Before Matt Antoszyk and Duncan Crystal had a company, they had summer camp. They met as children at Camp Becket, years before either had a reason to care about carrier integrations. Antoszyk returned as a staff member. So did Crystal. Jacob Barnett, who would become Herald's chief technology officer, went to the same camp. Some founding teams assemble through investor introductions. This one had already spent time in the woods.
It is a pleasing origin for a business devoted to making commercial insurance less cumbersome. Camp promises fresh air and a break from paperwork. Insurance tends to supply the paperwork. Yet the connection has become more interesting as Antoszyk's company has developed: a long acquaintance became a professional partnership, and a software business built around connections is now making an argument about the staying power of human trust.
Antoszyk is Herald's co-founder and CEO. He and Crystal started it in 2021 after working together at At-Bay, the cyber insurance company. Their initial proposition was straightforward enough to explain and difficult enough to occupy a company: give developers one connection to multiple insurance carriers. Since then, they have moved into AI tools that work inside a broker's email. The inbox, with its accumulated attachments and unfinished business, has become the front door.
A computer scientist in the insurance business
Antoszyk studied computer science at Yale, graduating in 2014. He joined McKinsey & Company as a business analyst, then moved to At-Bay in 2017 as a product manager. He later attended Harvard Business School, earning his MBA in 2021. The sequence put him in contact with code, business decisions and the practical difficulties of bringing a new insurance product to market.
At At-Bay, he was the first product manager. His work involved tools that helped brokers and clients understand and buy cyber insurance. In 2018, he also wrote about the deceptively simple question of what counted as a data record, and why the answer mattered to the cost of a breach. It was an early example of a recurring concern in his work: the distance between information a system contains and information a person can use.
His experience with Crystal supplied a more immediate business opportunity. They helped build and distribute At-Bay's cyber insurance API, the software interface through which other systems could connect. Their customers encountered different data models, uneven documentation and the continuing expense of maintaining integrations. The trouble repeated across carriers and insurance products. A connection that worked once still had to work tomorrow.
That repetition gave the founders something concrete to pursue. Developers at separate businesses were solving variations of the same connectivity problem. Herald could do that work centrally, leaving its customers to build their own experiences around it. Antoszyk had arrived at a problem that rewarded familiarity with both the software and the people using it.
One connection, many moving parts
Commercial insurance puts several businesses around a single purchase. A client supplies information. A broker looks for suitable coverage. Carriers evaluate the risk and offer terms. Moving data between those parties can mean sending documents, entering the same details into different systems and waiting for replies. Each participant may have perfectly serviceable software. Getting those systems to cooperate is another matter.
Herald began by addressing that cooperation. A developer could integrate with its API to obtain quotes and bind coverage from several carriers. The attraction was partly the initial connection and partly the maintenance that followed. The more carriers a brokerage wanted to reach, the more useful a shared connection could become.
Antoszyk announced an $8 million seed round in December 2021, led by Lightspeed Venture Partners, with Afore Capital and Underscore participating. Among the individual backers were people with direct experience of his work and his market, including At-Bay co-founders Rotem Iram and Roman Itskovich. Underscore had invested before the product existed. Its reasoning centered on the founders' experience of the integration problem and their relationships in insurance.
Two announced rounds. $20 million combined.
The Series A brought another $12 million in October 2024, co-led by Lightspeed and Brewer Lane Ventures. Afore and Underscore participated again. At that point, Herald reported working with eight of the 25 largest brokerages. The company was building infrastructure for firms that already had customers, working routines and substantial reasons to be cautious about changing either.
That customer matters to the rest of Antoszyk's story. A brokerage with established systems needs technology it can fit into its business. Its employees have clients to serve while a new tool is being introduced. Software that asks everyone to stop and learn a different way of working has given itself an additional job before it has done the first one.
The second product starts in the inbox
By 2025, Antoszyk was describing a change in what Herald could offer. Its connectivity layer was in place. Generative AI opened a route from the documents and emails brokers already used into that infrastructure. The company could take on more of the work surrounding a transaction, beginning where a submission arrived.
Herald had been partnering with Babbix, whose founders, Kim Mishra and David Elbaz, were building AI agents for brokers. In July 2025, Herald acquired the business. The founders joined to lead its new AI division. The combination put email-based data extraction alongside carrier connections, with quoting and cross-selling among the initial applications.
Antoszyk acknowledged a limit in the original offering: enterprise brokers needed more than the foundational infrastructure. That admission makes the acquisition consequential. It meant taking responsibility for more of what customers actually did, beyond supplying a connection that another developer could use. An email attachment could become the starting material for an insurance transaction.
- 01A submission arrivesEmail, documents and business details.
- 02The agent prepares itExtract information and apply the brokerage's rules.
- 03Connections carry itReach carriers and bring quotes into the workflow.
In August 2025, talking about the transition, he sounded interested in the questions that followed a fresh product launch: finding the first users, building volume and learning how to sell it. He described returning to that early discovery work as energizing. An established infrastructure business had handed its CEO another beginner's assignment.
By October, Herald was opening a waitlist for Babbix, starting with cyber insurance. Brokers could indicate which lines of business they wanted next and help shape the product. In March 2026, the company's explanation had expanded to strategic placement: getting a submission through intake, marketing and presentation while preserving the brokerage's own choices about how business should be handled.
Who gets to decide what the machine does?
There is a stubborn human question inside that technical progression. Two brokers can receive the same incomplete application and choose different next steps. One may ask a client for clarification. Another may use information already on file. A system needs instructions about when it can proceed and when it must stop. Reading a document accurately does not settle every decision that follows.
Herald's 2026 product thinking addresses that distinction through configurable rules for triage, extraction and action. A brokerage can specify which submissions go to particular teams, what counts as complete and which uncertainties require review. The point is to carry the firm's judgment into the workflow. A faster process still needs a view about where it should go.
That approach became more explicit in July 2026, when Herald joined Guidewire's Insurtech Vanguards program. Its Intelligent Placement System combined Babbix, a workbench for managing submissions and quotes, and connections to existing data and carrier systems. Antoszyk described placement as the messy middle of brokerage operations. It is an apt location for a product manager to keep finding work.
The next day, Herald announced that former USI chief information officer Stew Gibson had joined its advisory board. Gibson had worked with the company while at USI. His appointment brought the experience of a large brokerage's technology operation into the conversation, including the practical business of getting people to use a system after the demonstration ends.
The oldest relationship in the room
Antoszyk has also challenged the idea that AI will simply transfer insurance customers away from established brokers. In a recent public post, he argued that adoption increases the importance of trusting the people behind a new tool. He expects existing brokers to benefit from relationships built over time, even as clients expect faster work and better insight.
“As the use of AI expands, the need for trust expands.”
Matt Antoszyk · on the future of insurance brokerage
This remains his argument about the market's direction, rather than a settled outcome. It is also consistent with his own career. Crystal was a friend before becoming a co-founder. Former colleagues became investors. A customer-side technology leader became an advisor. Herald's software has changed substantially; these relationships have continued to supply knowledge, scrutiny and access.
When asked in 2025 what he had learned from founding a company, Antoszyk emphasized finding capable people who would challenge him and maintain demanding standards. He was candid about the difficulty of the job. His advice put the quality of daily working relationships at the center, alongside the technology those people were supposed to build.
The camp connection makes a charming opening. The adult version requires more effort: choosing partners, listening when an initial product proves insufficient and returning to customers with something new. Antoszyk's company now aims to move much more of an insurance transaction through software. Its CEO keeps returning to the people who must be willing to let it.