Here is a fact about government that is both boring and enormous: before a city can buy a fire truck, repave a road, or replace the software running its water system, someone has to write a document. Usually a very long document, from a template, by hand, describing exactly what the government wants, on what terms, at what price, and how bids will be judged. This is called procurement, and it is roughly a $2.7 trillion-a-year activity in the public sector. It is also, in most places, run on PDFs, email chains, and the institutional memory of one overworked purchasing officer.
Hazel is a company that looked at that and decided the paperwork should write itself. Founded in 2024 and based in New York, it builds what it calls AI-native software for government procurement - tools that help agencies define what they need, generate the solicitation, research the market, and evaluate the responses, all in one place. The pitch, in the company's own words, is that Hazel is "the AI superpower for procurement teams short on staff and time but hungry for innovation."
That framing matters, because the easy version of this story - AI replaces bureaucrats - is not the one Hazel is telling. The harder and more accurate version is that the bottleneck in public procurement was never the people. It was the tools. A good procurement officer knows exactly what her city needs. What she lacks is the forty hours to draft, cross-check, and publish the document that gets it. Hazel is software for those forty hours.
What the thing actually does
Strip away the category label and Hazel is really four moves stitched together. First, it helps a team scope a project - figuring out what they are even asking for. Then it auto-generates the solicitation itself: RFPs, RFQs, and RFBs, drafted from the agency's own templates, boilerplate, and the language of past awards, with a check for policy compliance built into the process. Then it does market research and surfaces vendors - including, notably, local, small, and minority-owned suppliers that often never see the opportunity. Finally, when the bids come back, it helps evaluate and score them.
The unifying idea is compliance-by-construction. Anyone can get a chatbot to write an RFP. The trick in government is writing one that survives legal review, follows procurement code, and produces an audit trail when someone inevitably asks how the decision was made. Hazel's whole design bet is that in the public sector, trust is the product and the AI is just how you deliver it. The company says its models are built to show their work.
Dallas is proving that good government and great technology go hand in hand.
The founders
Hazel was started by two Harvard classmates, August Chen and Elton Lossner, who between them had already worked on some of the more consequential software around. Chen, now CEO, studied mechanical engineering and computer science, co-founded MakeHarvard - the university's largest makeathon - and went on to build wildfire-prevention applications at Palantir. Lossner also came out of Harvard, co-founded MakeHarvard alongside Chen, and spent four years at Boston Consulting Group leading strategy and procurement cases in aerospace and defense, for both government and industry.
That combination is more relevant than it sounds. One founder knows how to ship serious software against messy real-world data; the other spent years inside the exact procurement processes Hazel is trying to fix. They could, presumably, have built almost anything. They chose the least glamorous, most consequential thing in government - how it buys - which tells you something about where the genuinely hard, unclaimed problems still live.
Who is buying it
For a company this young, the client list is the interesting part. Hazel says it works with more than ten government clients across state, local, and federal levels, and reports supporting over $2.5 billion in procurement. Among the named adopters: the City of Dallas, the City of Atlanta, PhilaPort - the Philadelphia regional port authority - and the U.S. Air Force, plus K-12 school districts and federal agencies. The company also reports that every surveyed user and chief procurement officer recommends it, a claim that is easier to make with ten clients than ten thousand, but not nothing.
The Dallas deal, announced in July 2025, is the clearest window into what Hazel changes. The city - the first major Texas city on the platform - is using it to compress the stretch from project scoping to a published RFP from months to days, on real work: airport upgrades, housing initiatives. Just as pointedly, the city framed it as an equity move, using Hazel to surface a broader pool of local and diverse suppliers, particularly small and minority-owned businesses. Fairer competition, in other words, was not a side effect. It was part of the sale.
Hazel is the AI superpower for procurement teams short on staff and time but hungry for innovation.
The backing, and the bet
Hazel went through Y Combinator's Winter 2024 batch, with group partner Michael Seibel involved, and raised pre-seed capital reported at around $500,000 from investors including Rebel Fund, Tusk Ventures, and Phoenix Investment Club. By venture standards these are small numbers, which is arguably the point: a five-person team, three of them founding engineers, supporting billions of dollars in public buying. The leverage is the story.
The broader bet is a contrarian one. Selling to government is famously slow, procurement cycles are long, and "govtech" has been a graveyard of good intentions. But that difficulty is also the moat. If Hazel can prove that AI-drafted, compliance-checked solicitations hold up in the one environment where every decision can be audited, contested, and FOIA'd, it will have done something most enterprise-software companies never have to: earn trust in public. That is a slower way to build a company. It may also be a more durable one.
None of this is guaranteed. Government sales stall, incumbents like the legacy e-procurement platforms are entrenched, and "AI you can trust" is a promise that gets tested the first time a model gets something wrong in a document worth millions. But the thing Hazel is pointing at is real, large, and strangely neglected. Most of a $2.7 trillion market still moves through documents nobody wants to write. Hazel wrote software for the document. Sometimes the biggest lever is the one everyone else walked past.