Recently, A designer on our team was missing the mark. Despite having access to all the project documentation, her work kept diverging from what the team and client expected. Her questions bounced between overly tactical ("What color should this be?") and frustratingly vague ("Does this work?"). In meetings, she'd present designs that left her teammates confused.

Everything changed when she learned to ask different questions:

• "What problem is this feature really solving for our user?"

• "How does this feature align to the user's journey?"

• "What assumptions am I making about the user's needs?"

The quality of her questions transformed the quality of her work. The principle of asking ‘good’ questions is that simple, and that powerful.

Thoughtful Provocations Versus Shots In The Dark

When stuck, try the "five whys" technique. Keep asking "why?" to dig deeper into any answer you receive.

But, be warned — don’t try this at home (or work)! As it may annoy other humans like a child incessantly asking ‘why’. We must contextually bring information from the surroundings of our area of inquiry.

Don't just ask "What problem is this feature solving for our user?" Instead, do the legwork first. Conduct exploratory potential user research or conduct desk research, then weave those insights into your questions. For example, when you mention a specific user’s behaviors, work deliverables, or pain points you've observed, you give the AI agent rich context to work with. "Given that our users spend 40% of their time reconciling data between spreadsheets, how might this feature reduce that friction of a specific data-set work deliverable?"

This is the difference between thoughtful inquiry and shooting in the dark. Your questions become provocations informed by real insights rather than unbounded shots in the dark.

Why This Matters Now

We're entering an unprecedented era where we collaborate with:

• AI agents that have infinite patience but no memory

• Remote teams across different time zones and cultures

• Complex systems that require deep understanding

• Stakeholders who need clear explanations

The ability to ask good questions isn't just helpful – it's becoming the key differentiator between those who thrive and those who struggle.

Question Makeovers: Real Examples

In Design Reviews

❌ "Do you like this design?"
✅ "How does this design solve our core user problems?"
💬 This question assumes we are capable of leveraging direct user insights from qualitative and quantitative user insights setting a baseline understanding of our customer

With AI Tools

❌ "Make it better"
✅ "What assumptions in this design might we be missing?"
💬 This question assumes we the agent has access to our design files. For example: an AI agent designer and art director directly within Figma able to write comments, create and edit components.

In Team Meetings

❌ "Does anyone have any questions?"
✅ "What risks do you see that we haven't discussed?"
💬 This question assumes our AI agent has access to project plans, roadmap and team room where program tasks are managed.

The Beautiful Paradox of AI Questions

Imagine having access to a brilliant colleague who:

• Never gets tired or frustrated by your string of seemingly repetitive questions

• Has no ego to bruise in asking pointed, critical questions

• Doesn't judge your "obvious" questions as uninformed or off topic

• Is willing to explain things a thousand different ways to clarify a topic

The Power of Questions Amplified

Questions have always been powerful tools that:

• Frame problems by setting the boundaries of what we're trying to solve

• Challenge assumptions by bringing hidden beliefs into the light to examine them

• Create possibilities by opening up new ways of thinking about old problems

• Drill down on topics as insights appear within the provided answers

The New Art of Asking

We’ve moved way past simple prompt and return experiences with early LLMs to contextual conversations with AI agents. Great questions for the AI age have unique characteristics of conversations:

1. They're Iterative

   • Start with broad, exploratory questions that help map the territory. Think of these as your initial reconnaissance.

   • Build thoughtfully on previous answers, using each response as a stepping stone to deeper understanding.

   • Explore multiple angles systematically, like a detective following different leads to solve a case.

   • Circle back to earlier points with new insights, creating a spiral of growing understanding.

2. They're Context-Rich

   • Paint the full picture upfront by providing relevant background that frames your question properly such as documents, articles, your initial hypothesis or outline.

   • Set clear parameters about your goals and any constraints you're working within.

   • Define your desired outcomes explicitly, helping guide the exploration toward practical results.

   • Include relevant history or previous attempts, creating a richer foundation for new insights.

3. They're Fearlessly Curious

   • Embrace the power of asking the same question multiple ways, each time uncovering new layers of insight.

   • Challenge your own assumptions openly, treating them as hypotheses rather than facts.

   • Explore edge cases and boundary conditions that might reveal hidden opportunities or problems.

   • Follow interesting threads without self-censoring, knowing there's no social cost to "obvious" questions.

When Voice Enters the Conversation

The medium reduces or adds friction to the message – and voice interaction is reshaping how we ask questions in fascinating ways:

1. Natural Flow vs. Composed Text

   • Written questions tend to be carefully crafted and edited

   • Voice questions flow more naturally and spontaneously

   • Speaking removes the friction of typing and editing

   • We're more likely to ask quick follow-ups when speaking

2. Context and Continuity

   • Written questions often front-load context

   • Voice allows for more natural back-and-forth

   • Tone and emphasis add layers of meaning

   • Conversation feels more like human dialogue

3. Different Mental Models

   • Writing encourages structured, linear thinking

   • Voice promotes more exploratory, associative questioning

   • Speaking reduces self-consciousness about "perfect" phrasing

   • Natural pauses create space for deeper reflection

I find voice input is more for on-the-go, mobile questions and typing I use on a laptop. These different modalities make voice more about in-context questions about a place or event I am seeing out in the world versus typed questions are more about my learning or critique of an interest area I am researching.

