Tier 1 Augmentation

Tier 1 — Augmentation (Summary & Positioning)

Definition

Tier 1 Augmentation is the foundational tier in the Agentic Tier Framework (v2026.1). At this tier, AI supports a human’s work but never acts on its own. The human initiates the interaction, interprets the output, makes decisions, and executes actions. The AI accelerates specific steps (e.g., drafting, summarizing, extracting, researching) inside a human-driven workflow.

Typical Tier 1 examples:

• In-context drafting assistants inside productivity tools
• Single-prompt research helpers (e.g., ask-and-answer interfaces)
• Structured-output extractors (e.g., turning text into fields) used only as inputs to human decisions

Tier 1 is often the entry point for institutional AI: it allows organizations to gain leverage while keeping responsibility and materiality clearly with the human operator.

Criteria for Tier 1 Classification

A capability is Tier 1 only if all of the following are true:

1. Explicit human invocation
   • The human chooses when to use the AI.
   • No automatic or background invocation.
2. Human review before any action
   • AI output is always read and interpreted by a human before anything is filed, sent, scheduled, transacted, or otherwise acted upon.
   • No downstream system is allowed to act directly on AI output.
3. No action interfaces
   • The AI can generate text, structured data, or recommendations only.
   • It cannot directly send emails, submit forms, trigger workflows, modify records, or change system state.
4. Human-controlled materiality
   • The materiality event is the human’s decision to accept and use the output.
   • The human remains the accountable actor for any consequential use.

If any of these conditions are violated, the capability is not Tier 1 and typically moves to Tier 2 (Co-pilot) or higher, where the AI is embedded in workflows with default or semi-automatic acceptance.

Canonical Controls at Tier 1

To be defensible at scale, Tier 1 deployments must operate under explicit controls:

• Prompt and source hygiene
  • Users are trained on what data may or may not be shared with the AI provider.
  • Constraints reflect contractual terms, jurisdictional rules, and confidentiality regimes.
• Verification responsibility
  • The human must verify any authority, citation, or factual claim before relying on it.
  • This is especially critical in legal, regulatory, and scientific contexts.
• Use case scope
  • Each capability has a named, approved workflow in the Risk Register.
  • Using the same tool outside that scope is itself a control violation.
• Telemetry on invocation patterns
  • Monitoring focuses on how the tool is used: prompts, audiences, volumes, and frequency.
  • At Tier 1, telemetry is about usage patterns, not autonomous behavior.

Common Tier 1 Pitfalls

• Implicit tier creep
  • Over time, humans may stop meaningfully reviewing outputs and begin treating them as default-approved.
  • Functionally, the capability behaves like Tier 2, but governance still treats it as Tier 1.
  • The Risk Register and AI Council Decision Log should be used to detect and correct this.
• Confidentiality breaches
  • The main risk is what users put into prompts (e.g., client secrets, regulated data).
  • Incidents typically arise from misaligned prompt hygiene, not from the AI’s internal behavior.
• Hallucinated authority
  • Fabricated case law, statutes, or references are especially common in Tier 1 research and drafting use cases.
  • The primary control is rigorous human verification before use.

Tier 1 in the Operating Model

Tier 1 is embedded in the institutional operating model via specific governance and lifecycle processes:

• USE-01 — Use Case Intake
  • Classifies the capability as Tier 1 and records the rationale (why it is augmentation only, not co-pilot).
• GOV-02 — Risk Register
  • Captures in-scope workflows, verification regimes, and materiality calibration (what is at stake if the human misuses or over-trusts the output).
• GOV-15 — Quarterly Cadence Retrospective
  • Reviews Tier 1 capabilities for:
    • Evidence of tier creep (declining review rates, default acceptance patterns)
    • Confidentiality incidents or near-misses
• AI Council decisions
  • Any promotion from Tier 1 to Tier 2+ is a formal decision, recorded on the Council’s log.
  • Capabilities do not drift up the tier ladder informally.