XMarts room

Summer 2025

burnt

Agentic Operating System for Food Supply Chain

Visit website

What they do

agent team that runs order entry and ERP work for food distributors

How close to us

5, very close

agents per department, learns each customer, checks with customers; one layer for one firm type, active

Wedges

  • W5 One layer for one type of small firm direct
  • W7 Find and fix margin leaks adjacent
L1 Company brain (knows the organisation) 2, clear
L2 Agent team (runs agents, tools, approvals) 2, clear
L3 Does and learns (finishes, checks, measures, updates the brain) 2, clear

Who buys

Firm size
mid-market
Buyer type
business
Role
operations chief or owner at a food distributor (sales and customer service teams use it; finance for credit control)
UK
yes
Their own description (from YC)

Food Supply Chain companies have been using legacy ERPs for decades. Everyone knows these are necessary evils that essentially run the business but take up a lot of operational capacity. They are not about to just rip and replace their existing system that runs their whole business with a system that takes years to implement. Imagine if they could implement AI that runs the ERP for them, in weeks?

Burnt’s building an agentic OS for the Food Supply Chain with AI Agents for each department.

Our AI Agent, Ozai, has started by replacing the whole order management job function at food distribution companies, so that sales and customer service teams can focus on what matters - the customer. Ozai monitors all incoming messages (emails, phone calls, voicemails, whatsapp, fax etc.), figures out if it's an order, learns each customer's habits, shorthand, and favorite products, picks up on any prep notes, and enters it straight into the existing ERP. It'll even go back to the customers on your behalf when it's not sure about something, clarify things and carry on working.

The more it sees, the smarter it gets.

Full teardown

burnt (YC Summer 2025): teardown vs XMarts

In one line: burnt sells AI agents to food distributors. The first agent, Ozai, reads orders from email, phone, WhatsApp and fax and types them into the distributor's old ERP [site] [yc]. Threat: 2, because it is the cleanest example of our W5 shape (one layer for one type of firm, agents that learn each customer), but it is one industry, three founders, and no named customer [yc] [press].

1. Fact file

Item Finding Source
What they say "Automate food distribution." YC line: "Agentic Operating System for Food Supply Chain." [site] [yc]
Founded, team, funding Founded 2023. YC page shows team of 3 (may be stale, UNVERIFIED for today). $3.8M seed, Sep 2025, led by Penny Jar Capital (Steph Curry's firm), with Scribble Ventures, Formation VC and angels incl. Dan Scheinman. Based in San Francisco. [yc] [press]
Founders Joseph "JJ" Jacob (CEO), Rhea Karimpanal (CPO), Chandru Shanmugasundaram (CTO). JJ is a fourth-generation food entrepreneur who says he has bought over 400M lbs of shrimp, and worked at Rekki (B2B restaurant marketplace). JJ and Rhea met at age 12 in Dubai. Chandru built his first ERP at 17. What YC likely saw: founders who live the industry, plus a CTO who has built an ERP (inference). [yc] [press] [site]
Customers No names anywhere. They claim large independent distributors of meat, seafood and specialty ingredients. TechCrunch says one of the UK's largest food groups (billions in revenue) is implementing it, unnamed. [yc] [press]
Traction (they claim) 97% of orders at their large distributor customers run with no person touching them; 90% of order-entry staff moved into sales. More than $10M of orders a month processed, six-figure revenue, launched in January (year not stated). [yc] [press]
Price Not public. Every call to action is "Get a demo." One open role, Founding Engineer, $125K to $250K and 0.10% to 1.50% equity. [site] [jobs]
Path Founded 2023, live with customers by January (inference: 2025), YC S25, seed Sep 2025. Pages now show three agents (sales and operations, procurement, credit control). Procurement and credit control are waitlist only. Newest blog post 12 Oct 2026. [press] [site] [blog]

