TL;DR
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
Claude Code creator Boris Cherny says effective prompts should explain the task, state how much effort it warrants and give the AI a way to verify its work. His conversational prompt for an interactive Home Depot project illustrates the approach, though the source does not report a controlled test of the advice.
Boris Cherny, creator of Claude Code, says a useful AI prompt needs three things: a clear task, guidance on how much effort the model should spend and a way to check its output. The advice followed attention to a conversational prompt for an interactive project about Home Depot, offering users a practical alternative to long, rigid prompting templates.
Cherny shared the project on X on Oct. 6, 2026, alongside an interactive website he said he had built using Claude. The project accompanied an episode of the Acquired podcast about Home Depot. The prompt attracted attention because it read more like a request to a coworker than a formal instruction set, according to XDA Developers’ account of the post and responses.
In the prompt, Cherny asked Claude to make an artifact that narrates the episode and communicates its business lessons. He requested an interactive, polished result, suggested watercolor illustrations and historical photos, and offered visual references such as New Yorker or New York Times data visualizations. He also asked the model to use many tokens and iterate until it was satisfied with the result.
Cherny’s first principle is to state what the model should do and give enough information about the desired result to reduce guesswork. His second is to specify the effort appropriate to the job: a simple request may not need extensive reasoning, while a complex task may benefit from more. His third is to provide a verification method. For a coding task, that could mean asking Claude Code to run tests rather than relying only on its assurance that the work is complete.
PROMPTING / FIELD GUIDE · OCTOBER 2026
Better Prompts Need Just 3 Things
Claude Code creator Boris Cherny’s advice: define the task, set the effort and give the AI a way to check its work. His Home Depot project shows the approach in action—while leaving its effectiveness untested.
Three inputs for a stronger request
The advice focuses on decisions that travel across tasks: what you want, how much work it deserves and what would count as a check.
State the task
Describe the intended outcome and include enough context to reduce guesswork. Say what a satisfactory result should contain.
GOAL + CONTEXTSet the effort
Match the depth to the job. A simple request may need a quick response; a complex one may warrant deeper work and iteration.
SCOPE + DEPTHDefine a check
Give the model a concrete way to review completion, such as running relevant tests or comparing the result with stated criteria.
EVIDENCE + REVIEWA conversational brief for Home Depot
Cherny shared an interactive companion to an Acquired podcast episode about Home Depot, built with Claude. XDA Developers reported on the post on Oct. 22, 2026.
“Make it beautiful and interactive and focus on communicating the nuggets of business wisdom throughout the episode.”
The larger request asked for an artifact that narrates the episode, suggested watercolor illustrations and historical photos, and named New Yorker or New York Times data visualizations as visual references.
— BORIS CHERNY, AS QUOTED BY XDATask
Create an interactive artifact that tells the episode’s story and business lessons.
Effort
“Use lots of tokens” and iterate until the result feels finished.
Visual direction
Use polished visuals, including watercolor illustrations and historical imagery.
Verification gap
The quoted prompt does not name a specific test or independent review.
An illustration, not a measured result
The source describes Cherny’s advice and project. It does not report a controlled comparison or an independent evaluation of the finished site.
No benchmark reported
No success rate or comparison shows how much the three instructions change output quality.
Iteration is not verification
Asking a model to refine its work does not independently confirm accuracy or completeness.
Project details are absent
The report does not say how the site performed or whether its images and episode summary were checked.
Results may vary
Tools, models and tasks differ. Users may still need tighter constraints and their own review.
Turn the advice into a usable request
Make the expected outcome visible, calibrate the work and name a check suited to the task. A check helps catch issues; it is not a guarantee.
Name the deliverable
Describe what you want produced, who it is for and which requirements matter most.
Calibrate the work
Say whether you want a quick answer, careful analysis or several rounds of refinement.
Ask for evidence
For code, request relevant tests. For other work, request a review against your criteria and inspect important claims.
