of AI deployment: capable models with no role definition, no authority boundaries, no institutional memory, no accountability structure, and no audit trail. Every session starts cold. Every output is unverified. Every delegation is unsupervised.

This is not a technology problem. The models are good. It is a management problem. The governance layer around the models does not exist.

What happens without a standard

The babysitter tax. Organisations deploy AI agents, then assign senior people to watch them. A compliance officer reviewing every output. An engineer checking every code change. The cost of supervision often exceeds the cost of just doing the work manually. The promise was delegation. The reality is a new layer of overhead.

The amnesia penalty. Every session starts from zero. The conventions your team follows, the mistakes made last quarter, the client preferences learned over months of engagement — none of it carries forward. You are re-onboarding the same worker every morning. A human employee with no memory of yesterday would be fired. We accept this from AI because we have no alternative.

The accountability vacuum. When an AI agent produces bad output, who is responsible? What authority did it have? What process did it follow? What knowledge did it draw on? Without structured governance, these questions have no answers. The output is a black box. The audit trail is “we prompted it and it said this.”

The lock-in trap. Build a worker in CrewAI and you cannot move it to Claude Managed Agents. Build it in LangGraph and you cannot export it to Bedrock. Every framework is a proprietary island. Your worker definitions should be as portable as your source code.

What DWS solves

The Digital Worker Standard applies the same governance infrastructure to AI workers that humans have always operated within.

Identity and authority. Every worker has a name, a domain, a role, and an authority level. Authority is graduated (escalate-only, restricted, supervised, autonomous) and enforced by the runtime at three checkpoints. The model cannot override its own authority.

Institutional knowledge. Workers accumulate learning across sessions. Patterns, conventions, failure modes, and lessons persist and compound. At run 1,000, the worker knows things it did not know at run one. This is what “tenure” means for a digital worker.

Independent verification. A separate worker evaluates output against the original intent. The verifier sees the output and the objective. It does not see how the worker arrived at its answer. This context isolation prevents rubber-stamping.

Structured workflows. Work flows through defined phases with verification gates, approval gates, and cost tracking. The routing is deterministic. The intelligence within each phase is not. Governance wraps the intelligence without constraining it.

Portability. A single job spec definition compiles to Claude Managed Agents, CrewAI, AWS Bedrock, or any compliant runtime. One definition, any platform.

Compliance by construction. A DWS job spec is a compliance artifact. Authority levels map to EU AI Act human oversight requirements. Event streams map to record-keeping requirements. Verification gates map to accuracy measurement requirements. The governance is not bolted on after the fact. It is the structure.

The test

Would you give this worker a job title?

A Contract Reviewer, a Compliance Analyst, a Due Diligence Associate — these are roles. They require institutional knowledge, defined authority, and an employment record. They need DWS.

Medical coding, report template population, taxonomy classification — these are stateless tasks. A well-built harness is sufficient.

If the worker needs tenure, accountability, and governance, it needs the Digital Worker Standard.

Learn more