AIThis post was created with the assistance of artificial intelligence (AI).

TL;DR

Claude Code creator Boris Cherny says useful prompts need three things: a clear task, guidance on how much effort the model should spend and a way to verify the result. An XDA Developers writer examined that advice alongside a prompt Cherny shared for an interactive Home Depot project. The source does not report controlled results showing how much the tips improve AI output.

Claude Code creator Boris Cherny says effective AI prompts should make clear what the model should do, how much effort to spend and how to check its work. In an XDA Developers article published Oct. 22, writer Mahnoor Faisal examines that advice alongside a conversational prompt Cherny shared for an interactive project based on a Home Depot episode of the Acquired podcast.

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

Cherny’s three recommendations are practical rather than a new prompting framework: state the task clearly, specify the level of effort appropriate to it and give the model a way to determine whether the result meets the request. The advice addresses both how users describe a goal and how they ask an AI tool to handle the work.

The example prompt asks Claude to create an artifact narrating the episode and foregrounding its business lessons. It also requests an interactive, visually polished presentation, with watercolor imagery and historical photographs, and encourages Claude to iterate on the result. The prompt uses ordinary conversational language rather than a rigid template.

Faisal describes the prompt as an illustration of Cherny’s approach. The article does not provide a controlled comparison of outputs with and without the advice, nor does it report measurements of time, accuracy or quality. Its examination of the example prompt is not evidence that the three tips will produce the same results on every task.

At a glance
reportWhen: XDA Developers article published Oct. 2…
The developmentAn XDA Developers article published Oct. 22 examines Claude Code creator Boris Cherny’s three-part prompting advice and a prompt he shared for an interactive project about a Home Depot podcast episode.
I Tested Claude Code’s Creator’s Advice For Better Prompts

Prompting · A practical checklist

I Tested Claude Code’s Creator’s Advice For Better Prompts

Boris Cherny’s advice boils down to three decisions: define the task, set an appropriate effort level, and give the model a way to check its work. An XDA Developers article explores the advice through one ambitious example.

Oct. 22Article published
Oct. 6Prompt shared
Home DepotPodcast subject
No trialControlled results

01 / The checklist

Three checks before you prompt

The advice is a straightforward way to make a request more actionable. Specify the outcome, describe the work expected, and say how success can be checked.

01 — DEFINE

State the task

Describe what the AI should produce or change. A clear goal gives the work direction and helps the model understand what matters.

02 — SCOPE

Set the effort

Indicate how much work the task calls for. Match the expected depth to its complexity; the article does not prescribe a universal prompt length.

03 — VERIFY

Plan a check

Give the model a way to assess its output, such as running tests for code or checking the result against stated requirements.

Goal → DirectionEffort → ScopeCheck → Evidence

02 / The example

A Home Depot story, built as an interactive artifact

Cherny’s conversational prompt asked Claude to narrate an Acquired episode about the retailer and bring its business lessons to life with a polished, interactive presentation.

Style direction in the prompt
“Make it beautiful and interactive and focus on communicating the nuggets of business wisdom throughout the episode.”
PurposeNarrate the full episode and surface its business insights.
Visual referencesUse watercolor illustrations and historical photographs.
Iteration“Use lots of tokens and iterate till you’re proud of it.”

03 / What the article establishes

An illustration, not a measured prompt test

The example shows how the three ideas can appear in an ordinary request. It does not establish that this style reliably improves results.

What readers can take from it

  • ✓One example connecting task, style, effort, and iteration.
  • ✓A practical reminder to define how output should be checked.
  • ~Tests can reveal some defects, but cannot guarantee software is problem-free.

What remains unproven

  • ×No controlled comparison with other prompt styles.
  • ×No independent quality, accuracy, time, or cost measurements.
  • ×No evidence the advice works the same across models and tasks.

04 / Put it to work

Turn advice into a checkable request

A detailed prompt alone does not guarantee success. Use a requirement or test to judge the result, then decide what needs revision.

1

Name the outcome

Say what you want the model to create, explain, or change.

2

Describe the scope

Set the depth and effort that fit the task.

3

Choose a check

Use tests, sources, or a requirements list as appropriate.

4

Review the result

Compare the output with the check and revise where needed.

Keep the limits in view: How much control users have over model effort, and how consistently models verify or revise work, can vary. The article reports no follow-up experiment.

05 / Key questions

What to remember

What are the three tips?

State what the AI should do, specify how much effort to spend, and provide a way to check whether the work meets the request.

What was the Home Depot project?

An interactive artifact narrating an Acquired podcast episode about Home Depot, focused on business lessons and visual storytelling.

Did the article prove the advice works better?

No. It examines an example prompt but reports no controlled comparison or measured improvement across tasks.

Why ask a model to verify its work?

A specific check gives the model a way to assess its output. For code, tests can help identify problems, though they do not prove the software is defect-free.

