What you will be able to do
- Break an existing process into steps before deciding what Claude handles
- Sort steps into ones that are safe to hand off and ones that must stay with a person, and explain why
- Write a pass/fail definition of done before delegating a step
Key concept
Delegation (AI Fluency) — Delegation is the competency of splitting a process into its individual steps and then choosing, step by step, which ones Claude handles and which stay with a person. It is decided per step, not as a single yes or no for the whole workflow.
1.Decide per step, not per process
When someone asks you to "add Claude to our process," it is tempting to treat it as one question: do we use AI here or not? The AI Fluency framework uses a smaller unit of decision. Its first competency, Delegation, is about deciding what work should be done by humans and what by AI, and you make that decision for each step of the work.
The builders' version of the course puts it plainly. Delegation is not "should I use AI here?" It is about how you break a problem down and what role AI plays in each part. You decompose first and assign second. Take a monthly reporting process. It might involve gathering figures, reconciling them, drafting commentary, and signing off. Those steps differ in volume, in how clearly they are specified, and in who has to answer for them, so they will not all fall on the same side of the line.
The framework splits Delegation into three sub-components. Each one asks a different question when you review a workflow:
| Sub-component | What the framework says | Question to ask of each workflow step |
|---|---|---|
| Problem Awareness | Understanding your goals | What is this step actually for, and what outcome does it protect? |
| Platform Awareness | Knowing your tools' capabilities and limitations | Can Claude do this step well, and where does it fall short? |
| Task Delegation | Distributing work thoughtfully | Given the first two answers, does this step go to Claude, stay with a person, or get split? |
Delegation also sits in a pair. The framework calls Delegation and Diligence the outer loop: together they cover the bigger decisions about when and how to use AI. Description and Discernment form the inner loop of day-to-day interaction. Integrating Claude into a workflow is mostly outer-loop work. You settle it before anyone writes a prompt.
Checkpoint 1 of 4· Check yourself
A team lead is listing what Claude can and cannot reliably do before deciding which onboarding steps to hand over. Which Delegation sub-component is this?
Platform Awareness means knowing your tools' capabilities and limitations. Problem Awareness is about your goals, and Task Delegation is the act of distributing the work.
“It includes Problem Awareness (understanding your goals), Platform Awareness (knowing your tools' capabilities and limitations), and Task Delegation (distributing work thoughtfully)”Source: academy.claude.com
2.What to hand off, and what stays with a person
With the steps laid out, the builders' course gives a short rule for sorting them: delegating implementation is safe, and delegating judgment is not.
Implementation steps are well specified and produce output you can check. Examples are turning a CSV into a formatted quarterly report, converting a Word document into a presentation, or pulling the tables out of a PDF into a spreadsheet. Claude's file-creation features are built for this kind of repetitive, text-heavy work, and they can chain several format conversions into one workflow. These steps are the strongest candidates for handing off, as long as someone checks the result.
Judgment steps are different. Examples are deciding what a customer actually needs, whether to grant an exception, or whether a document is ready to go out. You cannot simply hand these over, because someone has to answer for the outcome. The fourth competency, Diligence, says this directly: you own the final result fully, whatever role AI played. Its Deployment Diligence sub-component means verifying and vouching for outputs. The educators' course offers a blunt test: would you put your name on the final output? If the whole point of a step is that a named person vouches for it, that step stays with that person. Claude can prepare material for it but cannot take it over.
The sources draw this line around judgment and accountability. They do not give a fixed list of forbidden tasks. The table below applies that rule; it is not an official catalogue.
| Kind of step | Delegation decision | Why |
|---|---|---|
| Well-specified conversion or extraction (CSV to report, PDF tables to spreadsheet) | Hand to Claude, then check | Implementation work with checkable output |
| Drafting from existing context (a thread reply, a meeting briefing) | Hand to Claude as a draft a person reviews | The person still decides what is shared |
| Framing the problem or deciding what users need | Keep with a person | Delegating judgment is not safe |
| Final sign-off a named person must vouch for | Keep with a person | Diligence: verifying and vouching for outputs |
Checkpoint 2 of 4· Exam question
An insurer's manual claims process involves five separate teams handing a claim off through email at each stage. Leadership decides to rebuild the process so a single Claude-driven agent triages, requests documents, and routes each claim end to end. Which term best describes this change?
