Building AI Products That Actually Solve Problems: A Design Framework

Embedding AI capabilities into software or launching entirely new AI-driven offerings demands far more strategic thinking than simply deploying a conversational interface. Yet the mere presence of artificial intelligence does not guarantee value creation. In reality, many organizations pour substantial resources into AI implementations that yield minimal returns.

A curated set of design principles and research findings offers pragmatic methods for evaluating whether AI genuinely improves a product and provides concrete guidance for building AI-powered features based on empirical study.

The technology sector tends to oscillate between two extremes: unbridled optimism about AI's transformative potential and cynical dismissal of it as pure marketing noise. Both perspectives miss the mark. The right approach demands genuine engagement with AI's capabilities while remaining honest about situations where it fails to deliver meaningful impact.

Strategy and Value Proposition

Technology companies have long chased emerging innovations without fully understanding their practical applications. In 2021, before large language models like ChatGPT entered the mainstream, Nielsen Norman Group's former CEO Jakob Nielsen cautioned organizations against overcommitting to nascent technologies without clarity on how they address genuine user problems.

Viewing AI as a specialized tool within a broader product toolkit is reasonable. The fundamental mistake occurs when teams reverse this logic—selecting AI first and then searching for problems it might solve, rather than identifying user challenges and determining whether AI is the appropriate solution.

Organizations achieve stronger outcomes when they emphasize the concrete benefits AI delivers rather than assuming the technology itself generates value. Features built primarily for novelty rarely produce tangible results. Instead, teams should concentrate on addressing genuine user frustrations.

• Resist the urge to adopt chat-based interfaces without validating they meet actual user requirements
• Narrower, more focused AI capabilities tend to be easier for users to comprehend and adopt at higher rates
• Content generation, text condensation, elementary statistical analysis, and alternative viewpoint generation represent AI's core strengths
• Site-specific chatbots lack appeal unless they resolve problems that current product functionality cannot address

Henry Modisett, head of design at Perplexity, has offered insights into effective AI feature design. Emerging generative interfaces could construct personalized layouts tailored to individual user objectives. As AI agents now operate alongside humans within digital environments, organizations must reconsider their definition of "user" and ensure accessibility remains central. Success emerges when product development begins with user challenges rather than with available AI technology.

Prompt Assistance

The effectiveness of language model-based features depends significantly on users' capacity to formulate effective prompts. Although advanced models can produce useful results even from vague instructions, they cannot interpret unspoken thoughts. Additionally, users frequently lack clarity about their own objectives or the full scope of what AI systems can accomplish. Supporting users in their prompting process becomes essential.

• Combining prompt-based inputs with graphical interfaces enhances the usability of image-generation applications
• Thoughtfully crafted prompt suggestions boost user learning and creative exploration while establishing realistic expectations
• Suggestions must align with the specific task, match the user's expertise level, and reflect contextual relevance
• Prompt controls enhance feature visibility, provide education and inspiration, establish boundaries, and enable iterative refinement
• AI conversations can generate straightforward interface components to gather information, reducing user effort and producing more tailored outcomes

Product and Feature-Specific Recommendations

After determining that AI can meaningfully enhance a product, implementation quality becomes paramount. Research-backed observations reveal specific best practices for designing AI-powered offerings effectively.

• Review summaries generated by AI gain user acceptance when they demonstrate specificity and clarity about their origins
• Generative AI text frequently disregards established web-writing conventions and should maintain brevity, visual scanability, hierarchical structure, and accessible language
• Amazon's "Rufus" feature illustrates that even valuable AI capabilities underperform when users fail to discover them
• Workplace AI systems can enhance operational efficiency, facilitate professional development, and customize user experiences
• The sparkle emoji carries inherent ambiguity yet increasingly signals AI functionality
• Trustworthy product-specific chatbots exhibit five characteristics: readiness to escalate to humans, adaptability, initiative, emotional awareness, and openness about limitations

Different organizational roles require distinct explanations for identical AI outputs. Information-architecture methodologies applied to AI systems enable agents to better comprehend data and generate more coherent responses. Analysis of Qwen's AI agent identified four design priorities: enabling feature discoverability, leveraging established interaction patterns, managing sensitive information responsibly, and preserving user control.

Effective product-specific chatbots communicate their functional boundaries transparently, furnish contextually appropriate prompt examples, and quickly convey what information they can access. Users typically consult these chatbots for rapid answers rather than extended dialogue, so responses should be direct, easily scannable, and expandable for deeper information. Explanatory language in AI chat interfaces frequently falls short of helping users comprehend the reasoning behind outputs. AI-generated text must adhere to standard digital writing practices to accommodate how users actually read online.

State of AI-Product Design

The following observations capture recommendations and assessments from particular moments in AI's evolution. While certain findings remain relevant, others have shifted as AI capabilities have expanded and adoption within user experience design has accelerated.

• ChatGPT's agent successfully completed a restaurant booking task, though the execution was sluggish and prone to mistakes (2025)
• Prioritizing user needs surpasses chasing technological novelty; AI may enhance human language comprehension and amplify UX practitioner effectiveness (2021)
• Inflated marketing claims about intelligent assistants risk damaging user adoption if the actual experience disappoints (2019)
• Jakob Nielsen identified a fundamental constraint: AI advancement depends on simultaneous progress in language comprehension and intention recognition (2016)
• As AI-generated content becomes ubiquitous, users increasingly regard visibly human-created work as a trustworthiness indicator

Podcast Episodes

Modisett, Perplexity's design lead, has discussed strategies for developing effective AI features. Paige Lord explores how UX professionals can contribute to building AI products and services that prioritize ethics. Designers developing AI-driven experiences must emphasize both user value and safety considerations. Articulating user intent clearly enables AI systems to create solutions aligned with actual user requirements.