2. Their proposition, reverse engineered

Part What they do
Who buys Food distributors that sell to restaurants. Sales, operations and finance teams use it [site].
The pain Old ERPs are "necessary evils" that need many people to type orders (10 to 50 people's time, they claim) and are too slow and risky to replace [press] [yc]. Orders arrive as voicemail, WhatsApp, fax, free text [yc].
The promise Keep your ERP, add agents on top, go live in 1 to 2 weeks, scale without adding headcount [site].
How it works Reads every inbound channel, decides if it is an order, matches products to SKUs, checks stock, writes to the ERP, and asks the customer when unsure. Keeps a profile per customer (shorthand, specs, habits). Confidence thresholds decide what runs alone and what a person reviews. Exceptions go to sales, ops or finance with an audit trail [site].
The output A finished order in the customer's own ERP, plus confirmations to the customer [site]. Procurement agent writes purchase orders; credit agent runs collections calls (both waitlist) [site].
Proof style Percent of orders with no human touch, minutes saved per order ("10 minutes to six seconds", they claim), plus headline claims: 99.99% accuracy, 80% less admin time, 20% profit uplift. No baselines, no named case study [site] [press].
Who sells it Demo-led sales by the founders (inference from the site and the 3-person team). White-glove on-site onboarding, with agents built at the distributor's site [site]. Member badge of IFDA (food distributors' trade body); what the membership gives them is not explained [site].
Price model Not public [site]. Probably a subscription or per-order fee (inference: none stated).
The story Insiders, not outsiders: "The food world doesn't run on clean logic or perfect systems." Enemy is the legacy ERP and the "rip and replace" project. Belief: protect the human craft of food, use AI where it works [site] [blog].

3. Positioning

  • Category they claim: agentic operating system for the food supply chain, "AI watchtower for food distributors" [yc] [site].
  • Category a YC partner would put them in: vertical AI for one industry's back office, starting at order entry (inference).
  • What they want to be known for: the AI that runs a food distributor's day, one department at a time.
  • Where they sit on the map: layers L1/L2/L3 = 2/2/2, buyer = mid-market, business.

4. Three layers vs ours

Layer Them (score + evidence) Us (Marton's logs show)
L1 Company brain 2. A per-customer profile (shorthand, specs, ordering patterns) that later orders read. A "single source of truth" is claimed on the platform page. It is a customer memory, not a company-wide brain [site]. A vault of about 1,400 pages, 112 call transcripts and about 8,700 customer quotes, searchable by meaning.
L2 Agent orchestration 2. Several agents (sales and operations, procurement, credit control) with approvals, confidence thresholds, and routing of exceptions to people. Two of the three are waitlist only, so the "team" is partly a plan [site]. About 5,600 jobs run by agents from three makers, each checked by a different AI, then automatic tests.
L3 Execution and learning 2. Does the work (writes the order to the ERP), checks it with thresholds, keeps an audit trail, and learns from human fixes to cut exceptions. Results are claimed, not shown. No public sign that measured results change the next run beyond exception fixes [site]. Every failure gets a record and an automatic check (about 160). Offer feedback lines fell from 44 to 18 as rules were added. Only about 15% of jobs reached the main code, and merges were unreliable in October.

5. Same and different

Topic Same or different Them Us
Core belief Same Do not replace the old system. Put agents on top and learn the customer [site]. Same shape: one layer for one type of firm (W5 in the brief).
Kind of work Same Agents do the work and own the output (an entered order) [site]. Marton's logs show agents do and check work, on his own business.
Learning Same Learns from human fixes on exceptions [site]. Marton's logs show rules added from failures and feedback.
Buyer Different Food distributors, one industry, with an operator-founder from that industry [yc]. SMEs and scale-ups, no industry picked, no customer yet (brief).
Who owns the result Same The customer's ERP holds the result; burnt runs inside it [site]. We would run inside the customer's tools (unproven).
How they capture Different Agents built on site at the distributor, watching orders arrive [site]. Calls into a vault, plus checked job results (system inventory).
Size of team Different 3 people on the YC page [yc]. 2 founders; Marton's system has run 5 months.
Proof Different Percent of orders run end to end (97%, they claim) and a UK group implementing it [yc] [press]. No customer number yet.
Delivery Different Software plus white-glove on-site onboarding [site]. The system plus the two founders.
Price Different Not public; demo-led [site]. Not set yet.

