Tulsee Doshi, Head of Product for Google’s frontier AI models, emphasizes a 60-20-20 framework for shipping AI products: 60% on what works now, 20% on what’s almost working, and 20% on speculative future capabilities. This approach balances immediate value with innovation, crucial for navigating the rapid advancements in AI. The core PM challenge is defining and measuring progress across these distinct time horizons, especially for probabilistic outputs. In your next product planning meeting, define clear success criteria for each segment of the 60-20-20 framework to ensure balanced development. Watch for how other frontier model teams define and track speculative R&D progress.
What happened
Tulsee Doshi, head of product for Google’s frontier AI models like Gemini, outlines her 60-20-20 framework for balancing current capabilities with future innovation. This framework guides product development in rapidly evolving AI landscapes, emphasizing the need to ship products that are not only functional today but also position for future advancements. The approach acknowledges the inherent uncertainty in AI development by allocating resources to proven solutions, near-term improvements, and speculative future technologies.
A scenario you might face
Imagine a Series B startup building an AI-powered content generation tool. Applying Doshi's 60-20-20 framework, the PM must decide how to allocate engineering resources. The 60% might be dedicated to improving the core text generation quality and speed of existing models. The 20% could focus on integrating a new multimodal input feature that shows promising early results but isn't fully optimized. The final 20% might explore emergent agentic capabilities for automated content strategy planning. The tension lies in setting realistic timelines and metrics for each segment, especially the speculative 20%, to avoid over-promising and under-delivering.
What changes in your practice
In your next sprint planning, explicitly categorize upcoming AI features or research spikes according to the 60-20-20 framework. Ensure that tasks are tagged to reflect whether they fall into 'what works now,' 'almost working,' or 'speculative future.' This will provide immediate clarity on resource allocation and risk. Before your next roadmap review, update your roadmap to visually represent these three segments to better communicate the balance of innovation and stability to stakeholders.
Questions for your team
- 01What specific KPIs will we use to measure success for the 'almost working' and 'speculative future' components of our roadmap?
- 02How will we define and track the 'vibe' component of frontier model development, as mentioned by Doshi, and what does that mean for our product evaluation?
- 03What process will we establish for sunsetting or pivoting on 'speculative future' initiatives that do not show sufficient progress within a defined timeframe?
- 04How can we ensure our evaluation strategy for frontier models adequately covers both quantitative performance metrics and qualitative aspects like 'vibe'?