Gemini 1.5 Pro and the Multimodal PM Toolkit

Most of my AI workflow involves Claude 3.5 Sonnet for deep reasoning and GPT-4o for quick tasks. But Gemini 1.5 Pro has carved out a specific niche in my toolkit that neither of the others fills well: processing large documents.

The million-token context window means I can feed it an entire project charter, requirements document, and meeting transcript simultaneously and ask it to identify inconsistencies. That is genuinely useful.

My Use Cases

Meeting transcript analysis. I record stakeholder calls and run the transcripts through Gemini. "Summarize the key decisions, open questions, and action items from this transcript." The quality is surprisingly good, and the long context means it can handle hour-long meetings without truncation.

Document comparison. "Compare this SOW with this requirements document and flag any scope items that are in one but not the other." This used to take me an hour of careful reading. Gemini handles it in under a minute with reasonable accuracy. I still verify the output, but it is a much better starting point.

Sprint retrospective synthesis. I maintain retro notes across sprints. Feeding six months of retro notes into Gemini and asking "what themes keep recurring that we have not resolved" gives me patterns I might miss when reviewing one retro at a time.

The Limitations

Gemini's reasoning on complex multi-step problems is not as strong as what I get from o1-preview or Claude 3.5 Sonnet. For analysis that requires deep logical reasoning — budget trade-off scenarios, risk cascade modeling — I use other models.

It also has a tendency toward verbosity. I have learned to add "be concise, use bullet points" to every prompt.

The Bigger Picture

The PM toolkit in late 2024 is genuinely multi-model. Different models for different tasks, just like different tools in a toolbox. PMs who learn to match the tool to the task will have a meaningful productivity advantage over those who stick with one model for everything.