Vol. 01 · Field Guide Revised 2026
AI Brain Coach
C · T · R · L   —   A framework for delegating to AI agents

The CTRL
Framework.

For leaders who are done asking AI questions and ready to hand it work. A plain-language method for delegating to agents without losing control.

15 sections · ~15 min read
Published by AI Brain Coach · aibrain.coach
Author Ray Tan
AI Brain CoachThe CTRL FrameworkContents
Contents

What's inside.

Four controls. One framework. A short reference for handing real work to an AI agent without losing the wheel.

—Foundations
01 Introduction Delegation isn't prompting 03
02 The CTRL Framework Four controls, at a glance 04
—The Four Controls
03C  Context Set the world the agent works inside05
04T  Task Define what "done" means06
05R  Rules Keep the human in authority07
06L  Loop Tell the agent how to work, check, and stop08
—In Practice
07Agent vs Chatbot The relationship has changed09
08The Handoff Card Four questions for any task10
09Weak vs Strong Delegation Side-by-side examples11
10Human Review Points Where judgement belongs12
—Closing
11The Impact How each control affects outcomes13
12Next Steps Where you go from here14
13Credits Contributors & contact15
02aibrain.coach
AI Brain CoachIntroductionDelegation isn't prompting
01   Introduction

Delegation
isn't prompting.

Most leaders learned AI by prompting: ask better questions, get better answers. That worked when AI was mostly a conversation partner.

Agentic AI changes the relationship. An agent doesn't just answer. It plans, searches, compares, drafts, revises, calls tools, triggers workflows, and moves through a task across many steps.

You are no longer asking. You are delegating — and delegation needs a different discipline.

CTRL is that discipline. A structured, repeatable system for handing work to AI agents without losing the wheel. Four controls, written in plain language. No tooling expertise required.

The thesis

AI agents don't need better vibes. They need better handoffs.

What changes

From asking for answers to delegating real work.

What it costs

About thirty seconds more clarity before the agent moves.

Companion to PROMPT

PROMPT helps you speak to LLMs. CTRL helps you direct agents. Use both — one for asking, one for assigning.

Read this guide if

You're a leader, operator, founder, consultant, or assistant whose AI tools are starting to act on their own — and you want them to act on your terms.

Time on page

~15 min, cover to cover

03aibrain.coachThe CTRL Framework
AI Brain CoachThe FrameworkFour controls, at a glance
02   The Framework

Four controls for
agentic work.

A prompt asks for a response. A CTRL handoff assigns work. Four controls cover what the agent needs to know, what it must complete, what it must obey, and how it should check, correct, and stop.

C
Context

What does the agent need to know?

Agents fail when they work without real-world background. Give them the situation, audience, goal, files, prior decisions, and constraints that shape the work.

T
Task

What must the agent complete?

Agents need a finish line. Define the deliverable, the outcome, and what "done" looks like — otherwise the agent produces activity instead of completion.

R
Rules

What must the agent obey?

Rules cover permissions, limits, tone, privacy, tools, approvals, brand, legal boundaries, and escalation. This is where you keep authority.

L
Loop

How should it proceed and stop?

Tell the agent whether to plan first, ask before acting, verify sources, work in stages, or stop for approval. This is what prevents runaway execution.

Each control earns its letter. Skip one and the agent fails in a predictable way. The next four pages take them one at a time.

Turn the page →
04aibrain.coachThe CTRL Framework
AI Brain CoachControl 01 / 04Context
C
Control 01 · Context
01 / 04   ·   Context

Set the world
the agent works in.

Tell the agent what's real before it starts. The more autonomy it has, the more dangerous vague context becomes.

A chatbot can survive a vague question. An agent can't. Context is the operating environment — business goal, audience, situation, files, prior decisions, preferences, market, budget, stakeholders, risks, tone. Without it the agent guesses. With it, the agent works inside reality.

The Move How the Agent Reacts Example Handoffs
No context givenDefault state The agent fills gaps with assumptions — and may optimise for the wrong audience, goal, or business reality entirely. Weak
"Help me prepare for a client meeting."
Basic context providedSituated The agent understands the assignment well enough to produce relevant work. Good
"I'm meeting a mid-sized Canadian manufacturer about AI adoption. They're cautious, non-technical, and worried about cost. Prepare a one-page briefing I can use to guide the conversation."
Operational context providedExecution-ready The agent understands the goal, constraints, audience, assets, and decision environment before acting. Better
"Same brief — likely use cases are sales admin, reporting, and customer service. Position me as practical, not hype-driven. Include three use cases, likely objections, and smart questions to ask."
05aibrain.coachContext
AI Brain CoachControl 02 / 04Task
T
Control 02 · Task
02 / 04   ·   Task

Define what
"done" looks like.

A task is not a topic, an intention, or "look into this." It is an outcome the agent can complete and you can use.

Agents do work. Work needs a finish line. The task tells the agent what to produce, for whom, in what form, and by when — and makes completion obvious. If the agent can't tell whether it's done, the task isn't clear enough. Good tasks turn an energetic intern into a useful operator.

