This is how I'd work an Exa territory: the accounts I'd open first, why now, who I'd talk to at each one, and what I'd actually say. And since Exa's product is built for exactly this kind of research, I used it to build half the list.
Funding trackers and PR feeds scanned by hand. Sales Navigator saved searches. Cross-checking each company against its own site, news, and job posts. A spreadsheet as the source of truth. A day of sourcing per tier before writing a single message, and the list starts going stale the moment it's finished.
One plain-English query in Websets Agent mode. Thirty verified companies back in minutes, each row carrying its funding source and date. It caught something manual sourcing might have missed, one result was already an Exa customer, and the whole run re-executes in one click, so the territory refreshes weekly instead of decaying.
These companies build agents that do research for a living. When the agent's sources are weak or stale, their product is weak, so better retrieval isn't a convenience for them, it's a better product. That's what makes them the right first accounts. The ones Websets itself found lead the list. Tap any account to expand, and tap a name inside it for my read on that person.
$120M Series C at a $1.2B valuation, led by Khosla Ventures, closed July 7, 2026. Source: PR Newswire, via Websets run.
A compliance determination is only as defensible as the sources behind it, and the regulatory record changes daily. Agents making those calls need real-time, cited retrieval over live regulatory content. That is a grounding layer, not a crawler project Norm should build in-house.
Founder-led company, so the architecture call and the budget call are the same call.
He's selling regulated enterprises on trusting an AI with compliance decisions, which means he spends a lot of time answering some version of "how do I know the agent is right?" Every sales cycle probably hits that wall.
It gives him a concrete answer. Every determination carries cited, dated sources, so "trust the agent" becomes "check the agent's citations." That's a sales asset as much as an engineering one.
He owns the stack a retrieval layer would plug into, and nothing gets bought without him.
Post-Series C he's hiring fast and deciding what his team builds versus buys. Keeping web content fresh for the agents is the kind of work that eats engineers without making the product smarter, and my guess is he'd rather spend those hires on the reasoning layer.
Retrieval becomes an API call instead of a system his team maintains. The engineers he was going to put on crawling go to work Norm actually differentiates on.
Legal engineering is the team that turns regulation into something agents can execute, which makes him the person who feels source quality daily. He's the champion seat.
When a regulator updates guidance, someone on his team has to catch it, and I'd bet that's closer to a manual review cycle than a feed. The nightmare scenario is an agent citing last quarter's version of a rule.
Monitors watches sources for changes continuously, so freshness stops depending on someone remembering to check. His team finds out the day guidance moves.
$22M Series A led by AVP, closed July 2, 2026. Source: PR Newswire, via Websets run.
An alpha product lives on being first and being right. Agents surfacing market-moving signals need comprehensive real-time retrieval plus continuous monitoring, which maps directly onto Exa Search and Monitors. Sub-450ms latency matters when the signal has a shelf life.
Two-person committee at a Series A company, and he signs.
Fresh off the raise, he's proving to hedge fund clients that an AI analyst can be trusted with investment research. Speed and sourcing are the whole pitch, so anything that makes the product faster or better-sourced is a revenue conversation for him.
The retrieval layer stops being a thing his small team builds and maintains, and the product's answers come back cited, which matters when the reader is a portfolio manager.
He owns the agents that retrieval feeds. Technical evaluation starts and probably ends with him.
A signal product lives or dies on catching things first. If his agents miss a filing or a management change that a client's analyst caught manually, that's a churn conversation.
Comprehensive retrieval plus continuous monitoring means the agents watch more of the web than any manual process could, and latency under half a second keeps "first" realistic.
$45M Series B (TCV, Bain Capital Ventures, First Round), April 2026. Source: Business Wire, via Websets run.
GTM agents need exhaustive coverage of people and companies, which is exactly why Exa built comprehensive indexes over those verticals and why Monday.com built its lead agents on Exa. This is Exa's own stated priority vertical, applied to a non-customer.
