MMarketing Against The Grain
← All frameworks
Innovation

Assistant-Peer-Conductor Model

Match your operating model to AI's increasing level of autonomy.

Difficulty
Moderate
Time to result
~ongoing to results
Steps
6
Confidence
92%

The Assistant-Peer-Conductor Model describes three stages of AI operating maturity. In the assistant stage, AI supports a person with research, drafting, editing, or other supervised tasks. In the peer stage, a specialized agent can own a bounded job at roughly colleague-level quality, allowing one person to operate with several extensions of themselves. In the conductor stage, AI coordinates work, chooses or sequences specialized capabilities, and may outperform a person at the operational task. The model guides delegation: classify the current capability honestly, define what the human still owns, prove quality on bounded work, and expand autonomy only when evidence supports it. As systems mature, the human role moves from direct execution toward goal-setting, evaluation, exception handling, and orchestration.

Origin

Extracted from Marketing Against The Grain as Kieran Flanagan described three eras through which AI-enabled work may progress.

Core principles

  • 01AI capabilities progress through distinct levels of autonomy
  • 02Each maturity level requires a different human role
  • 03Delegation should expand only when output quality is proven
  • 04Vertical agents can become specialized extensions of a person or team
  • 05Humans shift from doing tasks to coordinating systems as autonomy increases

How to run it

  1. 1

    Define the Job

    Specify the outcome, constraints, inputs, and quality threshold for the work being considered. A vague activity cannot be assigned a meaningful autonomy level.

    Pro tip Choose an observable deliverable rather than a broad function such as marketing or strategy.

    Watch out Do not classify a workflow based on a product demo rather than its performance on your actual work.

  2. 2

    Start as an Assistant

    Use AI for supervised subtasks while a person directs the process and remains responsible for the finished result.

    Pro tip Capture which instructions and examples consistently improve the output.

    Watch out Assistance can appear autonomous when a human is quietly correcting most of the work.

  3. 3

    Prove Peer-Level Performance

    Give the system a bounded job and compare its quality, reliability, and speed with a competent colleague performing the same job.

    Pro tip Use repeated trials and explicit acceptance criteria rather than one impressive result.

    Watch out Do not promote a system to peer status when it succeeds only on ideal inputs.

  4. 4

    Build an Agent Team

    When specialized agents reach peer-level performance, assign them distinct roles and create explicit handoffs between their outputs.

    Pro tip Keep each agent's remit narrow enough that failures can be diagnosed.

    Watch out Multiple agents can compound errors when no stage validates upstream work.

  5. 5

    Introduce Conductor Behavior

    Allow AI to select, sequence, and coordinate specialized work after the underlying tasks and handoffs have demonstrated reliability.

    Pro tip Preserve human escalation for ambiguous goals, exceptions, and consequential decisions.

    Watch out Coordination autonomy should not exceed the reliability of the weakest critical component.

  6. 6

    Reclassify Continuously

    Review the workflow as models and tools improve. Move responsibilities between humans and AI according to demonstrated capability rather than fixed assumptions.

    Pro tip Track failure patterns as carefully as productivity gains.

    Watch out Neither permanent skepticism nor premature autonomy is a sound operating policy.

In the wild

Content Production Maturity

A team first uses AI as an assistant for research and headline options. After proving that a specialized editing agent consistently meets its standards, it grants that agent ownership of first-pass edits. Later, an orchestration layer routes transcripts through clipping, captioning, proofreading, and review agents while a human approves the creative concept and final release.

The workflow gains autonomy gradually without surrendering accountability or production quality.

Specialized Campaign Agents

A campaign lead deploys separate agents for audience research, channel adaptation, and performance reporting. Each agent first operates under supervision, then receives bounded ownership after repeated evaluations. A coordinating agent eventually schedules the handoffs and flags exceptions to the campaign lead.

The campaign lead shifts from manual task execution to goals, quality control, and exception handling.

Common mistakes

Skipping Directly to Conductor

Giving an unproven system control over coordination magnifies weak task performance and unreliable handoffs.

Confusing Fluency with Competence

An articulate output may still fail the real quality, factual, or operational requirements of the assigned job.

Using One Autonomy Level Everywhere

Different tasks mature at different rates, so a system may be a peer in coding or editing while remaining only an assistant in creative judgment.

Is it for you?

Best for

It is best for teams progressively introducing AI assistants and agents into established workflows.

Not ideal for

It is not ideal for unstructured deployments without measurable outputs, review standards, or accountable human owners.

From the transcript

there are really three eras of AI that I think we will go through there's the assistant era where the AI is an assistant to…

Kieran Flanagan · 20:30

there's the AI as a peer and we're see starting to see that in vertical with things like Devon where the AI is like a…

Kieran Flanagan · 21:00

and then AI is a conductor which is really where we want we are trying to get to

Kieran Flanagan · 21:00

From the episode

Will AI Automate 95% Of What Marketers Do?