TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Claude Code creator Boris Cherny says useful prompts should clearly state the task, the effort expected and how the result can be checked. An XDA Developers writer examined his prompt for an interactive Home Depot project and said it illustrated those principles, though the source provides no controlled test comparing results.
An XDA Developers article published Oct. 22 examines three prompting principles shared by Boris Cherny, the creator of Claude Code: state the task clearly, specify how much effort the model should spend, and give it a way to check its work. The advice is relevant to people using AI coding tools because it offers an approach to structuring prompts, though the article does not report a controlled test proving the method improves results.
Cherny shared an example prompt on X on Oct. 6, 2026, alongside an interactive website he had made with Claude for an Acquired podcast episode about Home Depot, according to XDA. The prompt asked Claude to create an artifact narrating the episode, make it interactive and visually appealing, and emphasize business lessons. It also suggested watercolor illustrations, historical photographs and a visual style inspired by publications such as The New Yorker and The New York Times.
XDA reports that the conversational prompt drew attention for its relatively straightforward wording. Cherny later described three elements he considers useful: explain what the model should do; indicate the desired level of effort; and give the model a means to verify its output. The article presents the Oct. 6 prompt as an example of the first two principles, including a request to iterate until the creator was satisfied.
For a Claude Code task, verification could mean asking the tool to run tests after building an application, the article says. That gives the model a concrete check against the requested outcome rather than relying only on its own assertion that the work is finished. XDA’s piece is an account of Cherny’s advice and an examination of one prompt, not a published benchmark of model performance.
Prompting · Claude Code · Field Notes
How to Improve Claude Code Prompts With Just 3 Things
Boris Cherny’s advice gives coding requests a simple shape: define the task, set the effort, and say how to check the result. One Home Depot project illustrates the approach; it does not prove a performance boost.
“A good prompt comes down to just three things.”
Boris Cherny · Claude Code creator · as quoted by XDA Developers01 / The framework
Three instructions, one clearer request
Cherny’s approach, as described by XDA Developers, focuses on what to make, how much effort to spend, and how to verify completion.
Name the outcome
Describe what the model should create or change. Include the subject, format, and the result you want.
Signal the depth
Say whether the work calls for a quick pass, careful iteration, or a more polished result.
Make success testable
Give the model a practical way to assess its work, such as running relevant tests and reporting what happened.
02 / The example
A creative brief for Home Depot
The prompt accompanied an interactive project about an Acquired podcast episode on Home Depot, according to the XDA Developers account.
Tell the episode’s story
The request asked Claude to narrate the episode as an interactive artifact and highlight its business lessons.
It also suggested watercolor illustrations, historical photographs, and a visual sensibility inspired by The New Yorker and The New York Times.
“Make it beautiful and interactive and focus on communicating the nuggets of business wisdom throughout the episode.”
“Make it really good, use lots of tokens and iterate till you’re proud of it.”
Effort and iteration instruction in the shared prompt03 / The useful check
Ask for evidence of completion
For an application-building task, the article suggests asking Claude Code to run tests and report the result. A check is most useful when it matches the requested outcome.
A passing test suite can reveal specific failures, but it cannot establish that every requirement is met. The check needs to cover what matters for the task.
04 / Reading the evidence
What this example can—and cannot—tell us
XDA examines Cherny’s advice and one prompt. The account does not report a controlled comparison or measured performance gains.
A prompt with concrete direction
The example identifies a subject, an interactive format, business lessons to emphasize, visual references, and a request to iterate.
That the method improves results
No baseline, controlled test, performance measurements, or independent evaluation of the finished artifact is reported.
How the project performed
The supplied account does not detail how long it took, whether extra instructions were used, or whether tests were run on the website.
Match the check to the task
For code, request relevant tests and a clear report. The value depends on whether those tests cover the requirements you care about.
05 / Put it to work
A compact prompt checklist
Before sending a Claude Code request, make each link in the chain explicit.
06 / Key questions
Quick answers
What are the three principles?
Clearly state the task, specify the desired effort, and give the model a way to verify its output.
What was Cherny’s example about?
An interactive artifact narrating an Acquired episode about Home Depot, with an emphasis on business lessons.
Does the article prove the advice works?
No. It presents Cherny’s advice and examines one example, without a controlled comparison or performance measurements.
How can Claude Code check its work?
For an application task, ask it to run relevant tests and report the outcome. Tests help check requirements but cannot guarantee every detail is correct.
Three Instructions for Better Checks
The advice addresses different approaches to prompting: some recommend providing extensive context, while others favor short, specific requests. Cherny’s three-part approach, as described by XDA, focuses on the task, the effort and the check. It offers a way to frame work without requiring a fixed template.
Verification is relevant to coding tasks because a request to run tests can reveal failures that a written account of completed work might not show. The value of a check depends on whether it tests the user’s requirements; passing a test suite does not establish that an application meets every requirement. The article presents the principle but does not provide evidence that a particular verification method guarantees correct results.
AI coding prompt engineering tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Home Depot Prompt Example
Cherny regularly shares Claude Code setup advice on X, according to XDA. The post cited in the article accompanied an interactive project about an episode of the Acquired podcast covering Home Depot. The prompt asked Claude to narrate the episode and communicate its business lessons, while suggesting visual treatments and asking for iteration.
The example includes creative direction as well as a broad objective. It names the intended subject and format, identifies what the work should emphasize, and gives visual references. The XDA writer uses it to illustrate how a prompt can be conversational without being empty of specifics. The article’s supplied account does not establish how the finished project performed, how long it took to make, or whether Cherny used additional instructions not shown in the excerpt.
“A good prompt comes down to just three things.”
— Boris Cherny, Claude Code creator, as quoted in the XDA Developers article
As an affiliate, we earn on qualifying purchases.
What the Example Cannot Establish
The source material does not describe a controlled comparison of prompts, report measured improvements, or specify what baseline Cherny used. It therefore remains unclear whether the three principles reliably improve Claude Code’s results across tasks, or how much each principle contributes on its own.
The article also does not include a complete account of the finished Home Depot artifact, independent evaluation of its accuracy, or details of the verification used on that project. Its suggested test-running example is general advice, not confirmation that tests were run on the interactive website. The effort instruction in the quoted prompt—asking Claude to use many tokens and iterate—also differs from a precise setting or quantified level of computational effort.
As an affiliate, we earn on qualifying purchases.
Apply the Advice to a Task
Readers can use the three principles as a checklist when writing a Claude Code request: name the intended outcome, state how much work or iteration is appropriate, and specify a practical way to check completion. For code, that may include asking the tool to run relevant tests and report their results.
No follow-up test, benchmark or new product announcement is reported in the source. Whether Cherny’s approach produces better results for a particular user will depend on the task and the quality of the checks available.
AI prompt testing and validation tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What are Boris Cherny’s three prompting principles?
As reported by XDA Developers, Cherny recommends clearly stating what the model should do, specifying the desired effort, and giving it a way to verify its output.
What was the example prompt about?
It asked Claude to make an interactive artifact narrating an Acquired podcast episode about Home Depot, with an emphasis on business lessons and suggested visual styles.
Does the article prove these tips improve Claude Code?
No. The article describes Cherny’s advice and examines an example prompt, but provides no controlled comparison or performance measurements.
How can Claude Code check its work?
For an application-building task, a user could ask Claude Code to run tests and report whether they pass. Tests can help check specific requirements but do not guarantee that every aspect of the result is correct.
Source: XDA Developers
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
