Home›AI Agency Academy›Lesson 04
Module 01 · Lesson 04

Decide whether AI belongs in the workflow

Use an explicit scorecard before investing in an agent build.

Last updated August 5, 2026•15–25 minutes•Free AI agent course
What you will learn

Make a clear, safer operating decision.

You will be able to make a disciplined go, redesign, or no-go decision for a proposed agent workflow.

Why this matters

Good agent work is useful before it is impressive.

AI is not a substitute for missing process design. A good candidate has recurring volume, enough trustworthy context, a bounded action, and a simple way for a person to catch mistakes.

Field note 04

Make the relationship visible.

AI AGENTS · FIELD NOTE 04Value + frequency − risk + fallbackTHE SCORECARD01Repeated work02Useful context03Action risk04Human fallbackOriginal visual framework for Decide whether AI belongs in the workflow.AI AGENTS · FIELD NOTE 04Value + frequency − risk + fallback01Repeated work02Useful context03Action risk04Human fallback
Use this framework to make decide whether ai belongs in the workflow visible before you build.
Core concepts

The language that keeps the work clear.

FrequencyHow often the problem occurs and whether a repeatable design will be used enough to justify maintenance.
Context qualityWhether the agent receives current, relevant, permitted information instead of guessing.
Action riskThe harm if the agent is wrong: inconvenience, cost, compliance exposure, safety, or loss of trust.
FallbackThe concrete route a person takes when the agent cannot proceed confidently.
The practical method

Work through the decision in order.

Score the opportunity

Rate frequency, value, context quality, action risk, and fallback strength on a simple 1–5 scale.

Read the imbalance

High value does not cancel high risk. Weak context does not become safe because the prompt is longer.

Reduce scope

If the score is mixed, change the proposal from acting to recommending, or from all customers to a controlled segment.

Write the decision

Record why the workflow is proceeding, redesigning, or staying manual. Revisit when evidence changes.

Worked example

A realistic, bounded implementation.

An ecommerce brand wants an agent to issue refunds. Volume is high, but product conditions, fraud risk, and payment consequences make the action risky.

The scorecard turns the first release into a recommendation workflow: the agent gathers order context, checks an approved policy, drafts the recommended resolution, and asks a support lead to approve the refund.

After a clean evaluation period, the brand may allow low-risk replacements within defined limits. It never treats a broad payment action as the default starting point.

Build it in practice

Use this copyable working template.

Adapt it to the client’s evidence, policy, people, and tools. Do not treat placeholders as approved instructions.

Frequency: [1–5]. Value if improved: [1–5]. Context quality: [1–5]. Action risk: [1–5]. Fallback strength: [1–5]. Decision: [go / redesign / no-go]. Reason: [one sentence].
Spacebrain implementation

Put the operating system around the agent.

Use permissions, task assignment, automation conditions, and approval-oriented workflows to keep a first release narrow and recoverable.

Practice

Before you move on

  • Score one proposed agent honestly.
  • Write one design change that lowers its risk.
  • Name the person who receives failed or uncertain cases.
  • The decision considers risk and fallback, not just upside.
  • The agent has the context it needs or the scope has been reduced.
  • A controlled pilot is preferred when uncertainty remains.
  • The decision is written down for future review.

Build the operating layer around your agent.

Use the free Spacebrain workspace to keep contact context, handoffs, tasks, automation, and reporting together.

Start for free →