Hexa Builds the AI Agents That Quietly Run Industrial Distribution

Somewhere right now a distributor of steel fasteners is losing a bid. Not because the price was wrong or the parts were out of stock, but because a request for quote came in as an emailed PDF at 4:47 in the afternoon, and by the time a person opened it, keyed the line items into an ERP built before the smartphone, and sent back a number, a competitor had already answered. In industrial distribution, speed is often the whole game. Hexa was built to close that gap.

Hexa, part of Y Combinator's P26 batch, makes AI agents for the mid-market companies that keep physical supply chains moving - distributors and manufacturers dealing in bearings, pipe and valve fittings, fasteners, electrical parts, food service supply. These are businesses with enormous transaction volume and back offices that still run on inboxes, spreadsheets, and decades-old enterprise software. Hexa's pitch is narrow and specific: point AI agents at the repetitive work, and let people do the rest.

It is worth pausing on who these customers are, because they are easy to overlook. A regional bearings distributor might carry tens of thousands of SKUs, field hundreds of quote requests a week, and run the whole operation on a system that predates cloud software. The work is not complicated so much as relentless: read the request, find the parts, check the price, confirm the stock, type it back. Repeat, all day, across sales, purchasing, and accounts payable. The people doing it are experienced and capable. They are also buried, and the burial is the problem Hexa is trying to solve.

Quote faster, win more bids

The company's own one-line description is plain enough: "Hexa helps industrial distributors automate their sales and procurement workflows to quote faster and win more bids." The first product most customers meet is the sales agent. It reads an inbound RFQ - from email, from a PDF attachment, from a spreadsheet - generates a quote, and enters the order directly into the ERP. What took a person a chunk of an afternoon happens in minutes.

From there the agents fan out across the functions a distributor actually runs on. Procurement agents watch inventory, send supplier RFQs, compare prices, and cut purchase orders. Finance agents match invoices to those POs, route approvals, track payments, and follow up on the ones that come in late. Customer-service agents field routine questions - order status, pricing, availability - using live data pulled from the ERP rather than a stale export.

The unglamorous truth of the back office is that most of its work is neither creative nor final. It is transcription. A number lives in an email and needs to live in the ERP. A line item on a PDF needs to become a row in a quote. An invoice total needs to reconcile against a purchase order that was cut three weeks ago by someone who has since moved on to the next hundred orders. Each hop is a chance to fat-finger a digit, and each error compounds downstream. By having one agent carry a request from inbox to order to invoice to collections, Hexa is not only saving keystrokes - it is removing the seams where mistakes and delays tend to hide.

No rip-and-replace

Most enterprise software has a hidden tax: to get the benefit, you first have to tear something out and put the new thing in. That is where a lot of industrial-tech deals go to die, because a distributor running SAP or NetSuite or Epicor is not going to gamble its order book on a migration. Hexa's design choice is to ask the customer to change nothing. The agents run natively inside the ERP and CRM systems already in place - SAP, NetSuite, Epicor, Microsoft Dynamics, Oracle, and on the sales side Salesforce, HubSpot, Zoho, Pipedrive. Implementation is typically live in two to six weeks.

That timeline is possible because Hexa pairs the software with people. Forward-deployed engineers embed with each customer to design and tune the automations to how that specific business works - its parts, its price lists, its quirks. It is a service-heavy way to sell software, and in a vertical this messy, the service is much of the point.

The forward-deployed model has become fashionable in AI circles, but it is not decoration here. No two industrial distributors run their ERP the same way. One has custom fields nobody documented; another prices off a matrix that lives, informally, in a senior buyer's head. A generic bot dropped into that environment fails quietly and expensively. An engineer sitting with the team for a few weeks learns where the bodies are buried and shapes the agent around them. The upfront cost is real. So is the payoff: an automation that fits the business rather than one the business has to bend around.

The unglamorous market, on purpose

Plenty of 2026's AI startups are chasing the same well-lit rooms: coding assistants, chatbots, marketing copy. Hexa went the other direction, toward a corner of the economy that is huge, essential, and almost entirely unserved by modern software. The alternatives it displaces are horizontal automation tools like UiPath or Automation Anywhere, brittle in-house ERP scripting, and - most often - a clerk working an inbox by hand. The difference Hexa is betting on is depth: an agent that knows one distributor's catalog, prices, and customers beats a general-purpose bot that knows none of them.

The expertise Hexa is assembling is less about any single model and more about the shape of the domain. Knowing that an RFQ can arrive as a photographed fax, that a distributor's margin logic changes by customer tier, that a purchase order and its eventual invoice rarely match line-for-line on the first pass - this is the accumulated texture of industrial operations. A general agent platform has to learn all of it from scratch for every deployment. Hexa is building it into the product, one customer at a time, and that compounding knowledge is the harder thing for a competitor to copy than any given feature.

Raised on a factory floor

Hexa was founded in 2026 by Ishaan Makkar, Auri Nayak, and Mann Patira. Makkar, the CEO, did not stumble into industrials from a demo-day whiteboard - he grew up around it. His family runs one of India's largest paper producers, and before Hexa he worked at a manufacturer building the very automations the company now sells. Nayak, the CTO, came through AI work at Bloomberg and software engineering at Autodesk, with a master's in machine learning and AI from Imperial College London.

It is a founder-market fit that reads less like a pitch and more like a homecoming. The people selling to distributors have sat where distributors sit. That matters in a business where the buyer has been pitched a hundred pieces of software that promised to fix everything and mostly created new work.

There is a credibility that comes from that history and cannot be faked in a sales deck. When a buyer at a distributor asks a skeptical question - what happens when the RFQ has a handwritten note in the margin, or the part number is one the catalog spells three different ways - the answer lands differently coming from someone who has actually processed that RFQ. Trust is the scarce resource in this market. Founders who grew up around the work have a head start on earning it.

What it costs, who it's for

Hexa is B2B SaaS with pricing that starts around $3,000 a month and scales with company size and how many workflow modules a customer switches on. The buyers are mid-market distributors and manufacturers - large enough to feel the volume of standard transactions, not so large that they have already built an internal automation team. The math the company sells is simple: an agent that helps you answer more bids faster, more accurately, is measured against the deals slow quoting used to cost you.

There is a plausible version of the next few years where this kind of company matters more than the flashier ones. The consumer AI products get the attention, but the dollars in a distributor's back office are concrete: bids won, hours returned, invoices collected on time instead of thirty days late. A tool that moves those numbers does not need to be exciting to be bought. It needs to work, fit the way the business already runs, and pay for itself. That is a narrower promise than most AI startups make, and a more legible one.

The larger bet under all of it is that the industrial economy will not modernize by throwing out its systems - it will modernize by having agents do the work inside them. If that is right, the least glamorous inbox in the building turns out to be one of the more interesting places to point an AI. Hexa is wagering that it is.