Boris Cherny, the creator and head of Claude Code at Anthropic, recently said something that runs against most assumptions about working with AI: Claude Code prompt engineering just isn’t as important as people think.
In a conversation with Y Combinator’s Diana Hu, Cherny explained that the model has become capable enough that meticulous prompt wording matters far less than giving it a genuinely hard task and a way to check its own work. This blog breaks down what he said and what Claude Code prompt engineering actually looks like today.
What Boris Cherny Said About Claude Code Prompt Engineering
Cherny’s central point is that Claude Code prompt engineering had its moment, much like the short-lived rise of “context engineering” before it, and both are already fading as the model’s raw capability grows. He described these as waves that come and go, driven by how quickly the underlying technology improves rather than any lasting methodology.
To illustrate this, he shared a story that went viral internally at Anthropic:
- An engineer gave Claude access to OpenCV, the open-source computer vision library, and simply asked it to draw images.
- Nobody expected that to work well, since drawing wasn’t something anyone had unlocked through careful Claude Code prompt engineering.
- It worked anyway, catching the team off guard and reinforcing Cherny’s point that hidden capability, not clever phrasing, is where the real upside lives.
Diana Hu specifically asked about model elicitation, the research area focused on discovering what a model can already do rather than assuming its limits in advance. Cherny’s answer suggested that elicitation isn’t a specialized skill reserved for researchers. Anyone practicing Claude Code prompt engineering can apply the same principle by simply handing Claude bigger, more ambitious problems than they’d normally attempt.
The Verification Skill Behind Claude Code Prompt Engineering
The part of the interview that stood out most was Cherny’s claim that verification, not prompting, is where most people fall short. He said the real skill today is less about Claude Code prompt engineering and more about handing Claude a task that feels a bit too hard, then giving it a reliable way to confirm its own progress.
A Real Example of Claude Code Prompt Engineering in Action
To demonstrate this, Cherny described giving Claude Tag, Claude working as an agent inside Slack, access to a Mac virtual machine and a genuinely difficult assignment. He asked it to rewrite the entire Claude desktop app from the Electron framework into native Swift, a task that typically takes a development team weeks.
Rather than relying on detailed Claude Code prompt engineering, his instruction was short:
- Run the Electron version
- Screenshot it
- Compare it pixel by pixel against the Swift rebuild
- Keep going until finished
That built-in comparison step did the heavy lifting, and it changes how agencies should think about Website Development projects that lean on AI-assisted coding. Automated verification, rather than longer and more detailed prompts, tends to produce more reliable results across large codebases, which is the same lesson at the heart of good Claude Code prompt engineering.
Why Verification Beats Longer Claude Code Prompt Engineering
Cherny noted that fancy tools weren’t the reason the Swift rewrite succeeded:
- No elaborate slash commands
- No multi-step planning frameworks
- Just a clear goal and a way to measure success against it
Claude even took the initiative to document its own progress in a Slack channel, posting screenshots every few minutes without being asked. That kind of self-directed behavior is exactly what Cherny means when he says Claude Code prompt engineering matters less than designing a task Claude can verify on its own.
Applying Claude Code Prompt Engineering to Client Projects and Search Rankings
Cherny’s argument has practical implications well beyond software teams, and it extends into how agencies approach client work more broadly. Agencies handling Website Development services in Udaipur and similar markets are already testing how far AI tools can go on real client builds, and the Claude Code prompt engineering lesson still applies: a well-defined build task with a verification step tends to beat obsessing over the exact wording of the request.
The same logic extends to content and search work. Teams offering SEO services are finding that Claude Code prompt engineering matters less when the task includes a clear way to check output, such as running a compliance script against a drafted page, rather than trying to perfect the initial instruction.
Claude Code Prompt Engineering vs Traditional Prompting
The table below summarizes how Cherny’s approach to Claude Code prompt engineering differs from the traditional, heavily-specified style many experienced professionals default to.
| Approach | Traditional Prompt Engineering | Cherny’s Claude Code Prompt Engineering |
| Instruction style | Long, highly detailed, step-by-step | Short, goal-focused |
| Core skill | Wording and phrasing precision | Task difficulty and verification design |
| Handles ambiguity | Poorly, needs constant re-prompting | Well, adapts and self-corrects |
| Best suited for | Narrow, repetitive tasks | Open-ended, complex projects |
| Failure mode | Model follows instructions too literally | Model may need better verification, not better wording |
This comparison shows why Cherny keeps returning to the same point about Claude Code prompt engineering: older models needed precise wording, but newer models reward clarity of goal and a working feedback loop far more.
What Claude Code Prompt Engineering Means for Marketing Teams
Cherny’s advice about giving Claude harder tasks with built-in verification translates directly into how marketing teams can put Claude Code prompt engineering to work day to day. Instead of writing increasingly specific prompts, teams get better mileage when they define a clear outcome and a way to check it, whether that’s campaign performance against a benchmark or content matching a brand’s compliance rules.
