Prompting & Task Execution

Screen 1: Prompting & Task Execution

Module 2Introduction·4 min

The same request, phrased two ways, produces two different levels of quality.

Ask Claude to "write something about our Q3 results" and you get a generic paragraph. Specify the audience, the three results that matter, the format, and the length, and you get a draft you can almost send. The model did not get smarter between those two requests. The prompt did.

This module treats prompting as a communication discipline with learnable structure, not a knack some people have and others do not. The structure has a name in the AI Fluency Framework: Description, the competency of telling Claude precisely what you want. Description is the backbone of this module and the prompting foundation the rest of the course builds on.

By the end of this module, you will be able to:

• 1Create effective prompts for business and technical tasks using a repeatable component structure.

• 2Apply task decomposition techniques to structure complex, multi-part requests.

• 3Iterate prompts diagnostically to improve output quality.

• 4Adapt prompting strategy to the task type: analysis, research, drafting, or brainstorming.

Consistent output quality comes from prompt structure, not cleverness or luck. Learn the five components, learn to decompose complex work, learn to iterate on the component that failed, and learn to match strategy to task type. Those four skills cover the entire Prompting section of the exam.

We built this Associate course Module 2: Prompting & Task Execution to help you get real work done with Claude. Treat it as educational content. It doesn't constitute legal, financial, or other professional advice, so adapt what you learn to your own situation. Our products and services evolve quickly, so certain content may contain errors or be outdated; remember to verify on Anthropic's website or docs. Examples and scenarios used in the course are illustrative and often fictitious. If the course material mentions a company or product, it doesn't mean Anthropic endorses them, they endorse Anthropic, or that we're affiliated. Also note your use of Anthropic products and services is covered by our terms, policies and documentation; if anything in this course conflicts with them, they control.

---

Screen 2: Anatomy of an Effective Prompt

TeachingAnatomy of a Prompt·12 min

A strong prompt is built from components, and most weak prompts are missing one or more of them.

Naming the components turns prompting from guesswork into a checklist you can run before sending any non-trivial request.

The component stack

Five components carry almost all the weight in a professional prompt. Flip each card to see what it controls.

Role

Who you want Claude to be for this task: a financial analyst, an editor, a policy reviewer. Role sets the vocabulary, depth, and assumptions Claude brings.

Context

The background Claude cannot know unless you provide it: the audience, the situation, prior decisions, the source material. This is the component professionals most often omit.

Task

The specific action, stated as a clear instruction. "Summarize," "compare," "draft," "identify": one primary verb, stated unambiguously.

Constraints

The boundaries: length, tone, what to include, what to leave out, what to avoid. Constraints are how you keep the output usable without heavy editing.

Output format

The shape of the result: a table, a bulleted list, a three-paragraph memo, a draft email. Stating the format up front saves an iteration.

Not every prompt needs all five. A quick question needs a task and maybe a constraint. A client deliverable needs all five. The skill is noticing which components a given task requires.

Description in practice

The Description competency is the habit of making each component explicit instead of assuming Claude will infer it. Unless you connect a source through a Connector, Claude cannot see your inbox, your org chart, or last week's meeting, and even with a connector, Claude sees only what you have allowed it to access. Anything that lives only in your head or outside a connected source is a context gap, and context gaps are the single most common reason a prompt underperforms for new users.

Diagnosing a weak prompt

Hold a disappointing prompt against the component stack and the gap usually becomes obvious. Missing context produces generic output. An ambiguous task verb produces the wrong action. Absent constraints produce output that is the wrong length or tone. The components are also a diagnostic checklist, which is why Lesson 4 returns to them when output falls short.

---

Screen 3: A Worked Build

TeachingAnatomy of a Prompt·included in L2

Weak prompt: everything left implicit

Claude produces three plausible paragraphs that could describe almost any company. No audience, no figures, no format, no sense of what matters. The output is not wrong. It is unusable, because the prompt specified almost nothing.

Strong prompt: components made explicit

Same model, same data. The second prompt produces a draft the analyst can refine in two minutes instead of rebuilding from scratch. The difference is entirely in the specification.

The habit to build

Run the five components in your head: have I given Claude the role, the context it cannot infer, an unambiguous task, the constraints, and the format I want back? Thirty seconds of specification routinely saves several rounds of correction.

---

Screen 4: Task Decomposition for Complex Requests

TeachingTask Decomposition·10 min

Some requests are too large to specify as a single instruction.

When a task has several distinct stages, packing it into one prompt produces shallow work on every stage. Decomposition is how you break a complex request into a sequence Claude can execute well.

Decomposition means splitting a multi-part problem into discrete, ordered steps, then running them in sequence rather than asking for everything at once. A vendor evaluation is a good example: it is really four tasks wearing one sentence.

