How to Actually Collaborate With AI (Not Just Use It)

By Jordan Hauge — Published June 17, 2025 — Category: AI Strategy, Product Management, Prompt Engineering

The article highlights that despite the growing AI market, a 2024 Salesforce study showed only 39% of business users frequently get what they need from AI, primarily due to unclear prompts, missing context, and unrealistic expectations. Ultimately, getting valuable AI output depends on the quality of your interaction and communication strategy.

Let’s get one thing straight: prompting AI isn’t about “telling it what to do.” It’s about collaboration and strategic communication. While the global AI market is projected to reach over $1.8 trillion by 2030 (Grand View Research), the effective use of AI in daily workflows is still a learning curve for many.Most people treat AI like a vending machine...Drop in a prompt, hope for a snack.But if you want outputs that are nuanced, on-brand, and actually useful, you need to treat it more like your business partner. Or better yet, a brilliant colleague who knows absolutely nothing about your specific project until you start talking.This seemingly small shift in mindset can 10x the quality of your results and transform AI from a basic tool into an indispensable thought partner.Prompting Isn’t a Quick Input. It’s a Strategic Setup For Strong OutputA lazy prompt is like saying to a contractor, “Build me something good.”Good luck with that. You'd likely get something generic, perhaps even off-target.AI thrives on clarity and context.It's only as smart as the information you give it. You wouldn’t ask a designer to make “a logo” without telling them your brand values, target audience, aesthetic, and goals.The same goes for AI.This concept ties into the idea of "prompt engineering," a rapidly evolving skill set in the AI landscape. It’s about crafting precise instructions to elicit desired responses.Give it:Specific outcomes: (“Summarize this as a tweet, not a blog post, highlighting the main benefit and including relevant hashtags.”)Real-world context: (“Imagine you’re giving this advice to a founder who just raised seed funding for a B2B SaaS startup aiming to disrupt a mature industry…”)Analogies or metaphors it can build on: (“Explain the blockchain like Gordon Ramsay teaching a 5th grader about baking a cake.”)Role-playing: (“Act as a senior marketing strategist for a Gen Z audience. Your task is to brainstorm viral campaign ideas for a sustainable fashion brand.”)The more you teach the model how you think, the more it can meet you there. Think of it as setting the stage for optimal performance.One-Shot Prompts Are Overrated; Try Iterative RefinementHigh-quality output is rarely one-and-done, especially for complex tasks.This isn't a flaw in the AI; it's a feature of effective collaboration.The most effective prompting strategies use “chain-of-thought” interactions and iterative refinement — a fancy way of saying: treat your conversation like a strategy session, or even a mini-design sprint.Start broad.Clarify.Then refine.Example:Prompt 1 (Broad): “Give me 3 frameworks for strategic product prioritization.”Prompt 2 (Refine/Clarify): “Good. Now rewrite the second one (ie. RICE scoring) using a metaphor a 12-year-old would understand, focusing on how to decide which video game to play first.”Prompt 3 (Expand/Refine): “Excellent. Now, add a bullet list of common mistakes people make when only using that framework, and suggest a complementary perspective.”Each step compounds, building on previous context.The result is layered, nuanced, and far more useful than a single, flat request.This approach can significantly improve outcome accuracy, as supported by research from Google AI demonstrating the effectiveness of chain-of-thought prompting in complex reasoning tasks.You Should Learn AI, and AI Should Learn YouWe don’t talk to our mom, our co-founder, and our mechanic the same way. AI shouldn’t be any different.With advanced features like Custom Instructions (OpenAI), memory (some proprietary models), or cross-session persistence (Gemini's growing capabilities), you can train the model to speak your language inclusive of your tone, your values ... even your specific humor.Ask it to remember how you write your internal communications. Give it examples of "a good answer" from your perspective.This isn't just about convenience; it's about creating a personalized AI agent that understands your unique style and needs, vastly improving efficiency over time.That’s not cheating, that’s personalization.That’s what makes it feel less like a generic tool and more like a true thought partner.Pick the Right AI for the JobNot all AI is created equal.Different models have different strengths, much like different specialists in a team. Understanding their core competencies is crucial for maximizing utility.Here's a quick breakdown:🔷 GPT-4 (OpenAI)Strengths: Deep reasoning, extensive context window, strong at complex creative and technical writing, robust code generation, sophisticated product thinking.Weaknesses: Can be overly verbose; slower inference speeds in some applications; can "hallucinate" if context is insufficient.Best For: Strategy sessions, content creation (articles, marketing copy), ideation, structured plans, coding assistance.🟡 Claude (Anthropic)Strengths: Exceptionally long context window (ideal for processing entire books or large datasets), excels at summarizing and debating dense information, known for a "friendly" and less assertive tone, strong ethical guardrails.Weaknesses: Can be overly cautious, sometimes vague in creative outputs; less direct than GPT for certain tasks.Best For: Digesting massive research documents, legal or compliance reviews, empathetic or customer-facing writing, literary analysis. Also excels at engineering tasks and code reviews.🔶 Gemini (Google)Strengths: Strong cross-tool integration (Docs, Sheets, Gmail, etc.), excellent visual reasoning, superior image and video-based tasks, increasingly strong multimodal reasoning across various data types.Weaknesses: Still maturing in very long-form, complex reasoning compared to highly specialized LLMs; performance can vary by tier. Gemini can hallucinate at times or make false promises (ie. I will provide aGoogle Sheet for you and let you know when it is complete, but it never even begins to work on the promised task)Best For: Multimodal work (analyzing charts, generating captions from images), workflows within the Google ecosystem, assistant-style tasks, quick factual lookups, real-time data integration.Using the right tool isn't just about preference; it's the difference between hiring a generalist for brain surgery and a domain expert for a highly specialized task.Mismatching the tool to the task leads to frustration and suboptimal results.The Perception vs. Reality of AI AccuracyA fascinating 2024 Salesforce study revealed that only 39% of business users said AI “frequently gives exactly what they need.” This number is critical. It underscores a significant gap between expectation and reality, often rooted in how users interact with the AI.The biggest reasons cited for this dissatisfaction:Lack of clarity in the initial promptMissing critical context provided by the userUnrealistic expectations of one-shot magic without refinementIn other words: the problem isn't usually the model's inherent capability.It's the message you gave it.As researchers from Stanford have frequently emphasized, the quality of AI output is directly proportional to the quality of the input and the interaction strategy.Final ThoughtsAI is not a genie granting wishes from a single, mystical command. It’s a powerful, evolving collaborator.To unlock its true potential, talk to it like you would a highly capable teammate you trust. Provide it with clear asks, shared context, layered back-and-forth iteration, and a willingness to teach and refine.Let it learn your voice and preferences. And critically, choose the right model for the job, leveraging its unique strengths.You’ll be shocked how much better, how much more nuanced, and how much more you the output becomes.It’s not just about using AI. It’s about mastering the art of collaboration with intelligence.