From: aidotengineer
Many current AI implementations today focus on adding chatbots to products, which often serves as an easy, albeit not always helpful, solution [00:00:00]. Arthur, a product designer at Evil Martians, advocates for challenging conventional thinking about AI in products and adopting practical principles for more effective integration [00:00:19].
The Problem with Reactive AI
The prevailing approach involves creating AI assistants that wait for user input, primarily through chat interfaces [00:01:09]. This means users must initiate interaction and explain their needs, which can be time-consuming and disruptive to their workflow [00:02:20]. Such reactive models often lead to generic “how can I help?” questions that don’t effectively move work forward [00:01:30].
Inspiration from the Past: Clippy
The concept of proactive assistance isn’t new [00:00:48]. The much-maligned Clippy, though often hated for its terrible execution and timing, had the right idea: to anticipate user needs [00:00:55]. With current technology, the execution can now be done effectively [00:01:03].
Introducing Proactive AI
Proactive AI, in contrast to reactive models, is designed to anticipate user needs without being explicitly asked [00:01:11]. It works by understanding the context of the user’s activity and intervening at the precise moment with relevant, contextual questions or suggestions [00:01:44]. This integration occurs seamlessly within the natural flow of work, eliminating the need for context switching or extra windows [00:02:51].
Examples of Proactive AI in Action (Tigon)
The AI issue tracker Tigon demonstrates proactive AI through different modes:
-
Suggestion Mode: This mode tracks what a user is writing in real-time, understands the context, and provides specific, non-generic questions exactly when needed, without a chat window [00:01:41]. For example, if a user reports a “legality share,” the AI immediately asks relevant questions to clarify the issue [00:01:25].
-
Action Mode: This mode allows the AI to identify complexity in a user’s input and suggest better ways to organize work [00:02:06]. For instance, if a user writes an issue that can be split into sub-issues, the AI identifies and suggests splitting it, leveraging previous data and understanding best organizational practices [00:01:58]. This is an example of AI that actually understands [00:02:29].
-
Question Plus Action Mode: This combines the previous two, allowing the AI to not only organize work but also consider timelines and resources, acting like a proactive project manager [00:02:35]. All interactions happen within the user’s workflow, without requiring new interfaces [00:02:51].
Users maintain control, with the ability to easily revert any AI-suggested changes with a single click [00:03:11].
Core Principles of Proactive AI
To foster proactive AI in products, three simple rules are followed:
- AI should supplement user agency, not replace it [00:03:47].
- AI should offer recommendations, never force them [00:03:50].
- AI should be part of the natural workflow, not disrupt it [00:03:56].
Applications Beyond Issue Tracking
The pattern of proactive AI can be powerful across various professional tools:
- Code Editors: AI can proactively monitor for common pitfalls and suggest improvements, which is particularly valuable for developers learning new languages or frameworks [00:04:02].
- Design Tools: Imagine a tool that suggests accessible design improvements as you work, eliminating the need for post-design checks [00:04:12].
- Communication Tools: AI could prepare relevant context before meetings or locate documents mentioned during calls, acting as a personal advisor [00:04:24].
Implementing Proactive AI in Your Products
To start thinking about implementing proactive AI:
- Look for friction points: Identify areas where users have to stop their work to ask for help; these are opportunities for proactive assistance [00:04:41].
- Identify user behavior patterns: Determine where users commonly need help or what questions they frequently ask; these provide clues for automation [00:04:51].
- Consider context: Focus on where users get stuck; this is where AI can provide the most help [00:05:03].
Instead of merely adding chat interfaces, which is a common pitfall in AI strategy, developers should be willing to experiment and challenge the status quo of UI solutions [00:05:20]. AI interface design is still in its early stages, lacking a fully formed playbook of best practices [00:05:13].