The Move How the Agent Reacts Example Handoffs
Vague taskDefault state The agent produces activity: notes, options, summaries — output without a clear deliverable. Weak
"Research AI tools."
Defined deliverableOutcome-led The agent knows what to produce and can stop when it's complete. Good
"Compare three AI scheduling tools for a five-person consulting team."
Decision-ready taskBusiness-ready The agent produces something a human can review, approve, send, publish, or use. Better
"Same comparison — return a table with pricing, core features, integrations, risks, and a recommendation for which one to test first."
06aibrain.coachTask
AI Brain CoachControl 03 / 04Rules
R
Control 03 · Rules
03 / 04   ·   Rules

Set the boundaries
before it moves.

Rules are how humans stay in authority. Without them, the agent decides its own limits — and may oversell, overshare, or act too soon.

Rules define what the agent can do, can't do, must avoid, and when it must escalate. Tone, brand, budget, privacy, sources, legal limits, approval requirements, tools, geography, risk tolerance. This is the difference between delegation and abdication.

The Move How the Agent Reacts Example Handoffs
No rulesDefault state The agent decides its own limits and frequently oversteps. Weak
"Draft replies to these client emails."
Basic rulesConstrained The agent avoids the obvious mistakes and keeps to the surface of what it's allowed to do. Good
"Draft replies. Do not promise pricing, timelines, discounts, or legal terms."
Authority rulesControlled execution The agent knows what it can do alone, what needs approval, and what must never happen. Better
"Draft replies in a calm, professional tone. Don't promise pricing, timelines, discounts, legal terms, or technical capabilities. Flag anything involving money, contracts, complaints, or risk for human review."
07aibrain.coachRules
AI Brain CoachControl 04 / 04Loop
L
Control 04 · Loop
04 / 04   ·   Loop

Tell it how to work,
check, and stop.

The loop is the operating rhythm. Without it the agent rushes, drifts, overproduces, or keeps working past the useful point.

Agents can move across many steps — useful when the loop is clear, risky when it isn't. The loop tells the agent whether to plan first, ask before acting, work in stages, verify sources, revise after feedback, or stop at a defined output. A strong loop turns a task into a controlled workflow.

The Move How the Agent Reacts Example Handoffs
No loopDefault state The agent may rush, drift, overproduce, or keep working beyond the useful point. Weak
"Create a launch plan."
Review loopCheckpointed The agent pauses at key moments for human approval before continuing. Good
"Create a launch plan. First give me the structure. Wait for approval before writing the full plan."
Quality loopSelf-correcting The agent checks its own work against the task and rules before returning the output. Better
"…then review the draft against audience, budget, timeline, and risks. Stop after the final draft and list the three decisions I still need to make."
08aibrain.coachLoop
AI Brain CoachIn PracticeAgent vs Chatbot
03   In Practice · 01

The relationship
has changed.

A chatbot answers. An agent acts. That single shift changes the skill the human needs to bring. Casual works for chatbots. Agents need a deliberate handoff.

A chatbot conversation

Asking.

The main risk is a weak answer. Stakes are low; revision is cheap.

  • "Explain this."
  • "Summarize this."
  • "Give me ideas."
  • "Rewrite this."
  • "Answer this question."
Prompting
An agent handoff

Acting.

The risk is weak execution: wrong steps, wrong assumptions, wrong tools, wrong audience — work that looks complete but fails in practice.

"Research this, compare the options, prepare a recommendation, draft the email, check for risks, and stop before sending."
Delegation

The more autonomy the system has, the more deliberate the handoff has to be.

CTRL exists for the second world.
09aibrain.coachAgent vs Chatbot
AI Brain CoachIn PracticeThe Handoff Card
03   In Practice · 02

Four questions
anyone can use.

CTRL should fit on a sticky note. Brief the agent the way you'd brief a competent assistant — no orchestration jargon required.

C

What does it need to know?

Context — the situation, audience, files, and constraints.

T

What must it complete?

Task — the deliverable, in plain language.

R

What is it not allowed to do?

Rules — limits, brand, privacy, escalation.

L

When should it check back or stop?

Loop — the checkpoints, review points, and finish.

Example
Handoff card

A one-page executive briefing

Context

Canadian executive audience. Interested in AI but sceptical of hype.

Task

A one-page briefing on how agentic AI will affect administrative work in 2026.

Rules

Plain English. No technical jargon. No legal, HR, or financial claims. Don't exaggerate.

Loop

Outline first. Wait for approval. Then full draft, with three risks to review.

The user doesn't need to understand agents, orchestration, memory, retrieval, or tool calling. They only need to brief the machine the way they would brief a competent assistant.

10aibrain.coachThe Handoff Card
AI Brain CoachIn PracticeWeak vs Strong Delegation
03   In Practice · 03

Most failures begin
with lazy handoffs.

