Claude 3 First Impressions for Project Managers

Anthropic released Claude 3 in March with three tiers: Haiku, Sonnet, and Opus. I have been using Opus alongside GPT-4 for the past several weeks, specifically for the kind of work project managers do. Here are my honest impressions.

Where Claude 3 Shines

The standout feature is the context window. Claude 3 Opus handles up to 200K tokens. In practical terms, I can paste an entire project charter, three months of sprint summaries, and a stakeholder map into a single conversation. With GPT-4, I was constantly hitting context limits and having to break conversations into pieces.

For summarization tasks, Claude 3 feels more precise. When I ask it to summarize a long requirements document, it captures nuance better. It is less likely to hallucinate details that are not in the source material. For a PM, that reliability matters.

Where GPT-4 Still Wins

GPT-4 is better at structured output. When I need a table, a specific format, or a template populated with data, GPT-4 follows formatting instructions more reliably. It also has a larger ecosystem of plugins and integrations that fit into existing workflows.

For brainstorming and ideation, I still prefer GPT-4. It is more willing to generate diverse options and push creative boundaries. Claude 3 tends to be more conservative, which is sometimes what you want and sometimes not.

My Recommendation

Use both. I use Claude 3 for analysis and summarization tasks where accuracy matters. I use GPT-4 for drafting, formatting, and creative work. Neither replaces the PM's judgment, but together they cover different parts of the workflow.

The important thing is to start building these tools into your practice now. The PMs who learn to work effectively with AI in 2024 will have a significant advantage over those who wait. The technology is not perfect, but it is good enough to save hours every week.