OP #307: Context Engineering

A great video that anyone who is trying to get up to speed on AI can learn from.

Author

Darren Herman
September 16, 2025

Happy Tuesday. Not much wild to report—other than my growing excitement for the NBA season. The Knicks have been making some under-the-radar moves, and I’m curious (and cautiously optimistic) to see how it plays out. Expectations are high, maybe too high, but that’s the beauty of being a Knicks fan: the hope is always bigger than the roster.

On the work front, a lot of people I know seem to be poking their heads up, exploring what’s next. I’m currently recruiting for a CMO/Head of Marketing at 1440 Foods—a New York–based CPG company that’s building what I like to call the P&G of Protein. We’re looking for someone who’s scaled multi-brand businesses from ~$500M to $1B+, and who still wants to roll up their sleeves and build real brands. The role is New York–tri-state based and will sit shoulder-to-shoulder with our new CEO. If someone comes to mind, I’d love to hear about the person. I can then forward their CV to the search firm that’s running the process.

Here are a few quick links to get us started:

Thanks for subscribing to The Operating Partner. If you want to share it with someone, just point them to www.opletter.com. That’s where the recent editions are indexed and searchable—and where folks can sign up for future letters.

Be well, do good.
Darren

10x More Productive

I was listening to Stanford’s Jeremy Utley recently—he teaches innovation and has been diving deep into AI. His framing stuck with me: AI isn’t some perfect guru. It’s more like an overly eager intern (me at i33 back in the day). Smart, fast, willing to help—but lacking boundaries. If you don’t give direction, you’ll get a mess of work back.

Utley calls the solution “context engineering.” Think of it as prompt engineering on steroids. It’s less about writing magic one-liners and more about coaching: being explicit about what you want, how you want it, and why it matters.

He lays out five techniques worth stealing:

  • Chain of thought: ask the AI to show its work, step by step.

  • Few-shot prompting: feed it examples—good ones and bad ones.

  • Reverse prompting: let the AI ask you clarifying questions.

  • Role assignment: tell it who to be (lawyer, designer, skeptical board member).

  • Roleplay: practice tough conversations before you have them live.

The bigger point? AI can’t read your mind. If you want quality output, you have to make the implicit explicit: share your brand guidelines, product specs, tone of voice, customer journey maps. The limitation isn’t the tech. It’s our imagination in how we choose to use it.

If you have 20 or so minutes, this is a great video to watch. Even though you may be an AI-expert, the video still has some goodies for everyone and I bet any subscriber of the OP can learn a thing or so.

Below are a few articles I came across this past week that I found interesting. While I may not agree with everything in each one, I think they're worth a read. If you stumble upon an article you think I or the Operating Partner community would enjoy, feel free to share it with me. Of course, I reserve the right to decide what gets featured in the OP.

Why Slow Is the New Fast (Fast Company)

Ephemeral Software (Tomasz Tunguz)

Thank you for reading (and subscribing to) the OP!

What did you think of today's OP letter?

Login or Subscribe to participate in polls.