Startup wisdom is a new TNW series offering practical lessons from experts who’ve helped build great companies. This week, Dainius Kavoliūnas, head of no-code platform Hostinger Horizons, shares his tips on vibe coding.
Vibe coding has become an indispensable tool, especially for entrepreneurial thinkers building apps and platforms for solving everyday problems, streamlining business processes, or enhancing digital experiences. It represents a paradigm shift in software development. Instead of writing lines of code, you can now describe your requirement and have AI bring it to life. Vibe coding is fast, intuitive and opens up a new realm of possibilities where code isn’t just functional, but expressive — bridging the gap between people’s ideas and machine abilities.
But like any new tool, getting good results takes some practice. Based on insights from the Hostinger Horizons team and users, here are five practical ways to prompt AI more effectively.
1. Keep prompts short and specific
Though vibe coding might seem improvised and freeform, it is still helpful to follow some basic rules to effectively communicate with AI. My foremost advice for prompt engineering is to be concise. Short, clear, and specific prompts consistently yield better results than vague or broad instructions, and being too general is one of the most common rookie mistakes.
For example, a general command like “Build a website” may produce generic output, while a detailed prompt such as “Create a one-page website for a freelance designer with sections for portfolio, services, and contact form” delivers a more relevant, tailored result.
The inclusion of layout specifications, feature lists, and stylistic preferences further refines output quality. Precision in prompting also minimises ambiguity and reduces the need for excessive revisions later in the workflow.
2. Break down prompts into smaller steps
Being specific doesn’t mean overloading the prompt and packing all the instructions into a single command. Testing the approach shows that it’s more effective to break development tasks into a sequence of smaller, interrelated prompts than issuing a single, complex instruction.
This stepwise approach allows the AI to focus on specific components, which leads to cleaner code and fewer logical errors. For instance, one prompt might focus on generating the structure for a landing page, while subsequent prompts add styling or functionality. This modular process improves maintainability and makes it easier to identify and correct issues early.
3. Encourage AI to propose improvements
The best prompt engineering practices involve using AI not only as a passive output generator but also as an active participant in the development process. Beyond responding to direct commands, generative AI tools can be used as creative collaborators.
Prompting the AI to critique, optimise, or propose alternative solutions can uncover more efficient methods or inspire new design directions. For instance, you can ask the AI to suggest features, content, or sections.
4. Test frequently
One thing common between traditional and vibe coding is that timely detection and early fixes of errors can save a lot of time later and help build functional applications more quickly. Therefore, I advise testing AI-created projects early and often, using realistic data or practical scenarios. That allows for rapid feedback and continuous refinement.
When you ask AI to add a feature — like sending an email notification to someone who fills out a contact form on your website — you should always test that it works as expected. In most cases, the functionality will be implemented correctly, but you may still need to refine the user interface. Occasionally, the feature may not work at all, and you’ll need to give the AI clearer instructions or additional context to get it right.
5. Maintain human oversight
While AI accelerates development, human judgment remains essential, especially for solving complex problems, ensuring long-term maintainability, and verifying the quality of the final product. If an AI-generated result is unsatisfactory, revising the prompt or approaching the problem from a new angle often leads to improvement.
It’s also advisable to preserve working versions before making significant changes. This provides a fallback point and lets you experiment without losing previous progress. Sometimes, some technical expertise can also be needed — consultation with an experienced developer can help resolve deeper architectural or logic-related challenges.
No team, no skills — no problem
Based on feedback from Hostinger Horizons users, these strategies — clear prompting, modular thinking, collaborative refinement, consistent testing, and human oversight — form the backbone of effective AI-assisted coding. As this space evolves, mastering prompt engineering will become a core competency for anyone working with generative development tools.
Take, for example, Prashant Maurya, a 20-year-old student. Without any coding skills and just intuition empowered by AI, he built a set of tools under his own X 247 Project, including an Ads Checker, QR Generator, and Shop Biller. Each took only five to 15 days to build at a fraction of the cost that he would have otherwise incurred if he had gone the traditional way. Over 1,000 users have already tried his tools, built by simply prompting AI.
“As a student with no team, no budget, and just a vision to solve real problems, AI coding tools transformed how I build and launch apps,” Prashant told me.
“The availability of free deployment tools and AI-powered assistance made it possible to create multiple powerful platforms quickly and affordably, and now I’m inspired to help others do the same.”
The power of AI-powered coding tools doesn’t lie in productivity, but in accessibility. For people like Prashant, vibe coding is breaking down barriers, enabling anyone with a vision to build powerful apps quickly and affordably. But this democratisation hinges on one key skill: writing effective prompts. As building becomes easier, standing out through quality and functionality becomes the new challenge, and it all starts with how you prompt.