He's the door. A GTM founder at a GTM-agent company will take a call about retrieval quality because it's his product's ceiling, and a good outbound email to him doubles as a demo of the craft his product sells.
He's in customer conversations where the objection is some version of "your agent missed half my market." Coverage complaints land on his desk before anyone else's.
His agents research accounts against an index built for exactly this, which is the same reason Monday.com put its lead agents on Exa. Coverage stops being the objection.
He signs, and at a company this size he's still close enough to the product to have opinions about infrastructure.
Post-Series B, the pressure is proving the "superintelligence" framing holds up against cheaper lead tools, and the difference is research depth.
Depth gets cheaper. His agents can read the whole web per account instead of whatever a scraper reached.
Second technical founder, likely the one deepest in the agent architecture. If Karr opens the door, Gupta is the evaluation.
He owns the tradeoff between research quality and cost per account, and that math decides their margins.
Usage-based retrieval with verified results moves that tradeoff in his favor, and he can test it against their current setup in an afternoon.
25,000+ agents in production per Harvey's own published numbers, with research workflows expanding across jurisdictions and into client environments via the Microsoft 365 work.
Legal research agents win or lose on retrieval. Case law, filings, and the live web sit beyond any internal document system, and lawyers need results they can cite. Comprehensive, verified retrieval is the difference between an answer a lawyer trusts and one they re-check by hand.
Siva Gurumurthy, CTO · Sridhar Reddy, VP Engineering · Anique Drumright, Chief Product Officer. Siva owns the architecture call, Sridhar owns implementation at agent scale, and Drumright owns whether retrieval quality shows up in the product lawyers touch.
$250M Series D at a $12B valuation (January 2026, Thrive + DST), earmarked explicitly for its multi-AI agentic architecture. Reported ~$300M annualized revenue as of July 2026.
Medical literature doubles every five years, with 1.5M new papers annually. As the agentic architecture reaches beyond curated journals toward live evidence, the retrieval layer has to be comprehensive, current, and citable, because the end reader is a physician mid-consultation.
Zachary Ziegler, Co-Founder & CTO · Micah Smith, VP Engineering · Evan Hernandez, Chief Scientist. Research-proud org, so Ziegler and Hernandez are peer conversations about retrieval architecture, with Smith owning what ships. Founder Daniel Nadler stays out of the first motion.
Bigger committees, longer cycles, worked in parallel with the developer segment rather than after it. The Websets find leads here too.
Seer Agent entered open beta April 28, 2026: an agent that investigates application errors and telemetry through the UI or Slack. Source: Sentry blog, via Websets run.
An agent investigating a production error reaches for the same things an engineer does: docs, changelogs, issue threads, and vendor advisories across the live web. Exa's code-tuned search exists for precisely that lookup, and coding agents are Exa's other named priority vertical.
Seer just went to open beta, so someone owns making an investigation agent genuinely useful, and they're measured on whether it resolves errors faster than an engineer would.
An agent investigating an error wants the same things an engineer opens in twelve tabs: docs, changelogs, issue threads, advisories. Building that lookup in-house is a distraction from the investigation logic itself.
Code-tuned search over the live web as an API. The agent gets the twelve tabs without anyone building a crawler.
Ramp's internal coding agent merges a majority of production pull requests through MCP servers, per their own public engineering talks, and their price intelligence product is built on live web pricing data.
Two lanes in one account. Coding agents map to Exa's embedding models fine-tuned for code and technical documentation, and vendor price intelligence is structured web retrieval at scale. Both are Exa's named strengths hitting one buyer bench.
Rahul Sengottuvelu, CTO · Lewis Drummond, Head of Infrastructure. Rahul is two months into the seat, the window where inherited tooling gets reassessed, and Drummond owns the infrastructure the coding agents and price intelligence actually run on. Two threads, two lanes, one account.
An AI assistant handling the majority of consumer service volume, plus a shopping product whose recommendations depend on current merchant and product data from the live web.