The Future Belongs to the Question-Askers

In a world where AI can provide instant answers, the true value lies in asking the right questions. It's tempting to fall back on vague prompts like "make it better" – and sometimes they even work. But this is like using a calculator without understanding math: you might get results, but you miss the deeper learning and growth that comes from truly engaging with the problem.

Those who master this art will:

• Learn faster by targeting their questions to uncover fundamental principles rather than just surface facts

• Adapt better by developing the ability to quickly frame new situations in ways that reveal key insights and opportunities

• Lead more effectively by using questions to align teams and surface hidden concerns before they become problems

• Solve harder problems by breaking them down into precisely framed questions that build on each other

• Build stronger relationships with both humans and AI by mastering the art of dialogue that moves beyond surface-level interactions

AI agents best trick isn’t doing tasks for us, but rather helping us to quickly understand and the key to unlocking this super power is to ask good questions.

The quality of your questions shapes the quality of your solutions. In an age of instant answers, knowing how to ask is becoming more valuable than knowing the answers.

Next time you find yourself reaching for your AI tool, ask yourself: "What could I learn if I asked this differently?"

Meta's Explore Theory-of-Mind program could serve as a powerful guardrail tool by generating diverse, challenging scenarios that specifically test an AI system's ability to understand and appropriately respond to human mental states and intentions. Its program-guided approach to creating adversarial examples could help identify potential misalignments in AI systems before deployment, as demonstrated by the significant 27-point accuracy improvement on the ToMi benchmark when fine-tuning Llama-3.1 7B - suggesting it could be particularly valuable for evaluating and improving AI systems that need to engage in complex human interactions.

Andrew Ng's presentation at Snowflake Build reveals a crucial paradox in modern AI development: while generative AI has dramatically accelerated prototyping speeds (from months to days), the true challenge and competitive advantage lies in sophisticated data engineering, particularly for unstructured data like images and video that power AI agents. The rise of agent orchestration layers (like LangChain and Landing AI's Vision Agent) is making application development more accessible, but their effectiveness fundamentally depends on well-engineered data pipelines and robust evaluation systems - suggesting that organizations must invest as heavily in their data infrastructure as they do in their AI model development to build effective agents.

Perplexity's acquisition of Carbon AI represents a pivotal shift in AI-powered search, as it transforms from a web-only search engine into a comprehensive knowledge assistant that can seamlessly connect and synthesize information across both public and private data sources (web, Notion, Docs, Slack). This evolution is particularly significant because it addresses a core friction point in knowledge work - the fragmentation of information across multiple platforms - by enabling AI agents to conduct unified searches and synthesize insights across previously siloed data sources, potentially revolutionizing how professionals access and utilize their organization's collective knowledge.

The rapid generation of 3D worlds by AI agents represents a transformative capability that's currently revolutionizing gaming and entertainment through tools like Explorer, World Labs and Google’s Genie 2 which can instantly convert 2D images into navigable 3D scenes without complex capture processes – demonstrating how AI can dramatically reduce the friction in creating immersive digital environments. While these early applications are impressive, the true paradigm shift lies ahead in industrial applications, where this technology could enable the instant creation of digital twins for complex systems, fundamentally changing how we approach data analysis, training simulations, and system optimization across manufacturing, urban planning, and infrastructure management.

Suggestion box

This newsletter is part of our ongoing series exploring the future of specialized AI through a human-centered lens. Subscribe to join a community of designers and developers shaping purposeful AI agents.

Until next time, keep innovating and stay curious!

Jason 

When Voice Enters the Conversation

The medium reduces or adds friction to the message – and voice interaction is reshaping how we ask questions in fascinating ways:

1. Natural Flow vs. Composed Text

   • Written questions tend to be carefully crafted and edited

   • Voice questions flow more naturally and spontaneously

   • Speaking removes the friction of typing and editing

   • We're more likely to ask quick follow-ups when speaking

2. Context and Continuity

   • Written questions often front-load context

   • Voice allows for more natural back-and-forth

   • Tone and emphasis add layers of meaning

   • Conversation feels more like human dialogue

3. Different Mental Models

   • Writing encourages structured, linear thinking

   • Voice promotes more exploratory, associative questioning

   • Speaking reduces self-consciousness about "perfect" phrasing

   • Natural pauses create space for deeper reflection

I find voice input is more for on-the-go, mobile questions and typing I use on a laptop. These different modalities make voice more about in-context questions about a place or event I am seeing out in the world versus typed questions are more about my learning or critique of an interest area I am researching.

The Future Belongs to the Question-Askers

In a world where AI can provide instant answers, the true value lies in asking the right questions. It's tempting to fall back on vague prompts like "make it better" – and sometimes they even work. But this is like using a calculator without understanding math: you might get results, but you miss the deeper learning and growth that comes from truly engaging with the problem.

Those who master this art will:

• Learn faster by targeting their questions to uncover fundamental principles rather than just surface facts

• Adapt better by developing the ability to quickly frame new situations in ways that reveal key insights and opportunities

• Lead more effectively by using questions to align teams and surface hidden concerns before they become problems

• Solve harder problems by breaking them down into precisely framed questions that build on each other

• Build stronger relationships with both humans and AI by mastering the art of dialogue that moves beyond surface-level interactions

AI agents best trick isn’t doing tasks for us, but rather helping us to quickly understand and the key to unlocking this super power is to ask good questions.

The quality of your questions shapes the quality of your solutions. In an age of instant answers, knowing how to ask is becoming more valuable than knowing the answers.

Next time you find yourself reaching for your AI tool, ask yourself: "What could I learn if I asked this differently?"