The single-prompt version that underperforms

Claude has to invent criteria, apply them, weigh trade-offs, and recommend, all in one pass. It will do all four shallowly and you will not see the reasoning behind the recommendation.

The decomposed version

Each step produces a checkable intermediate result. If the criteria in Step 1 are wrong, you catch it before scoring, not after the recommendation. Decomposition also makes the work auditable, which matters when someone asks how the recommendation was reached.

One conversation or several

Keep sequential steps that build on each other in one conversation, so each step sees the prior results. Move to a separate conversation when a step is genuinely independent, or when the conversation has grown long enough that early context is degrading. That judgment connects directly to the context-management skills from Module 1.

---

Screen 5: Decompose a Parallel Case

TeachingTask Decomposition·included in L3

Three deliverables, one foundation: sequence the shared extraction first.

A communications manager needs to turn a dense 20-page policy change into an internal announcement, an FAQ for staff, and a short briefing for executives. Before reading on, decompose this into an ordered sequence of steps you would run with Claude.

Model decomposition

Step 1: Extract the substantive changes from the policy document and what each one means in practice.

Step 2: Confirm the extraction is complete and accurate before building anything on top of it.

Step 3: Draft the staff announcement from the confirmed change list, tuned to a general audience.

Step 4: Draft the FAQ, anticipating the questions staff will likely ask about those changes.

Step 5: Draft the executive briefing, compressed down to decisions and impact.

Why this order

Steps 1 and 2 build a verified foundation that the three deliverables all draw on. Drafting any deliverable before the change list is confirmed risks propagating the same misreading into three documents. Sequence the shared, high-stakes extraction first; let the parallel drafts follow.

---

Screen 6: Iterating Prompts to Improve Output

TeachingIterating to Improve·8 min

A first draft from Claude rarely lands perfectly. The skill is not rewriting the whole prompt when output disappoints. It is reading the output to diagnose which component fell short, then fixing that one thing.

Output deficiencies are prompt diagnostics

Each kind of disappointment points back to a specific component:

Symptom

Likely cause

Fix

Output is generic or off-base

The context was thin

Add the background Claude could not infer

Output answered the wrong question

The task verb was ambiguous

Sharpen the instruction

Output is the wrong length, tone, or shape

A constraint or the format was missing

Add it

Output is close but misses on one section

Iterate on that section only; do not discard a draft that is mostly right

Targeted revision, not wholesale rewriting

When you rewrite the entire prompt, you lose the parts that worked and you cannot tell which change fixed the problem. Change the one component the output told you to change, resend, and compare. The same diagnostic discipline applies to troubleshooting any underperforming workflow, covered in depth in Module 7.

A live iteration cycle

Watch the diagnose-and-fix loop run on one deliberately weak prompt.

Round 1 prompt

"Write a follow-up email to the client about the delayed deliverable. "

Round 1 output

A generic, slightly defensive three-paragraph email that does not say when the deliverable will arrive or why it slipped. Diagnosis: the context is thin (no reason, no new date) and there is no tone constraint.

Round 2 prompt

"Write a follow-up to the client about the two-day delay on the analytics deliverable. The cause was a data-quality issue we have now fixed; new delivery is Thursday. Tone: accountable, not over-apologetic. Keep it under 120 words. "

Round 2 output

A tight, accountable note with the cause, the new date, and a confident close. Diagnosis: strong; only the subject line is missing.

Round 3 prompt

"Good. Add a subject line that signals resolution, not just delay. "

Three rounds, each changing exactly the component the previous output exposed. No round threw away working text, and by round three the improvement was marginal, the signal to stop.

Knowing when to stop

Iteration has converged when each additional round produces marginal change rather than improvement. At that point, further prompting yields less than a quick manual edit. Recognizing diminishing returns is part of the skill: the goal is a usable result, not a perfect prompt.

---

Screen 7: Adapting Strategy by Task Type

TeachingStrategy by Task Type·8 min

The component stack applies to every prompt, but the emphasis shifts with the task.

Analysis, research, drafting, and brainstorming each reward a different balance of specificity and creative latitude. Using one fixed style across all four costs quality on every task it does not fit.

Analysis. Wants tight constraints and explicit criteria. Tell Claude what to measure, against what standard, and how to handle ambiguity. Low creative latitude; high specification.

Research. Wants clear scope and source discipline. Define the question, the boundaries, and whether current sources are required. Quick currency needs can be met by turning on web search in chat, while deep multi-source investigation points to Research (available on paid plans). Ask for citations so claims are checkable.

Drafting. Wants audience, tone, and format specified, with room for Claude to find the phrasing. Medium latitude: you control the shape, Claude fills it.

Brainstorming. Wants loose constraints and high latitude. Over-specifying kills the divergence you are after. Give the goal and the boundaries, then ask for volume and range before you narrow.