Three Inputs for Better AI Prompts
Cherny’s advice matters because people often receive conflicting instructions about prompting: provide extensive context, keep requests short, or use a particular framework. His three-part guidance focuses instead on decisions that apply across many tasks: define the goal, set the expected effort and check the result.
The verification step is especially relevant when AI output has consequences beyond a conversation. A generated application can appear complete while still containing errors; asking the tool to run tests gives it a concrete check, although passing tests do not prove that every feature is correct. For writing, research or design work, users can similarly define what would count as a satisfactory result and ask for a review against those criteria.
The approach is guidance from a tool creator, not evidence from a reported experiment. It may give users a useful starting point, but the source material does not establish that these three instructions reliably improve results across models, tasks or users.
As an affiliate, we earn on qualifying purchases.
The Home Depot Prompt Behind the Advice
The advice emerged from a specific example rather than a newly announced Claude feature or formal prompting study. Cherny, who regularly posts setup tips for Claude Code, shared a Claude-made interactive companion to an Acquired episode about Home Depot. The simple, conversational wording of the prompt prompted discussion, according to the XDA report.
The example combines a broad goal with several concrete preferences. It identifies the subject and purpose of the artifact, asks for interactivity, names the business insights to emphasize and suggests visual treatments. It also asks Claude to iterate. That makes it an illustration of Cherny’s first principle, while the request for substantial effort and repeated refinement touches on the second. The prompt does not specify a particular test or other independent check, so it does not show all three principles equally clearly.
Claude Code is Anthropic’s coding tool, but Cherny’s three points are framed as general prompting advice. XDA published its account on Oct. 22, 2026. The supplied material describes the original post and the author’s interpretation; it does not provide independent evaluation of the finished website.
“A good prompt comes down to just three things.”
— Boris Cherny, creator of Claude Code, as described in the XDA report
As an affiliate, we earn on qualifying purchases.
What the Example Cannot Establish
The supplied report does not say whether Cherny’s three principles were tested against alternative prompts or measured across different tasks. It also provides no benchmark, success rate or comparison showing how much the advice changes output quality. The principles are presented as practical guidance, not as a guarantee.
Some details of the finished Home Depot project are also absent from the source material. It does not report how the website performed, whether the historical images were verified, whether the episode was represented accurately or what checks were applied to the completed artifact. The prompt asks Claude to iterate, but iteration alone is not an independent verification process.
It is also unclear how consistently the advice applies to different AI systems. Models and tools may respond differently to requests about effort, and a task that is easy to describe may still require more specific constraints. Readers should treat Cherny’s example as an illustration, not a controlled demonstration that one prompt style works best for every use.
As an affiliate, we earn on qualifying purchases.
How Users Can Apply the Three Steps
For users trying Cherny’s approach, the immediate next step is to make each part of a request explicit: describe the intended outcome, say whether the task merits a quick response or deeper work, and name a way to check completion. For coding, that could include running relevant tests; for other tasks, it could mean checking the response against stated requirements.
The source material does not identify a planned follow-up study, product change or further announcement from Cherny. Whether the approach produces better results in a given case will depend on the task and the quality of the check. Users may need to refine their instructions or inspect the output themselves, particularly when accuracy matters.
As an affiliate, we earn on qualifying purchases.
Key Questions
What are Boris Cherny’s three prompting tips?
He recommends clearly stating the task, specifying how much effort the AI should spend and giving it a way to verify its work.
What was Cherny’s Home Depot prompt for?
He asked Claude to create an interactive artifact narrating an Acquired podcast episode about Home Depot, with an emphasis on business lessons and visual elements such as watercolor illustrations and historical photos.
Does the Home Depot example prove the advice works?
No. The report describes Cherny’s prompt and project but gives no controlled comparison, benchmark or measured results. It is an example of his approach, not proof that it improves outputs in every case.
What does verification mean when prompting an AI?
It means giving the model a concrete way to check its output. For a coding task, a user might ask Claude Code to run tests. Test results can help identify problems, but they do not by themselves confirm that all requirements are met.
Source: XDA Developers
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
