Burnt Puts AI Agents Inside Food Distribution ERPs and Claims 97% of Orders Run Without Humans

The Macro: Food Distribution Is a $1.5 Trillion Industry Running on Fax Machines

I want you to think about the last meal you ate at a restaurant. The chicken came from a distributor. The vegetables came from a different distributor. The cooking oil, the napkins, the to-go containers, all of them came from distributors. The food distribution industry in the United States alone is worth over $1.5 trillion annually, and it is one of the most operationally complex verticals in existence.

It is also, somehow, still running on fax machines and phone calls.

I am not exaggerating. Walk into a mid-size food distributor and you will find order intake processes that look like this: a restaurant manager calls in at 4 AM. Someone at the distribution center writes the order on a notepad. That order gets manually entered into an ERP system that was built in the early 2000s. If the call gets garbled, if the handwriting is bad, if the data entry person misreads “40 cases” as “4 cases,” nobody catches it until the truck shows up at the restaurant with the wrong delivery.

The ERP systems in this industry are ancient and deeply entrenched. Produce Pro, Edible Software, Seasoft, BlueCart. These are not systems you rip out and replace. They are wired into every operational process from purchasing to invoicing. The switching cost is measured in years and millions of dollars. So distributors keep using them, patching around the edges, and accepting a level of operational friction that would be unthinkable in most other industries.

Sysco and US Foods, the two giants, have invested heavily in modernizing their tech stacks. But the industry has roughly 17,000 food distributors in the US alone, and the vast majority of them are small to mid-size operations running on thin margins with no internal technology teams. They cannot afford a digital transformation. They need their existing systems to work better without replacing them.

This is exactly the kind of problem that AI agents are well-suited to solve. Not replacing the ERP. Not asking the distributor to learn a new system. Just sitting inside the existing workflow and handling the repetitive, error-prone tasks that humans currently do manually.

The Micro: Fourth-Generation Food Family Builds the Agents Their Industry Needs

Burnt deploys AI agents into food distribution ERPs. The flagship agent, Ozai, processes incoming orders from every channel a food distributor deals with. Email, phone, text, WhatsApp, and yes, fax. The agent reads the order, interprets it in context, and enters it directly into the distributor’s existing ERP system. No system replacement. No migration. No retraining staff on new software.

The founding team is the story here. Joseph Jacob is CEO, and his family has eighty-five years in the seafood distribution industry. Rhea Karimpanal is Chief Product Officer and grew up around restaurants. Chandru Shanmugasundaram is CTO and has built backend systems across legal tech, banking, and restaurant operations. They are out of San Francisco and came through Y Combinator’s Summer 2025 batch.

This is founder-market fit in its purest form. These are not Silicon Valley outsiders who read a McKinsey report about food distribution inefficiency and decided to build a startup. They grew up in this industry. They know why a 4 AM phone order from a chef who is half asleep and shouting over kitchen noise is hard to process. They know why distributor ERPs look the way they do and why nobody is going to replace them.

The 97% end-to-end automation claim is bold. That means 97 out of 100 orders flow from intake to ERP entry without a human intervening. The remaining 3% presumably get flagged for manual review. If that number is real and sustained across different distributors with different ERPs and different order patterns, it is genuinely impressive. Most order automation tools in adjacent industries hover around 70-80% because edge cases are brutal.

Beyond order intake, Burnt has agents for procurement, inventory management, and credit control. The procurement agent handles purchasing decisions to keep inventory stocked while optimizing costs. The credit control agent automates collections and follow-ups. The vision is not a single-feature product but an operating system layer that sits on top of the distributor’s existing technology and handles all the workflows that currently require manual data entry and repetitive decision-making.

The white-glove onboarding model makes sense for this market. Food distributors are not going to self-serve their way through an AI deployment. They need someone to come on-site, understand their specific ERP configuration, train the agents on their specific order patterns, and make sure the system works before walking away. That is expensive to deliver but it builds deep switching costs once installed.

The competitive landscape includes companies like Choco, which is building a digital ordering platform for food supply chains, and Cut+Dry, which focuses on digital commerce for foodservice distributors. But those companies are trying to replace existing workflows with new platforms. Burnt is augmenting the workflows that already exist. In an industry that resists change with the ferocity of a line cook defending their station, augmentation is a smarter entry point than replacement.

The Verdict

I think Burnt is one of the more interesting vertical AI companies I have looked at recently. The market is enormous, the pain is acute, the founding team actually understands the domain, and the product approach of working with legacy ERPs rather than against them is strategically sound.

The risk is execution at the integration layer. Every food distributor’s ERP is configured differently. Order formats vary. Product catalogs vary. Business rules vary. Building agents that generalize across this variation without breaking is an ongoing engineering challenge, not a one-time problem.

In thirty days, I want to see the accuracy metrics from multiple distributors, not just the flagship customer. Ninety-seven percent at one distributor is promising. Ninety-seven percent across twenty distributors with different ERPs would be remarkable. In sixty days, the question is expansion beyond order intake. If distributors are adopting the procurement and credit control agents, the operating system thesis is working. If they are only using order intake, Burnt is a feature, not a platform. In ninety days, I want to understand the unit economics. White-glove onboarding is expensive. If the average contract value supports that delivery model with healthy margins, this is a real business. If onboarding costs eat the first two years of revenue, the model needs work.