Product Trade-Offs When Building AI Features and Assistants

Always-On vs. User-Initiated Assistance, Defining the Scope of the AI Assistant, Proprietary vs. Open Source Models

An important part of developing Product Strategy is identifying, evaluating and making decisions on product trade-offs - mutually exclusive options that need to be specified in order to provide good guidance for Engineering and UX to build a cohesive and compelling product.

Product trade-offs tell you what you WILL focus on and what what you WILL NOT build. The latter is the opportunity cost and it needs to be articulated, together with the rationale for the decision.

Being specific about product trade-offs contributes a lot of value to a PRD and is an integral part of Product Strategy. If you are not forced to make trade-offs, the product will be too generic and not differentiated enough. It will blend in with other offerings in the same category and will not be more compelling to any target audience.

For Product Managers tasked with AI-powered features, defining specifications for this new user interaction paradigm that is built on top of an entirely new tech stack comes with a new set of questions that need to be answered and trade-offs to be considered.

They range from user experience specifications, such as how to signal to the user that there are AI features available and what their scope is, to technical architecture considerations, such as whether to use a proprietary, licenseable model or an open source one.

This post looks at three essential product trade-offs when building AI applications, their advantages and disadvantages that apply across both Consumer and Enterpise-focused implementations. An important note is that these are not binary, ON or OFF options, but rather ends on a spectrum, and each AI assistant can be built at different levels on each of them, depending on the product and the market context. This includes the model selection trade-off between proprietary, licenseable and open source models, as not all open models are fully open source.

Read below about:

• Balancing the AI presence: always-on vs. user-initiated assistance

• Defining the scope of the AI Assistant: broad vs. use case-specific

• Selecting the AI model: proprietary vs. open source