Running Campaigns With Less Claude Code Prompt Engineering
Agencies running performance marketing campaigns can apply the same principle Cherny described with the Swift rewrite. Rather than crafting an exhaustive prompt for every ad copy variation, a Claude Code prompt engineering approach that hands the model a target metric and a way to test against it, such as click-through benchmarks from past campaigns, tends to produce sharper results with less manual back-and-forth.
Producing Content With Simpler Claude Code Prompt Engineering
The same idea holds for social media marketing, where content volume and consistency matter more than any single perfectly-worded prompt. Claude Code prompt engineering works best here when it’s paired with a review step, like checking drafts against brand voice guidelines, rather than trying to front-load every stylistic detail into the initial instruction.
Providers of social media marketing services in Udaipur are seeing similar gains from this shift toward Claude Code prompt engineering built around verification. For context on how quickly AI adoption is moving across marketing functions generally, HubSpot’s 2026 marketing report found that a majority of organizations are already using AI tools in some part of their workflow, a trend that lines up with Cherny’s point about giving these tools bigger, less rigidly-defined tasks.
Choosing an Agency That Understands Claude Code Prompt Engineering
Cherny closed the interview with advice on Claude Code prompt engineering that applies just as well outside of software: stop chasing the “one weird trick” that influencers claim will unlock better output.
His recommendation is to learn empirically:
- Give Claude something hard
- Watch where it struggles
- Fix the gap through better prompting, a custom skill, or added context rather than endless rewording
An SEO services in Udaipur provider that follows this same empirical habit, testing and verifying rather than guessing at the perfect prompt, tends to build more durable content workflows over time.
Businesses evaluating a digital partner should look for the same empirical mindset Cherny described when it comes to Claude Code prompt engineering. Agencies like Digital Locus that treat Claude Code prompt engineering as one part of a larger workflow, rather than the whole solution, tend to deliver more consistent results because they build in verification at every stage instead of relying on a single perfectly-worded instruction.
For a deeper technical look at how Claude Code actually works, Anthropic’s official documentation covers installation, configuration, and the underlying architecture behind good Claude Code prompt engineering. You can also read Search Engine Journal’s full coverage of the original conversation with Diana Hu for the complete transcript and video.
A performance marketing services in Udaipur team evaluating a new agency partner should ask specifically how that agency approaches Claude Code prompt engineering, since the answer usually reveals whether they are chasing prompt tricks or building real verification into their process.
Conclusion: The Future of Claude Code Prompt Engineering
Boris Cherny’s message marks a genuine shift in how professionals should think about Claude Code prompt engineering. It still has a role, but it’s a smaller one than most people assume.
The bigger win comes from handing Claude a task that stretches its real capability and building in a way to verify the result, whether that’s a pixel-by-pixel comparison, a compliance check, or a performance benchmark. Businesses that adapt this mindset early, treating Claude more like a capable coworker than a search bar to be phrased perfectly, are likely to get more value out of every project they hand it, and Digital Locus builds this exact approach into how it handles client work.
FAQs
Is prompt engineering actually important?
Prompt engineering is still useful, but its importance is changing as AI models become more capable. Boris Cherny, the creator of Claude Code, argues that users often overthink prompts and give models overly detailed, step-by-step instructions. Instead, users can focus on clearly explaining the desired outcome, setting boundaries, and letting the model determine the best approach. In other words, good communication matters, but complex prompt formulas are becoming less essential.
Why is Claude Code so good at coding?
Claude Code is effective because it works more like an autonomous coding agent than a simple chatbot. It can:
- Understand a codebase
- Use tools
- Modify files
- Run commands
- Test its work
- Iterate based on the results
This allows it to handle larger tasks with less back-and-forth from the developer. Anthropic’s research also shows that Claude Code sessions often involve substantially more delegation than traditional chat-based workflows.
How do Claude engineers prompt Claude?
Claude engineers increasingly focus less on writing elaborate prompts and more on creating systems that allow Claude to work autonomously. Boris Cherny has said his role has shifted toward building “loops” that prompt Claude and determine what to do next. For everyday users, the practical approach is simpler: describe the task, explain the desired outcome, establish important constraints, and allow Claude to decide how to execute the work rather than micromanaging every step.
Can I use Claude Code without coding knowledge?
Yes, you can use Claude Code without being an experienced programmer, especially for simpler tasks such as explaining code, creating small scripts, troubleshooting errors, or making basic changes. However, coding knowledge becomes increasingly valuable as tasks become more complex. You still need to understand what you want built, review the output, and recognize when something is incorrect or unsafe. Claude Code can handle much of the implementation, but human judgment remains important when working on real software projects.