Weak delegation sounds efficient. It isn't. When you say "handle this," the agent invents the missing structure — context, priorities, boundaries, finish line. That's where the errors begin.

Weak
"Plan my content."
Strong
"Create a five-day LinkedIn content plan for Canadian executives who are curious about AI but not technical. Position me as a practical AI advisor. Avoid hype, fear, and jargon. Give me post topics first; wait for approval before writing the posts."
Weak
"Deal with these emails."
Strong
"Sort these emails into urgent, waiting, informational, and ignore. Draft replies only for urgent and waiting. Don't send anything. Flag anything involving money, contracts, complaints, or personal data."
Weak
"Research competitors."
Strong
"Find five Canadian AI consulting firms serving mid-market businesses. Compare positioning, services, pricing signals, proof points, and gaps. Return a table and a short recommendation on how we should differentiate."

Strong delegation isn't longer instructions. It's complete instructions.

Weak delegation creates cleanup.
Strong delegation creates usable work.
11aibrain.coachWeak vs Strong
AI Brain CoachIn PracticeHuman Review Points
03   In Practice · 04

Where judgement
still belongs.

CTRL isn't about letting agents run wild. It's about knowing where human review belongs. Decide before the agent starts.

Review before action

Human approves first.

  • Sending emails
  • Publishing content
  • Contacting clients
  • Making purchases
  • Changing or deleting records
  • Scheduling external meetings
  • Submitting forms
  • Claims about law, finance, health, safety, employment
Review after drafting

Human reviews output.

  • Briefings
  • Reports
  • Content drafts
  • Research summaries
  • Meeting prep
  • Internal notes
  • Training materials
  • Proposal drafts
Review by exception

Agent can run alone.

  • Formatting
  • Sorting
  • Summarising
  • Deduplicating
  • Renaming
  • Organising
  • Extracting data
  • Preparing first-pass options

The agent can prepare. The human approves impact.

Protect the user.
Protect the business.
Protect the brand.
12aibrain.coachHuman Review Points
AI Brain CoachThe ImpactHow each control affects outcomes
04   The Impact

Agentic AI fails
differently.

With prompting, the most common failure is a poor answer. With agents, it's poor execution — and that raises the stakes. CTRL reduces those failures by making the handoff explicit.

4×
more impact per task when context, task, rules, and loop are all set before the agent starts.
~30s
average added prep time per handoff. Pays back in fewer revisions and less cleanup.
0
agent autonomy without human review on impact actions. The agent prepares; the human approves.
Contribution to agent effectiveness

How much each CTRL control matters.

Context
30%
Task
25%
Rules
25%
Loop
20%

Context carries the most weight because the agent acts inside whatever reality it's given. Task and Rules define the finish line and the boundaries. Loop controls execution quality once the work begins.

The four in one breath
  • C
    Context. Set the world.
  • T
    Task. Define done.
  • R
    Rules. Keep control.
  • L
    Loop. Check, correct, stop.
What CTRL improves
  • Less drift. The agent works inside reality.
  • Less busywork. The task defines completion.
  • Less risk. Rules prevent overreach.
  • Less rework. The loop self-corrects.
13aibrain.coachThe Impact
AI Brain CoachNext StepsWhere you go from here
05   Next Steps

Where do you
go from here?

Agentic AI isn't just a new tool category — it's a new work relationship. You're no longer only a question-asker. You're a manager of machine work. Three principles keep CTRL useful as the technology keeps moving.

01

Handoff over prompt.

Don't start by asking "what should I type?" Start by asking "what work am I handing off?" The question reframes the whole interaction. You stop chatting and start delegating.

02

Control over autonomy.

Don't give the agent freedom because the software can handle it. Give the agent freedom only where the rules, review points, and stopping conditions are clear. Autonomy without a leash is just risk wearing a productivity costume.

03

Judgement over automation.

The agent can move the work forward. The human still owns the decision. CTRL keeps that relationship intact — and that's the relationship clients, teams, and regulators will all eventually ask you about.

Set the context. Define the task. Write the rules. Run the loop.

End of guide
14aibrain.coachNext Steps
AI Brain CoachCreditsContributors & contact
Closing

AI mastery now
requires better
delegation.

Thank you for being part of this next stage of the work. PROMPT helped define how leaders speak to AI when they need better answers. CTRL extends that discipline into the agentic era — where AI doesn't merely respond but assists, executes, and moves through work.

Continue the work

For coaching, frameworks, and tooling that turn AI into a leadership instrument, visit aibrain.coach or reach the team at hello@aibrain.coach.

Contributors
Concept & Framework

Ray Tan

Built as a companion to The PROMPT Framework.

Research & Strategy

Ray Tan

Content & Writing

Ray Tan

Contact

Studio

AI Brain Coach

aibrain.coach

Reach out

aibrain.coach

hello@aibrain.coach
@aibraincoach

AI Brain Coach

The CTRL Framework · Vol. 01 · Revised 2026

© AI Brain Coach 2026
All rights reserved

15aibrain.coachCredits