Tech Corner: Context Engineering is The Real Magic Behind Cut+Dry’s AI Solutions

It seems every month, there’s a new AI term - from “chain of thought” to “prompt engineering” to “vibe coding”.  The latest?  “Context Engineering”, a term recently popularized in Andrei Karpathy’s X post.  

His point was simple but powerful: Large Language Models (LLMs) are incredibly smart, but they become most powerful when you pair them with highly specific, domain-focused context. At Cut+Dry, this principle is fundamental to our success, evident in several of our tools that seamlessly integrate internal data with the general intelligence of LLMs. A couple of our AI tools, Yes Chef and Order Desk, are built on this exact approach—merging internal data with general LLM capabilities to deliver smarter, more trustworthy results.

At Cut+Dry, we help food distributors and manufacturers sell smarter by combining AI with the deep industry knowledge and data that they’ve built over time.  We've learned that generic AI predictions just aren't enough. Restaurants are nuanced. Distributors and manufacturers know their customers intimately—they know what sells during certain seasons, what's popular regionally, how trends shift over time, and what products specific customers are most likely to order.

Our Yes Chef product allows a distributor to generate sales proposals to restaurants by predicting which of the distributor’s products is most likely to be purchased.  We do this not just by accessing that restaurant’s menu, but also by looking at what similar restaurants bought from that distributor.  That’s the strength of mixing Cut+Dry specific context with the general intelligence of an LLM.

Similarly, our Order Desk product automatically creates order drafts from voicemails, emails, text messages, handwritten notes, and images.  While AI is smart enough to transcribe voicemail and to convert handwritten notes to text, combining those outputs with the knowledge of what the customer typically orders allows us to get it right. It’s like knowing that when the customer asks for “10 lbs of apples” they are likely asking for 10 lbs of sliced Granny Smith apples and not 10lbs of whole Fuji apples. 

The art of context engineering is feeding the right context into the LLM to allow it to make the best decision.  Ordering histories, menu details, regional and seasonal buying patterns—all of that is context that can go into making the best decision.

Keeping Humans in the Loop

But even with all this context, it's essential to include feedback from the folks who know best; distributor teams on the front lines. They see firsthand what works and what doesn't. While Yes Chef and Order Desk give our best recommendations, we also allow humans to view and override the recommendations.  And, Yes Chef incorporates their direct feedback, creating an ongoing, human-in-the-loop improvement cycle, to steer us continually toward more accurate predictions.

For instance, we had a distributor client who initially noticed Yes Chef wasn’t accurately predicting specific meat cuts for Taco stands. They provided feedback directly in the tool, and the system immediately integrated this knowledge into future predictions. The results improved significantly, and the next round of predictions was spot on. This directly translated to happier customers and higher sales.

Of course, none of this works if people don't trust us with their data. Privacy and security aren't optional—they're foundational. We go to great lengths to ensure every distributor’s data stays separate, secure, and private. Data isolation and robust privacy measures aren't just checkboxes; they're core to how we operate.

Ultimately, our experience at Cut+Dry highlights exactly what Karpathy meant. AI alone isn't magic. The real magic happens when we combine AI’s power with rich, carefully engineered context and continuous human feedback. That’s how Yes Chef helps our customers succeed—not just with good predictions, but with predictions they can genuinely trust and rely on.

Looking Ahead

We're excited about pushing context engineering even further. Better predictions, happier customers, and smoother operations—that's what we're aiming for. After all, the best AI is AI with the right context.

---

This blog was written by Josh Silver, VP of Engineering and Head of AI at Cut+Dry. At Cut + Dry, he leads the company’s AI strategy and engineering efforts—actively building tools like Yes Chef that combine LLMs with real-world foodservice data. A former Amazon leader and startup founder, Josh loves diving deep into emerging tech and turning it into practical AI products that work at scale.