Five Levels of AI Adoption for Teams
Adapted from Dan Shapiro, “The Five Levels of AI-Assisted Software Development,” 2026. Level names and descriptions rephrased for internal use in businesses.
| Level | Name | In one line |
|---|---|---|
| L0 | Manual | People do the work; AI is at most a reference. |
| L1 | Assisted | AI handles discrete sub-tasks; people drive and own the work. |
| L2 | Collaborative | People and AI work together; most AI-fluent teams are here today. |
| L3 | Supervised | AI does most of the work; people review and approve the output. |
| L4 | Orchestrated | People set the standards and guardrails; AI runs within them; people validate by sampling. |
| L5 | Autonomous | AI runs the activity end-to-end; people set intent, priorities, and quality bars. |
Example Outcome Narratives
L0 — Manual: risk-scoring dashboard, but every decision — hold, release, or A travel ops team manually reviews flagged bookings for fraud. They have access to a risk scoring dashboard, but every decision — hold, release, escalate — is made by a person. AI generated the score; a human reads it like a reference doc and acts independently.
L1 — Assisted: An engineer is writing a data pipeline. They use an AI coding assistant to autocomplete boilerplate, generate a unit test scaffold, and suggest variable names. The engineer still drives every architectural decision, reviews every line, and owns the output entirely.
L2 — Collaborative: A product manager drafts a feature brief. They write the problem statement, then ask an AI to help structure the success metrics, push back on scope, and propose edge cases. The PM rewrites, reacts, redirects. Neither the human nor the AI could have produced the same doc alone. Most teams doing AI-assisted knowledge work live here right now.
L3 — Supervised: A content team uses AI to generate the first draft of product description copy at scale — hundreds of listings per day. A human editor reviews a sample, approves batches, flags patterns that need prompt tuning. The AI is doing the heavy lifting; the human is a quality gate, not the author.
L4 — Orchestrated: An engineering org configures an AI agent to handle on-call triage: it monitors alerts, correlates signals, opens incidents, and pages humans only when a threshold is crossed. People defined the runbooks, set the escalation thresholds, and audit the outcomes weekly. The AI operates continuously within those guardrails without per-task human approval.
L5 — Autonomous: A pricing system runs continuous optimization end-to-end — ingesting competitor signals, modeling demand curves, updating prices, and logging rationale — without human review of individual decisions. People set the business objectives, define the constraints, and review aggregate outcomes. The system runs the activity; humans own the intent.