6. Wedge overlap

Wedge Fit Why
W5 One layer for one type of small firm direct Exactly this model: one industry, an operating layer above the old system, agents that learn each customer [yc] [site]. Their firm type is food distributors; ours is not chosen (brief).
W7 Find and fix margin leaks for commercial or finance leaders adjacent Their credit-control page claims a 45% fall in days sales outstanding and the procurement page a 15% cost cut, both waitlist-only and without baselines [site]. Same finance buyer problem, much narrower scope.

7. Whitespace (where they stop)

  • Buyer gap: food distribution only. Every page, customer type and job ad is food [site] [jobs]. To own other firm types we need a named person with the pain in a different sector, ideally UK.
  • Number gap: headline claims (99.99% accuracy, 20% profit uplift) with no named customer and no baseline [site]. To own it we need one workflow, 4+ weeks, a before/after number a finance person would sign.
  • Price gap: price is hidden and delivery is on-site and white-glove [site], which suggests an expensive way to serve each firm (inference). To own it we need a price a 50-person firm can pay, tied to the result.
  • Layer gap: not a clear gap. They already cover do, check and learn on order entry. Their weak point is breadth: two of three agents are waitlist only [site].
  • Speed gap: they say 1 to 2 weeks to go live [site]. That is hard to beat. We should not claim this gap.

8. For the YC application

  • Proven today: Marton's logs show his system runs agents that do and check work: about 5,600 jobs, each checked by a different AI, and about 160 failure records that each added an automatic check. burnt's public results are claims about one job (order entry) in one industry [site].
  • Draft answer to "how are you different from burnt?": DRAFT. "burnt is a good example of the shape: one industry, agents on top of the old system. They chose food and the order-entry job because the founders lived it. We would pick the firm type from calls with buyers, and our system already runs the whole loop of memory, agents and checks on our own business." Founders say it their own way.
  • Future claim (needs proof): "We run one workflow for a firm type like burnt does for food, with a number that moved." Needs one customer workflow run for 4+ weeks, with a before/after number.
  • What we understand that they don't: nothing proven against them on their ground. They are ahead on a live customer in one industry. Our edge is the checked-work loop across layers, needs proof on a customer.
  • Copyability (M3): medium. The tech is in reach: they run agents with approvals and learn from fixes [site]. But their edge is industry depth and family trust in food [site] [yc], and moving to other industries would pull a 3-person team off its niche (inference). Not hard, so M3 can not score above 3 against burnt alone.
  • Why now (M4): None found in this teardown. Their funding date (Sep 2025) and a new "agentic" label are not a dated change that helps us more than them.
  • Do not claim: that we have a customer or number, that we are ahead on live results, or that W5 is whitespace (burnt is funded and live on it for food). Do not say they are only a point solution: they have three agents on the site.

9. Ideas worth taking

  1. Pick the firm type from the founders' own life. burnt's pitch rests on family history (85 years in seafood, they claim) [yc]. Trust comes before the product.
  2. "Autonomy with control": confidence thresholds, humans on exceptions, audit trail, customer sets the level [site]. This answers the "how is client data kept safe and who checks" question in plain words.
  3. A blog that teaches the buyer's world (beef prices as a credit risk) and ties it to an agent [blog]. It sells without a pitch.

10. Unknowns

  • Who the UK group is and what "implementing" means. Look for press naming it, or ask a distributor contact.
  • Real price and contract shape. Ask a buyer who has seen a quote.
  • Current team size and revenue. The YC page may be old; check new job posts and any 2026 funding news (a search found none).
  • Whether the 97% figure is across all customers or one site.

Sources

Technical detail
Slug
burnt
Lens hash
179096c479
Pass
deep
Stale
no