Boon is the second employee in the back office.

There is a certain kind of company that looks, from the outside, like it is selling artificial intelligence, but is really selling something older and more boring: the removal of paperwork. Boon is that kind of company, and it is refreshingly honest about it.

Here is the pitch, stripped of adjectives. Somewhere in a trucking company's back office, a human being is retyping an order from an email into a transportation management system, then into an ERP, then checking a fuel card, then confirming an appointment slot. It is a job. It is not a good job. Boon's founder, Deepti Yenireddy, calls the product that does this instead "the second employee in the back office" - "another teammate doing critical work." That framing is doing a lot of work, because it sidesteps the anxious question ("is this replacing me?") and replaces it with a comfortable one ("what if the annoying part of my job just... did itself?").

Yenireddy did not arrive at this idea from nowhere. She spent time as a senior director of product at Samsara, the fleet-operations giant, where she ran product, telematics and international. This is a useful biographical detail, because Samsara's whole business is putting sensors on trucks and turning them into data. If you sit at that company long enough, you notice that the data is only half the problem. The other half is that the humans downstream are still doing enormous amounts of manual coordination. Boon is, in a sense, Yenireddy's answer to the thing she couldn't fix from inside a hardware company.

She had also done the startup thing before - she built an AI company in the HR space and sold it to Phenom People, a recruitment platform. Founders who have sold once tend to be less romantic about the technology and more interested in whether anyone will pay for it. That temperament shows up in Boon's numbers.

The numbers, as of the company's December 2024 fundraise: paying customers representing roughly 35,000 drivers and 10,000 vehicles, and an annual revenue run rate of about $1 million after nine months in business. A million dollars is not a large number in absolute terms. But "nine months" is a small number, and the ratio between them is the thing venture investors actually buy. Redpoint Ventures and Marathon bought it, leading a round that totaled $20.5 million - a $15.5 million Series A on top of a previously undisclosed $5 million seed.

What the money bought was a specific bet: that the winning form factor for AI in old industries is not a chatbot you talk to, but an agent that quietly does a defined task and plugs into the software you already run. Boon integrates with TMS, ELD, ERP, WMS, telematics and fuel-card systems - the unglamorous plumbing of a fleet - and automates the work that spans all of them. The company says its agents can learn to replicate a worker's task in roughly a week. For fleets, it claims average savings of $7 to $10 per truck per day, which is the kind of number a fleet manager can multiply in their head and then approve a purchase order.

Then something interesting happened, which is that Boon's public face moved. Visit getboon.ai today and the headline is not about trucks. It reads: "The intelligence layer for construction." The tagline: "AI agents that run preconstruction workflows from blueprint to bid, so your team can focus on winning work." This is a company that started in logistics and expanded into construction - and if you squint, you can see why the leap is smaller than it looks.

Preconstruction is, mechanically, a paperwork problem wearing a hard hat. Before anyone pours concrete, estimators do "takeoffs" - counting every fixture, conduit and fitting off a set of drawings. They do "clash detection," hunting for places where the plumbing runs through a beam that isn't supposed to be there. They do "bid leveling," comparing subcontractor bids that are all formatted differently to figure out who is actually cheapest. Each of these is a task where a human reads a large pile of documents very carefully and transcribes findings into a spreadsheet. Each is exactly the shape of problem Boon built its engine to eat.

The company's own scorekeeping tells the story: more than $4.5 billion in project value analyzed, and over 66,000 drawing pages read by its agents. A drawing page is dense - a single sheet can hold hundreds of items. Reading 66,000 of them is not a thing you want a person doing at 11 p.m. before a bid deadline, which is precisely when it tends to happen.

Boon's product surface reflects this generalized "agents that read and act" thesis. There is Agent Studio, a no-code way to build agents without engineers. There is a set of runtime pieces - Pulse for live visibility, Manage for oversight, Trust and Vault for the enterprise-security and document-storage reassurances that any serious buyer demands before letting software touch their data. And there are the construction-specific tools: Takeoff, Clash Detection, Bid Tracker, Bid Summary, Bid Leveling. The platform underneath is the same; the vocabulary changes with the customer.

The team is the other tell. Boon's roughly 73 people include alumni of Apple, DoorDash, Google, Samsara and Shell - a deliberate blend of consumer-grade product polish and industrial domain knowledge. Shell, in particular, is a name you don't put on a slide unless you mean it about understanding how fuel and fleets actually work. The customer list on the construction side skews toward electrical and general contractors: E.J. Electric, Rosendin, Marathon Electrical Contractors, Joeris, Pete King, Lemartec, HJD Capital Electric. These are not tech-forward startups. They are firms that win or lose on whether their estimate was right, which is a good market for a product that promises the estimate will be right and arrive faster.

None of this is guaranteed to work. Vertical AI is a crowded pitch in 2025 and 2026, and "we automate a boring workflow" is a sentence a hundred companies are saying at once. Boon's advantage, to the extent it has one, is unglamorous: it entered through a real workflow, charged real money for it, and grew the run rate fast enough that investors extrapolated. The pivot from fleets to construction could read as a lack of focus, or as evidence that the underlying engine travels well across industries that are drowning in documents. Which one it turns out to be is the whole question - and it will be answered by contractors and dispatchers, not by pitch decks.

For now, Boon occupies a clear and slightly funny position: a company built on the frontier technology of the moment, whose actual product is best described as "the part of your job you would happily hand to someone else." That is not a knock. It might be the most durable kind of AI business there is.