Just watched this video and wanted to share some key insights that stood out to me:
The smartest models won’t matter if they lack your business context. The real value comes from context engineering - connecting your data and knowledge systems to LLMs through MCP and similar mechanisms.
Their approach to content retrieval considers recency and authorship seniority (social graph) to ensure the most relevant information answers your queries.
Interesting to learn that OpenAI does quarterly planning too, but they emphasize direct customer conversations and go-to-market team input when deciding what to build.
I appreciated hearing that the “canvas” product originated from an individual contributor who pitched the idea and rallied others around it. I’m sure there was a lot more involved in taking initiative and driving it forward from within the organization, but it was still good to hear.
As someone in Trust & Safety, I was glad to hear their perspective that moving quickly doesn’t mean cutting corners on safety. They’re focused on shipping quickly AND responsibly - these aren’t competing priorities but both contribute to maintaining high quality standards. Does that always happen in practice? I’d hope so.
At 30:40, they discuss the emergence of internal “AI champions” within companies. These aren’t people pushing adoption for adoption’s sake, but individuals driving bottom-up cultural change to help others develop AI and tool fluency. The focus is on augmenting and extending capabilities, not replacing them.
The future of work seems to be heading toward building shareable custom GPTs and AI workflows that tap into your company’s institutional knowledge.
Check out what Peter‘s video on what it looks like for a company to leverage AI-first:
This post was originally on LinkedIn.