Korva is a scroll of the internet’s best ideas.
One email tool fits every business
HOA email work
In homeowner association management, a lot of the work happens in email. A board asks about finding a snow removal company, and the reply is not simple: the manager has to map the other ten communities, pull the five vendors, check what each charges, see how people reviewed them, and get schedules. Crafting that response and proposal can take forty-five minutes to an hour or longer.
Long Lake, which owns HOA management companies and deploys its own AI platform called Nexus in them, gives Nexus access to all that information. A draft can be waiting in the manager's inbox before they wake up, about ninety-nine percent accurate. The manager edits it into their own voice in five or ten minutes, and Nexus is getting better at drafting in each manager's own voice.
Travel inboxes
The capability transfers. Fifty-five percent of travel agent inbound queries from travelers arrive by email, covering new bookings, changes, and cancellations. Because the problem was solved in HOA management, the tool has a day-one impact at American Express Global Business Travel.
What carries over
Roughly seventy to eighty percent of Nexus is shared across industries, including model routing, security, and the layer that orchestrates agents. The last twenty or thirty percent means mapping each business's workflows and connecting its data sources, which is why about seventy percent of Long Lake's engineers sit with team members in the field across fifteen or twenty states.
Why has email outlasted every replacement?
Email runs on open standards, so a message can travel between any two providers with no company owning the network. Everyone with an address can reach everyone else, which makes it the default across organizations. The simple mail transfer protocol, still the basis for sending mail, dates to 1982. Ray Tomlinson sent the first network email on ARPANET in 1971 and chose the @ symbol to separate user from host.
How can AI write in someone's voice?
Language models learn patterns in text, so a sample of someone's past writing shows them that person's habits. Word choice, sentence length, greetings, and sign-offs all carry a style. A system can place a person's earlier emails in front of the model so the draft imitates them. The approach rests on how these models are trained: they predict the next word from the text they have been shown.
How does routing between models cut costs?
Model routing sends each request to the cheapest model that can handle it, and keeps the expensive model for hard tasks. Smaller models cost far less per token than frontier models. A 2024 paper from UC Berkeley researchers on a system called RouteLLM reported cost reductions of more than 85 percent on one benchmark while keeping most of a top model's quality.
More from David Senra
