Digital Employee, or Personal Assistant?¶

Evidence Level: F (directly proven by source code)
Analysis Baseline: 4f843556

TL;DR¶

The README claims "Digital Employees … just like a new hire," but in the code, an Agent's identity is derived from a creator_id foreign key, RBAC has no agent role, and quotas live on the User table—architecturally, this is a personal assistant, not an organizational colleague.

1. The Official Promise¶

Clawith's README defines its product positioning in the opening lines (README.md:49-50):

Clawith agents are digital employees of your organization. Every agent understands the full org chart, can send messages, delegate tasks, and build real working relationships — just like a new hire joining a team.

2. Industry Baseline: What Is a "Digital Worker"?¶

"Digital Worker" is not a concept invented by Clawith. In industry practice, this concept has well-defined boundaries:

2.1 Identity Independence¶

Non-Human Identity (NHI) is a foundational concept in modern IAM. Gartner's Identity-First Security framework explicitly requires that non-human identities be managed as first-class entities with their own lifecycle—they must not expire when their human creator leaves the organization[^1].

Google's BeyondCorp / Zero Trust architecture's core principle: every access subject—human or non-human—must have its own independent, verifiable identity[^2].

In the RPA and AI platform space, UiPath Robots and ServiceNow Digital Workers each possess:

Independent machine identity (Robot account in AD/AAD, not bound to a human)
Independent license allocation
Audit logs attributed to the Robot itself, not its registrant

In contrast, Clawith's Agent derives its identity through a creator_id foreign key—the Agent's "existence" depends on its creator. If the creator is deleted, the Agent's ownership relationship breaks directly.

2.2 Authorization Model¶

NIST SP 800-162 (ABAC Guide) classifies entities in a system as Subjects and Objects[^3]:

Subject: An active entity that initiates access requests, possessing an independent set of attributes and security identifiers
Object: A passive entity that is accessed

A digital employee, within its scope of responsibility, must be a Subject—it initiates messages, invokes tools, and accesses resources on its own behalf. But Clawith's RBAC enum has no agent role; the Agent is an object subordinate to agent_admin in the permission model.

The OWASP Non-Human Identity Top 10 (2025 draft) lists "lack of independent authorization model for non-human identities" as a core risk under CHER-1 (Credential Handling and Entity Resolution)[^4].

2.3 Accountability¶

ISO/IEC 22989:2022 (AI terminology standard) defines accountability for AI systems as: "the state of an entity being answerable for its actions and consequences"[^5].

If an Agent's work records, quota consumption, and cost attribution are all bound to its creator (a human), then the Agent has no architectural basis for being held accountable. A genuine digital employee should have independent:

Cost center attribution
Performance metrics
Work records (regardless of whether the creator is still employed)

These industry practices collectively define the minimum architectural standard for a digital employee: independent identity, role-based authorization, and independent accountability. These three criteria are not our invention—they are a distillation of industry consensus.

[^1]: Gartner, "Identity-First Security: The New Perimeter," 2023. [^2]: Google Cloud, "BeyondCorp: A New Approach to Enterprise Security," 2014. [^3]: NIST SP 800-162, "Guide to Attribute Based Access Control (ABAC) Definition and Considerations," 2014. [^4]: OWASP, "Non-Human Identity Top 10 (Draft)," 2025. [^5]: ISO/IEC 22989:2022, "Information technology — Artificial intelligence — Artificial intelligence concepts and terminology."

3. Three Criteria¶

If an Agent is truly a "digital employee" rather than merely "my assistant," it must satisfy at least three criteria:

Criterion	Meaning	Industry Basis
Independent Identity	Exists independently of its creator, with its own organizational identity	Non-Human Identity (Gartner), BeyondCorp (Google)
Role-Based Authorization	Permissions derive from position/role, not from who created it	NIST ABAC Subject/Object distinction, OWASP NHI Top 10
Independent Accountability	Has its own quotas, cost center, and work records	ISO/IEC 22989 accountability definition

These three criteria distinguish a "digital colleague" from a "digital assistant." Each has a corresponding industry standard as its basis.

4. Criterion 1: Independent Identity ❌¶

backend/app/models/agent.py:43-44:

creator_id: Mapped[uuid.UUID] = mapped_column(
    UUID(as_uuid=True), ForeignKey("users.id"), nullable=False
)
tenant_id: Mapped[uuid.UUID | None] = mapped_column(
    UUID(as_uuid=True), ForeignKey("tenants.id")
)

An Agent's identity is a database record with a non-nullable creator_id foreign key. The Agent has no:

Independent cryptographic identity (key pair, certificate)
Organizational position (position/title/department)
Delegation chain (who authorized this Agent to act on whose behalf?)

Its identity is "a record under someone's name."

5. Criterion 2: Role-Based Authorization ❌¶

backend/app/models/user.py:75-76:

Enum("platform_admin", "org_admin", "agent_admin", "member",
     name="user_role_enum"),

The four roles are: platform admin, org admin, agent admin, and member. All are human roles.

There is no agent role. In Clawith's permission system, Agents are not first-class citizens—they are objects managed by agent_admin and used by member.

6. Criterion 3: Independent Accountability ❌¶

backend/app/models/user.py:92,96:

quota_message_limit: Mapped[int] = mapped_column(Integer, default=50)
quota_max_agents: Mapped[int] = mapped_column(Integer, default=2)

The quota subject is the User (message limits, max Agents). The Agent layer only has runtime throttling:

max_llm_calls_per_day: Mapped[int] = mapped_column(Integer, default=1000)

An Agent has no independent budget, headcount, or cost center. How many Agents you can create and how many messages each can send depends on who you are—this is "my assistant" governance, not "digital colleague" headcount management.

7. The Capability Upper Bound¶

Agent's actual capability = creator's permissions ∩ tenant RBAC ∩ quota ceiling

An Agent can never exceed the capabilities of its creator. This is the governance formula of a personal assistant.

8. Comparison: Three Types of Entities¶

Entity	Level	Core Characteristic	Clawith's Agent
My Assistant	Individual	Identity derived from creator; permissions bounded by creator	✅ Matches
Digital Employee	Organizational	Independent identity, role-based authorization, independent accountability	❌ Does not match
Autonomous Agent	Independent	Independent goals, autonomous decisions, independent resources	❌ Does not match

Clawith's Agent is architecturally the first type (personal assistant), but the README wraps it in the language of the second type (digital employee).

9. Relationship to AI Coding¶

This is not an AI Coding problem—it is a product positioning decision. But AI Coding's involvement is visible in:

The code faithfully implements the "personal assistant" model (creator_id foreign key, User quotas, human RBAC) and never deviates—the AI did not "correct" the wrong product positioning; it merely faithfully executed it
The gap between README and code suggests marketing language and engineering implementation were produced by different people/processes, with AI as the executor of only one side
This serves as the baseline in the study of AI Coding's engineering consequences: the positioning falsehood is a project-origin problem; AI Coding's contribution is faithfully implementing a wrong positioning, and amplifying the architectural cracks in the process