Commerce answers age in hours. An assistant recommending products at Klarna's scale needs retrieval that reflects the web as it is today, not as it was at index time. Enterprise SLAs, latency, and SOC 2 are the qualifying bar, and Exa clears all three.
AI platform / engineering leadership. Enterprise cycle: multi-threaded from the first touch.
The flagship account worked the way I would actually work it: one thread per committee member, each on the pain that seat owns, three touches building on the same thread toward a meeting. Different proof in every thread, so the committee never sees the same email twice. These three emails are touches 1, 2, and 3 of the first chain. The full cadence they sit inside is in the next section. Tap a persona to expand.
John,
A $120M round to scale compliance agents means more determinations, in more regulated domains, in front of more scrutiny. Each one is only as defensible as the sources behind it, and the regulatory record changes daily.
That makes it a retrieval problem before it's a reasoning problem. Exa gives agents real-time, cited search over the live web, the same layer Cognition runs under Devin for all of its web access, at under 450 milliseconds.
Worth 20 minutes on how other agent teams ground their determinations without building retrieval in-house?
Izzy
John,
One more on the grounding thread. When a determination gets challenged, the question becomes what the agent read and when it read it. Exa returns cited, dated sources with every search, so each determination carries its own evidence trail from the start.
How is Norm handling source citation today, or is that a layer still being built?
John, quick one. If a client's regulator challenged a determination tomorrow, is the evidence trail behind it something Norm can produce today, or something that would take assembling?
Scott,
Regulatory AI has a quiet infrastructure tax. Crawling, indexing, and keeping web content fresh is a full product on its own, and every engineering week it absorbs is a week not spent on the reasoning layer that actually differentiates Norm.
Exa exists so agent teams skip that tax. A search API over an index of 500 billion pages, built for AI from scratch, so your team consumes retrieval instead of maintaining it.
Worth 20 minutes to compare how other agent teams drew the build-versus-buy line on retrieval?
Izzy
Scott,
The part that usually settles the build-versus-buy question is what scale does to it. Exa is engineered toward hundreds of thousands of searches per second, which is the kind of number that turns an in-house crawler from a side project into a permanent team.
Is keeping retrieval current something an engineer at Norm owns today?
Scott, quick one. Is retrieval infrastructure something your team is actively building right now, or handled well enough that it's not front of mind?
Paul,
Legal engineering sits closest to the failure mode that matters most here: an agent citing guidance that a regulator updated last month.
The fix lives at the retrieval layer. Exa returns real-time results with the source attached to every one, so what your team encodes and what the agents cite both trace to something current and checkable.
Worth 15 minutes on how cited retrieval slots into a legal engineering workflow?
Izzy
Paul,
Related piece worth knowing about. Exa Monitors watches sources for changes continuously, so when a regulator updates guidance, the platform can know the same day rather than at the next review cycle. For a team encoding regulation, that turns freshness from a manual audit into a feed.
Curious how change detection works on your side today?
Paul, quick one. When a regulation changes, roughly how long before it's reflected in what the agents cite, hours or a review cycle?
The three emails above are the first chain. Here's everything else those personas would experience over three weeks: two email chains on two different angles, calls stacked with the emails, and LinkedIn opened early but not used until it's earned.
Two rules run underneath the calendar. Effort follows engagement, so an open, a reply, or an accepted connect pulls a call forward regardless of what day it is. And calls stack with emails on purpose, since paired touches connect at roughly three times the rate of lone ones.
None of these three people should hear the same opener. Openers only; the rest of the call follows discovery.
Open where the trigger is freshest and the buyer is clearest: Norm Ai and LinqAlpha both closed rounds within the last four weeks and both have identifiable owners. The developer segment gets daily motion; the enterprise accounts run as multi-week, multi-threaded builds in parallel.
And every week starts with a fresh Websets pass, because the run that produced half this map takes minutes, and territories built on live retrieval do not go stale. That's the advantage of selling a product you actually use.