First Impressions of Claude 3.5 Sonnet for PM Work

Claude 3.5 Sonnet dropped in June 2024, and I've been running it hard for the past few weeks alongside GPT-4o. As a PM who writes a lot — PRDs, stakeholder updates, post-mortems — I wanted to see if it could genuinely save me time.

Where it shines

Technical documentation. I fed it our API specs and asked it to draft integration guides. The output was surprisingly coherent and required maybe 20% editing. That's a meaningful time save on docs I used to spend hours on.

Summarizing meeting notes. I dump raw meeting transcripts and get structured summaries with action items. It catches commitments I missed. This alone justifies the subscription.

Rewriting for audience. Taking a technical post-mortem and rewriting it for executive stakeholders used to take me 45 minutes. Now it takes 10 minutes of prompting plus 5 minutes of editing.

Where it falls short

Strategic thinking. It can't tell you what to build next. It's a writing accelerator, not a product strategist. PMs who expect AI to replace judgment are going to ship bad products.

Context windows matter but aren't magic. Even with the larger context, it loses the thread on complex multi-document analysis. I still need to be deliberate about what I feed it.

Hallucination is still real. I caught it inventing API endpoints that don't exist when writing integration docs. Trust but verify, always.

My current workflow

I use Claude 3.5 Sonnet for first drafts of everything written. PRDs, emails to leadership, retrospective summaries, onboarding docs. Then I edit aggressively. The AI writes faster than me, but my judgment about what matters is still the irreplaceable part.

The PMs who will thrive aren't the ones ignoring AI or the ones outsourcing thinking to it. They're the ones using it as a force multiplier for the tedious parts.