A shared ladder for marketing AI

The Levels of Marketing Autonomy

Self-driving cars have L0–L5 so everyone knows what "autonomous" actually means. Marketing AI doesn't — so every vendor calls a chat box an "agent." Here's the honest ladder. Most tools are stuck at Level 2. crm.care runs at Level 3 today — and can prove it.

Level 1

Assist

Writes when asked.

Generates a draft. You brief it, edit it, place it, send it, measure it — everything around the words is still you.

Input A prompt, every time.
Who's here Generic AI writers (Jasper, Copy.ai, ChatGPT).
Level 2

Copilot

Suggests inside the platform.

Offers suggestions in the tools you already run. Helpful, but it waits for you at every step and never owns an outcome.

Input You, still driving.
Who's here HubSpot Breeze, Salesforce Agentforce, Adobe AEP agents — stuck here.
Level 3

Conditional autonomy

You are here

Runs the campaign within guardrails.

Generates the campaign, publishes it into your MAP, sends, and reads attribution back — acting on its own inside approval gates and budget caps you control.

Input A brief + the limits you set.
Who's here crm.care — today.
Level 4

Portfolio autonomy

On the roadmap

Runs to a goal.

Proposes the quarter's campaign portfolio from CRM gaps, ships within standing authority, kills what's losing, doubles what's winning, and reports a P&L.

Input A pipeline target + a budget.
Who's here The destination crm.care is built toward.

Most "AI agents" are Level 2 with a Level 4 keynote.

The gap between what's announced and what ships is the whole story of 2026 — Gartner expects 40%+ of agentic-AI projects to be cancelled by 2027. crm.care doesn't demo autonomy; it runs it: campaigns generated in your brand voice, published into Account Engagement or HubSpot, sent, and attributed back from your CRM — with approval gates and per-tenant budget caps as the safety rails that make Level 3 trustworthy.

  • Generates + publishes into your MAP — not just a draft in a chat window
  • Acts on its own inside guardrails you set — approvals, budgets, schedules
  • Closes the loop — reads attribution back so it knows what worked