DeepVista

It's Time for Relational Intelligence

Jing Conan Wang, Founder of DeepVista · Former Google Brain researcher

AI can reason about the world. But it lacks the relationship history needed to act in it.

What we can do in the world depends on our relationships with other people.

A former colleague who has seen your work can vouch for it from experience. Someone you met once may know what you do and have little basis for recommending you. The history between you shapes what each person can credibly say and do on your behalf.

We call this relational context. It’s how people know each other, what they’ve done together, and how those relationships change over time. It lives in careers (where people worked, when, and alongside whom) as well as in the meetings, emails, and projects they share.

As AI agents help us make more decisions in the real world, they’ll need access to that context. Any agent acting on your behalf eventually asks three questions:

  • Who is the right person?
  • When is the right moment?
  • How should it communicate?

Relational context is how an agent answers them, and it changes what becomes possible.

Relational context shapes what becomes possible

Here are a few examples of how relational context enables an action, and the opportunity that follows.

Relational contextAn employee previously worked closely with a candidate and knows the quality of their work.

Action it enablesAsk for an informed referral.

OpportunityA promising hire.

Relational contextA mutual contact knows a founder’s work and has an active relationship with an investor whose thesis fits.

Action it enablesRequest a credible introduction.

OpportunityA relevant investor conversation.

Relational contextA former customer champion joins a company facing a problem your product helped them solve.

Action it enablesReconnect with shared history and a relevant reason.

OpportunityA new customer conversation.

In each case, knowing who someone is provides a starting point. Understanding the relationships around them helps determine what to do next.

This context also changes. People move companies, priorities shift, collaborations deepen, and relationships become less active. Useful recommendations must reflect those changes.

The Dynamic Trust Graph

DeepVista is building the Dynamic Trust Graph: a structured, continuously updated record of relationships and the evidence behind them.

Two kinds of evidence build it: experience and activity.

  • Experience gives a starting estimate. Experience is the history of where people have been and who they were with. A career is one example. Three years on the same team at a 15-person startup is strong evidence; a brief overlap at a company of 8,000 is not. The estimate weighs duration, company size, proximity, role, seniority, and recency, and it is labeled inferred.
  • Activity turns guesses into evidence. Activity is what people actually do together, recorded as it happens. Meetings, email, and shared documents are examples. A shared employer stays inferred. Shipping a project together is observed. New evidence is added beside what is already there: a job change, a long silence, a contradiction, or collaboration that upgrades an inferred overlap.

Use the graph against a goal

A trust score says how strong a relationship is. It does not say whether that relationship moves a goal. The agent starts from the goal, walks the graph, and scores each path by how much it would change the outcome.

The goal impact score is how much this relationship, used this way, moves this goal. It combines the trust in the path with how directly the action serves the goal.

Goal: hire a staff engineer. Path: you, Priya, Sam. Trust is high for working together — observed, three years, a meeting note from March. Goal impact: high for a referral, low for a warm intro to an investor. Ask Priya first.

Where the evidence runs out, the agent says so instead of guessing.

The professional network, rebuilt for agents

Every professional network today is a directory plus a feed. The platform holds the network, and an algorithm decides what you see. We expect personal agents to change who holds it, in two stages.

Who holds the network

TodayThe platform

Stage 1: better toolsThe platform

Stage 2: agent-nativeEach person’s agent

How you reach someone

TodayA cold message

Stage 1: better toolsA warm path, with evidence

Stage 2: agent-nativeYour agent asks theirs

What you see

TodayA feed an algorithm picks

Stage 1: better toolsChanges in people who matter to you

Stage 2: agent-nativeAnswers to your question

What it needs

TodayNothing

Stage 1: better toolsOne person with an agent

Stage 2: agent-nativeEnough people with agents

Stage 1

Your agent works next to the networks you already use, drawing on your own data, what’s public, and the messages you send yourself. A profile might only say someone is a second-degree connection. Your agent can say the path runs through a former colleague: they worked together for three years, they haven’t spoken in two, so ask them first. That record is the graph you bring into stage 2.

Stage 2

A professional network profile only keeps a snapshot: a title, a company, a list of connections, as of the day someone last updated it. The richer record lives elsewhere. The meetings you shared, the threads you worked through, what you built together, and when it went quiet. None of that is on the profile.

An agent doesn’t have to piggyback on existing profiles. It can preserve that shared context as it happens and start building it today, so by the time agents talk to each other, the history is already there.

In an agent-native network, the agents are the network. Each person’s agent holds their own relationships, knows their history and what they’re open to, and talks to other agents on their behalf. Instead of browsing profiles and a feed, you ask a question and agents answer it. This becomes possible once enough people have an agent that knows them well. In this version, the network belongs to the people in it.

Say a founder is raising a seed round. Her agent asks around:

Founder’s agent: She’s raising a seed round. Do you know an investor whose thesis fits?

Contact’s agent: He has an active relationship with Dana. I’ll check with him before I say more.

Dana’s agent: Understood. Send who vouches for the work and why it fits, and I’ll take it to Dana.

What comes back is a conclusion and the evidence for it, not the raw history. The founder gets a credible introduction. Dana gets one pitch worth reading instead of fifty cold ones. The same exchange works for a hire or a first customer.

Agents answer for the same reason people do: reciprocity. In the real world you answer a stranger because someone you trust vouched for them, or because helping builds goodwill you can draw on later. Agents can keep that memory honestly: who helped whom, how it landed, whose introductions were worth making. Answering well builds standing, and standing is what gets your own questions answered. It isn’t a single score. Like everything else in the graph, it lives in each relationship.

It also changes what “warm” means. Today it means a mutual connection. With agents, warm is shared context built together, so a stranger with the right history can be warmer than a friend of a friend.

Every person who joins makes it easier for everyone else: one more person who can confirm a history, vouch for a piece of work, or say honestly that they can’t.

A network you hold

When everyone has an agent that knows them well, the missing piece is the record of who knows whom, and how well. Today that record is scattered across your head, your inbox, your notes, and old job titles. The Dynamic Trust Graph puts it in one place, with the evidence behind it: a referral from someone who has seen the work, an intro through someone who knows the investor, a reconnect with a real reason.

We’re starting where relationships decide the outcome: hiring, fundraising, and customers. The larger shift is in how professional networks work. When agents can carry that record and talk to each other, the network stops being a feed someone else controls and becomes something you hold, with a history behind every introduction that you can examine.

It's time for relational intelligence.

Turn your network into outcomes.

Your network already holds the opportunities. We help you find them and act on them.

No credit card required.