Strategy quick reference

Task type

What to tighten

What to loosen

Analysis

Criteria, standards, scope

Phrasing

Research

Question, sources, citations

Synthesis approach

Drafting

Audience, tone, format

Word choice

Brainstorming

Goal and guardrails only

Quantity and direction

Four mini-demos, one per task type

Analysis. "Compare these two vendor contracts on payment terms, termination rights, and liability caps. For each, state which contract is more favorable to us and why, in a three-row table. " Tight criteria, defined output, no room to wander.

Research. "Using current sources, summarize how three named competitors positioned their Q2 launches. Cite each source. Flag anything you cannot verify. " Scope and citation discipline up front; Research (or web search in chat for lighter needs) supplies the currency.

Note: citations are checkable when they come from a grounded source (web search or Research results); citations produced from training memory alone can look equally confident and should be independently verified.

Drafting. "Draft a 150-word LinkedIn post announcing our new reporting feature, aimed at operations managers, in a confident but not salesy voice. " Audience, length, and tone fixed; phrasing left open.

Brainstorming. "Give me 20 angles for a campaign around faster month-end close. Range widely; do not self-edit yet. " Goal and one guardrail only; the constraints come later, after the range exists.

The same prompt structure underlies all four, but the dial between control and latitude moves with the task. The underlying move is the same each time: decide where you need control and where you need range, then set constraints accordingly. Matching strategy to task type is what separates a competent prompter from one who gets the same mediocre output on every task.

---

Screen 8: Checkpoint: Diagnose the Prompts

CheckpointStrategy by Task Type·5 min

Four prompts below each have one dominant weakness. For each, identify the missing or mishandled component and state the single change that would most improve the output.

---

Screen 9: Exercise: Repair the Underperforming Prompt

ExerciseRepair the Prompt·6 min

This exercise puts the whole module together: a weak prompt, its disappointing output, and your job to diagnose what is missing and repair it using the component stack.

Below is a prompt, the output it produced, and the author's actual goal. Work in three steps: find the specification gaps, match each fix to the component it repairs, then assemble the repaired prompt. The ideal answer is revealed at the end.

"Summarize the customer feedback and tell me what to do. "

A generic five-bullet list of themes ("customers want faster support," "pricing is a concern") with vague advice ("consider improving response times"). Nothing tied to the actual data, no priorities, nothing actionable.

A product manager has 200 survey responses and needs the top three issues by frequency, each with a representative quote, ranked so she can decide what to fix this quarter.

Select every component the prompt is missing or under-specifies.

Drag each fragment to the component it satisfies. One fragment does not belong.

Order the five fragments into a clean prompt sequence.

---

Screen 10: Module 2 Quiz: Prompting Foundations

QuizModule 2·5 min

Five scenario-style questions. Each presents a situation; select the response that best applies the module's prompting framework. Approximately five minutes.

---

Screen 11: Key Takeaways

Module 2Key Takeaways·5 min

Five things that hold across this module:

Structure drives quality, not cleverness.

Five components (role, context, task, constraints, format) carry almost every professional prompt. Run them before sending anything that matters.

Context is the component you will forget.

Claude cannot see what lives only in your head. The most common cause of generic output is a context gap, not a model limitation.

Decompose complex work into ordered steps.

Multi-stage requests succeed when each step produces a checkable result and the high-stakes foundation is built first.

Iterate on the component that failed.

Read the output as a diagnostic, change the one thing it points to, and stop when rounds stop improving the result.

Match strategy to task type.

Analysis wants constraints; brainstorming wants latitude. Decide where you need control and where you need range, then set constraints to fit.

All product behavior descriptions are based on claude. ai features as of June 2026. Feature availability and behavior should be verified against current Anthropic documentation at publish:

• Anthropic docs: prompt engineering overview and best practices, platform. claude. com/docs

• AI Fluency Framework: Description competency, for the prompting backbone framing

• Claude Help Center: Code execution for verified calculation, support. claude. com

---

Screen 12: Congrats! You’ve successfully completed this module.

Module CompleteAssociate Path·2 min

You can now structure prompts that get consistent, high-quality results from Claude. Master the prompt, and Claude does exactly what you need, every time.

M1: Product & Model Selection

Choose the right entry point, model, and features for any given task.

M2: Prompting

Build structured prompts and adapt them to the task type.

M3: Output Evaluation

Validate output and know when human review is non-negotiable.

M4: Workflow Integration

Map a workflow against Delegation criteria and redesign it safely.

M5: Configuration

Configure and maintain Projects, instructions, and knowledge.

M6: Governance

Apply use-case, data, policy, and ethics judgment responsibly.

M7: Troubleshooting

Diagnose underperformance and optimize workflows when results fall short.

M8: Course Summary & Next Steps

Recap the journey, prepare for the exam, and recognize escalation boundaries to the Developer and Architect tracks.