Three Checks Before You Prompt

The advice offers AI users a three-part checklist for making requests more specific: state the desired outcome, indicate the level of effort appropriate to the task and set out how to check the result. A clear goal gives the model direction; an effort instruction describes the expected scope; and a verification step provides a way to assess whether the output meets the request.

Verification is relevant when an AI tool is asked to create or modify software. Cherny’s example is to ask Claude Code to run tests after building an application. Tests can identify some defects, but they do not establish that software is free of problems. For other tasks, verification could mean checking a result against specified requirements or sources.

Cherny’s example includes details about the desired outcome and style while using conversational wording. The article does not prescribe a universal prompt length; it describes the need to include information relevant to the task.

Products Worth Considering

LG 27GS60QC-B Ultragear 27-inch Curved Gaming Monitor (QHD, 180Hz, 1ms, 1000R)

LG 27GS60QC-B Ultragear 27-inch Curved Gaming Monitor (QHD, 180Hz, 1ms, 1000R)

Bose QuietComfort Ultra Earbuds (2nd Gen), Wireless Noise-Cancelling Earbuds, Black

Bose QuietComfort Ultra Earbuds (2nd Gen), Wireless Noise-Cancelling Earbuds, Black

As an affiliate, we earn on qualifying purchases.

The Home Depot Prompt Example

Cherny shared the Home Depot project prompt on Oct. 6, 2026, according to the XDA article. He had built an interactive website using Claude to accompany an Acquired podcast episode about the retailer. In response to the prompt, Cherny described his three-part approach.

In the prompt, Cherny asks Claude to narrate the full episode and communicate its business insights. He also names desired features and visual references, including watercolor illustrations and historical photographs. The requests describe the project’s purpose and presentation while leaving implementation choices to the model.

The article discusses the advice alongside differing public approaches to prompting, including recommendations to provide extensive context and to use short, tightly specified instructions. Cherny’s three points focus on the task, effort and verification. The source does not establish a universal rule about prompt length or show that one approach works best across different models and jobs.

“A good prompt comes down to just three things.”

— Boris Cherny, Claude Code creator, as quoted in the XDA Developers article

What the Prompt Test Cannot Show

The source does not describe a controlled test comparing Cherny’s approach with other prompt styles. It provides no independent score for the resulting artifact, no benchmark of factual accuracy and no evidence that the prompt’s visual or interactive features worked as requested. The article therefore illustrates the advice but does not establish its effectiveness across tasks.

It is also unclear how much control a user can exercise over a model’s effort in every AI product, or how consistently a model will follow requests to verify and revise its output. The article discusses effort as a way of matching a model’s work to task complexity, but does not quantify the cost, time or quality effects of choosing different effort levels.

Applying the Advice to Other Tasks

Readers can apply Cherny’s checklist by stating the desired outcome, describing the expected level of work and identifying a way to check the result. For coding, that could mean running tests; for other work, it could mean checking the output against a list of requirements. The result can then be assessed against those requirements; a detailed prompt alone does not guarantee success.

The XDA article does not announce a follow-up experiment or provide a schedule for further testing. Whether these tips improve results will depend on the task, the model and the verification method; the source does not address those outcomes.

Key Questions

What are Boris Cherny’s three prompting tips?

He recommends stating what the AI should do, specifying how much effort it should spend and giving it a way to check whether the work meets the request.

What was the Home Depot project?

Cherny’s prompt asked Claude to create an interactive artifact narrating an Acquired podcast episode about Home Depot, with a focus on business lessons and visual elements such as watercolor illustrations and historical photographs.

Did the article prove that Cherny’s approach works better?

No. XDA’s article examines the example prompt, but it does not report a controlled comparison or measured results showing that the advice improves output across tasks.

Why ask an AI tool to verify its work?

A verification step gives the tool a specific way to check its output rather than relying on an assumption that it succeeded. For code, Cherny’s example is asking Claude Code to run tests; tests can help identify problems but do not guarantee defect-free software.

Source: XDA Developers

You May Also Like

7 Best PC Processors for Prime Day Deals in 2026

Discover the best PC processors to buy during Prime Day 2026, including AMD and Intel options for gaming, productivity, and future upgrades.

The New Personal Agent Layer

OpenClaw and Hermes introduce a new layer of persistent personal action agents, transforming how AI interacts with digital environments. Details are emerging.

Three Sites Made 215,128 “Best Software” Pages For AI. Perplexity Cites Them

Perplexity reports that three sites have generated 215,128 pages ranking as ‘best software’ for AI tools, highlighting a surge in AI-related content.

SWE-1.7 Reach Near GPT 5.5 And Opus Intelligence

SWE-1.7, the latest AI model, has reached performance levels close to GPT 5.5 and Opus Intelligence, signaling significant advancements in AI capabilities.