Their Headache Is Your Revenue – If You Can Offload the Pain to AI

01 /The Concept

Someone made the Panta website intentionally serious – so serious that it stopped being interesting. But the startup itself is considerably more interesting than its homepage suggests, and that's what this review is for.

Panta is an insurance broker for "the industries that build our world." For reference: an insurance broker is an intermediary who helps a client find the right coverage from insurers at terms that actually protect them.

The industries Panta considers "world-building": construction and renovation, industrial manufacturing, agriculture, logistics, and transportation.

Panta's tagline is "Smarter. Faster. Done." The startup deliberately goes after specialized markets and niches that most brokers avoid – because the complexity isn't worth their time.

Panta also makes a point of doing more than placing policies. The company says it will fight for its clients – pushing insurers to actually pay out when claims arise. Under the hood, AI agents are doing most of the heavy lifting.

Panta was founded in 2024 and is currently in Y Combinator's accelerator. The company published its platform listing on the YC site a few days ago – with a note that the platform runs "OpenClaw-style on a rack of Mac Mini computers," which is what made this worth a closer look.

02 /Why It Matters

Most insurance policies are straightforward. A broker receives a request, pulls from a set of standard options, compares terms, and recommends the best fit. Routine.

But complex cases are a different matter. A broker handling a specialized or high-risk policy needs to identify willing insurers, negotiate bespoke terms, and produce documentation precise enough that the client can actually collect when something goes wrong. That process involves roughly 50 distinct steps spread across two weeks.

Here's the catch: of that total time, maybe 5% actually requires a broker's judgment. The remaining 95% is pure coordination overhead – sending follow-ups, responding to queries, processing incoming documents, comparing terms across options, proofreading contracts, marking up revisions, proofreading again.

Most brokers won't touch complex cases for exactly this reason. Their time is their only resource, and standard cases are far more efficient. Even so, a top broker can only handle around 400 clients.

And yet the complex-case segment is substantial. It even has a name: the Excess & Surplus (E&S) market – insurance for non-standard and hard-to-place risks. According to Panta's figures (which Gemini confirms), the E&S market in the US alone generated $125B in annual premiums – and grew 12.5% in 2024, accounting for 9.2–12.3% of the total property and casualty insurance market.

Facing a market that size, Panta's founders could have built a software tool to help brokers move through those 50 steps faster. That's actually what they did first.

But they found themselves earning pennies on it. Making a broker 10% faster doesn't generate meaningful margin. Getting a broker from 400 clients to 440 doesn't change the business model – it's not worth much to them, and there's nothing to sell.

The bottleneck isn't step-level speed. It's the total size of one human's brain and calendar. So how do you scale a broker from 400 to 4,000 clients? By eliminating the human from as much of the process as possible – starting with the 95% that doesn't actually require judgment.

Once the founders internalized this, they stopped selling software and became an AI broker instead.

Technically, Panta's AI broker is a set of agents running on the OpenClaw platform, deployed on a full rack of Mac Mini computers. Each agent has access to a file system, browser, email, and phone – which means they operate just like human brokers would:

- Finding insurer websites, filling out submission forms, using specialized platforms, sending emails.

- Responding to underwriter questions, generating and sending documents, following up until they get a clear accept or decline.

- Creating Certificates of Insurance (COIs) with coverage descriptions, then iterating with clients and insurers until they're finalized.

- Preparing renewal documentation – including updated exposure summaries – and getting pre-approval from all parties before the current policy expires.

Human specialists at Panta stay in the loop for key decision points – but they're reviewing a complete interaction history already assembled by the AI agent, and mostly choosing from agent-proposed options, overriding suggestions, or giving approval. The context is already there; the specialist just needs to decide.

One example from the Panta team: a security firm responsible for a 100,000-person stadium had been turned down by six different brokers over a month. Panta took it on – and delivered binding offers from underwriters within three days, with the AI agent composing, sending, and responding to more than 60 emails to get there.

03 /Opportunities

The point here isn't to rush out and build an insurance AI broker. That's just one instance of a much broader trend – the "AI agency" model – which keeps showing up because it's genuinely powerful.

Even YC, in its latest Requests for Startups list, broke out "AI agencies" as a separate category – noting that traditional agencies were always hard to scale because more clients required proportionally more headcount.

With AI, a startup can use the same infrastructure it would have packaged as software – and instead sell the outcome directly, at 100x the per-unit economics of a SaaS subscription. And it scales without adding headcount. Panta is a textbook illustration of this.

A [recent review on AI agencies](/review/to-zhe-samoe-no-v-100-raz-dorozhe) covered a range of examples from design, advertising, and SEO. But a couple of less obvious cases are worth repeating here.

Last fall, WorkHero ([related review](/review/prodavaj-vot-takoj-servis-vmesto-it-platformy)) raised $5M in its first round for a service that places AI-powered office managers inside HVAC companies – running faster and cheaper than anything a human hire could deliver.

In late January, Phoebe ([related review](/review/platforma-dlja-samoj-bolshoj-professii)) raised $9.5M in a first round for a platform that turns any home care or eldercare company into a "self-piloting agency."

So: the direction is clear. Build an AI agency in a domain you understand well enough to deliver real outcomes cheaply. The question is which domain.

One particularly sharp strategy: don't target the whole market. Target the segment of complex cases that everyone else is turning down. In a large market, even 10% of rejected business is a lot of revenue – with minimal competition, if you move early enough.

The more painful and hassle-heavy the work, the better. Human specialists hate it. AI agents don't care.