🔀 Incoherent: Rethinking Product Development in the AI Era

When building AI-native product development or plugging AI in your product, it requires a mental shift in how we approach software creation. AI-native development means designing with AI as the core foundation, while “plugging in” refers to enhancing existing products with AI capabilities. Both approaches are valid but require different strategies.

Many of us still default to traditional thinking patterns even though AI allows for radically different approaches. The context you feed to AI is crucial — not just as input but also for the outcomes it produces and the dependencies it affects throughout your system.

Two Fundamental Problem-Solving Approaches

When thinking about leveraging AI from scratch in product building, consider that problems can be solved in one of two ways:

• Deterministic approach: Produces consistent, predictable results every time (think banking transactions or medical systems)
• Non-deterministic approach: Produces results that may vary when tried multiple times (think creative content or recommendation systems)

In some industries, consistency is often business-critical. Should you leverage AI in such scenarios? Yes, but only if it adds value without compromising consistency. The scope for AI here typically falls into two categories:

1. Accelerating operational processes that don’t directly impact outcomes
2. Enhancing use-cases around the core functionality — for example, generating real-time insights from transaction data while keeping the transactions themselves deterministic

Embracing Uncertainty Where Appropriate

There’s a second class of problems where a best guess or guesstimate is perfectly acceptable. Here, AI-native approaches truly shine. When designing a mood board, generating initial content drafts, or brainstorming ideas, the non-deterministic nature of AI becomes a feature, not a bug.

We’re also seeing an entire array of products where users haven’t reached their productivity potential because software interfaces don’t “speak their language.” Think of domain experts who need to learn complex UIs rather than expressing their intent naturally. AI can finally bridge this gap, allowing software to understand user intent rather than forcing users to learn rigid interfaces.

The Critical Elements: Prompting and Context

The journey doesn’t stop with identifying where AI fits. While AI can fast-track both paths (informed-guesses or improving human-system interaction), prompting and context are two crucial elements for success.

When designing prompting interfaces, you have three main options:

• Invisible prompts: Hide them behind the scenes, letting users interact with familiar CTAs
• One-shot prompts: Users get one attempt, with new prompts overwriting previous work
• Conversational iteration: Users refine outcomes through continued conversation with persistent context

Finding the right interaction model is essential. Rule of thumb: if users must learn how to craft effective prompts for your product, you’ll likely struggle to generate excitement. One-shot prompting creates limitations for any use-case where iteration would produce better results (though cost is certainly a consideration).

The Power of Context

Context matters not only in what you feed to the AI model but also in how you present its output. Cursor, for example, has implemented this better than early versions of GitHub Copilot. Making code changes in-line is often limiting since users need to track related changes across multiple files or functions — creating cognitive load.

The ideal solution identifies dependencies and shows users all the places that need modification to make a change complete. Think of requesting a feature change and being told, “Here’s the main implementation, and here are three other places that need updates to make it work.” This comprehensive approach delights users and builds trust.

While maximum input and output context with conversational capabilities sounds ideal, it isn’t always the most valuable approach. For simple use cases, it may be overkill or financially unsustainable. Analyze your specific user needs and business constraints before deciding on implementation.

Moving Forward

What other critical elements should we consider when designing software in an age where AI is becoming the default? How do we measure success, balance automation with user control, and create interfaces that feel both magical and trustworthy?

The answers will vary by domain, but one thing is certain: those who thoughtfully integrate AI capabilities rather than simply bolting them on will create the most compelling products in this new era.