Correct answer: A — Redesigning the workflow, since the multi-team handoff process is being replaced by an agent-driven process
- A. Correct. Replacing the multi-team, email-driven handoff structure with a single agent that owns triage, document requests, and routing end to end is a redesign of the workflow rather than a small addition to it.
- B. Incorrect. The five-team handoff structure is being removed, not preserved, so this is not augmentation of the existing process.
- C. Incorrect. Nothing in the scenario describes grouping claims for batch processing; the change is about who owns each step of the process.
- D. Incorrect. The scenario does not describe reusing prior claim decisions as cached context; it describes restructuring who performs each step.
Checkpoint 3 of 4· Check yourself
Claude now drafts the summaries a manager reads before approving large customer refunds. What should happen to the approval step itself?
Drafting the summary is implementation and can be delegated. The approval is a judgment that someone has to vouch for, so it stays with the manager.
“Deployment Diligence (verifying and vouching for outputs)”Source: academy.claude.com
3.Define "done" before you delegate
A step is only safe to hand over if you can tell whether Claude did it well. The builders' course says to write acceptance tests before code, so that you and AI share a definition of done. The same discipline works outside software. Before delegating a step, write a few statements that settle pass or fail, each concrete enough that a stranger could apply it.
The course contrasts two statements. A good one: "A patient with a basic smartphone can find the current wait time in under 30 seconds without creating an account." A bad one: "The system is easy to use." The second statement can never clearly fail, so it can never be checked either.
The course's practice exercise follows the same order. Before opening AI, you write down which competencies apply and what "good enough" looks like. Then you take one round with AI and review what came back. Defining done first gives the inner loop something to measure against. Discernment means evaluating what AI gives back before acting on it, and without written criteria that evaluation is just opinion.
Defining done also changes where people put their effort. As AI speeds up implementation, the course argues, your value moves to framing problems and raising the bar. In a redesigned workflow, the person who used to do a step often becomes the person who writes its acceptance criteria.
Checkpoint 4 of 4· Check yourself
A team is about to hand Claude the weekly status summary. Which definition of done actually lets them check Claude's output?
Only the second option lets a stranger say whether a given summary passes or fails. The others are impressions, not criteria.
“Each must be concrete enough that a stranger could tell you whether the finished product passes or fails.”Source: academy.claude.com
Exam traps
Each one states something that sounds right. Open it to see what is actually true.
1.Delegation is a single decision about whether a team or process will use AI.Why is that wrong?
Delegation is made step by step. You first break the process into steps, then decide what AI handles in each, so one workflow usually ends up with both delegated and retained steps.
Covered in Decide per step, not per process
2.Once Claude has produced the output, responsibility for it shifts to the tool.Why is that wrong?
Diligence keeps ownership with the person. You own the final result fully whatever AI contributed, so sign-off and vouching steps stay human.
Sources
Every claim above is drawn from one of these pages, quoted as it was written on the date shown.
- 1.
“Delegation involves deciding what work should be done by humans versus AI.”
↩︎ Decide per step, not per process“It includes Problem Awareness (understanding your goals), Platform Awareness (knowing your tools' capabilities and limitations), and Task Delegation (distributing work thoughtfully)”
↩︎ Checkpoint“Deployment Diligence (verifying and vouching for outputs)”
↩︎ Checkpoint - 2.https://academy.claude.com/courses/ai-fluency-for-builders/delegation-the-builder-s-toolkitOfficial docs
“Delegation means decomposing the problem first, then deciding what AI handles at each step.”
↩︎ Decide per step, not per process“Explain why delegating implementation is safe and delegating judgment is not”
↩︎ What to hand off, and what stays with a person“Write acceptance tests before code. They give you and AI a shared definition of done.”
↩︎ Define "done" before you delegate“Delegation means decomposing the problem first, then deciding what AI handles at each step.”
↩︎ Key concept“Delegation means decomposing the problem first, then deciding what AI handles at each step.”
↩︎ Exam trap 1“Each must be concrete enough that a stranger could tell you whether the finished product passes or fails.”
↩︎ Checkpoint - 3.
“Claude can even support workflows requiring multiple file format conversions.”
↩︎ What to hand off, and what stays with a person - 4.
“Would you put your name on the final output?”
↩︎ What to hand off, and what stays with a person - 5.
“Discernment means evaluating what AI gives back before acting on it.”
↩︎ Define "done" before you delegate“Diligence means owning the final result fully, regardless of AI’s role.”
↩︎ Exam trap 2