Earlier this year, Sift launched an internal GenAI enablement program we call AI Works. The goal isn’t to “add AI” everywhere, it’s to build fluency across our teams, experiment thoughtfully and learn where AI can meaningfully improve how we work.
Every week, our teams explore real use cases, test tools, share what works (and what doesn’t) and develop a point of view on how to apply AI responsibly in day-to-day work. It’s part of how we’re building a culture that is curious, practical and grounded in judgment (not hype).
This piece is a small example of the kinds of conversations happening inside Sift. These discussions help us understand when AI accelerates clarity and collaboration, and when a human needs to take the lead.
We’re doing this work intentionally to work smarter (not louder), improve decision-making and focus and to be a stronger partner to the health systems we serve.
Unedited AI
Great intern. Terrible finisher.
You know the move, your favorite LLM spits out a draft in 14 seconds, and you think, "wow, I am a genius of efficiency," and then you drop it directly into Slack, your boss’s inbox, or (worst) LinkedIn.
But here’s the thing, AI doesn’t know your audience. It doesn’t know the politics of your team. It doesn’t know that your CFO is allergic to adjectives (innovative synergy). And it doesn’t know that your partner reads tone like an FBI profiler.
So, your LLM draft? It may look polished and feel helpful, but it’s not actually done.
Before you hit send, ask:
1) Is this true? (Annoying question, I know.)
2) Does it say one thing clearly, or 17 things vaguely?
3) Does it sound like me, or some sort of corporate wellness blog?
ChatGPT calls its outputs “a helpful intern’s first pass,” which is cute (and GPT goes hard into the intern metaphors), but the intern is not allowed to email the board. The part we humans do best is the judgment layer, including context, priorities, subtext and the sense of when to push and when to soften. AI doesn’t have taste (IMO) and it doesn’t know where the real stakes are.
So yes, use the intern, but don’t let the intern run the meeting.
**Harvard Business Review more bluntly called this "AI workslop."
