Agents can now do anything a human does on a computer. More and more knowledge work will keep moving to them.
But human relationships will continue to decide deals, hires, and investments. Critical business outcomes will still depend on people trusting each other.
Part of that trust comes from knowing someone chose to invest their time in you. No AI agent can automate that choice. Genuine human attention will remain scarce and become more valuable as more work is automated.
Deciding who deserves your attention becomes an increasingly important part of professional judgment. The question shifts from “what task deserves my time?” to “who should I spend my time with?”
Which customer needs a conversation? Who could help us reach the right buyer? Which professional relationship have we neglected?
noticed is an applied research team focused on one core question: Who deserves your attention?
We’re building AI that understands who can help you and what kind of engagement is appropriate.
Why agents need noticed
Frontier agents like Grok Bot, Claude, or Muse will always outclass noticed at most tasks. They have more resources, broader autonomy, persistent memory, better tool use, and access to similar data.
noticed does not need to beat them at being an agent. We need to be the specialized intelligence they call when they need to understand professional relationships or decide who deserves attention:
- Who should I talk to about this?
- Does anyone here genuinely know this person?
- Is this an appropriate ask?
- Who are we neglecting?
Where others fall short
Our bet is that a general agent can have access to all your accounts and still struggle to identify who matters most and what the next step should be. We need to prove that this is true.
Access to interaction data is not the same as understanding a relationship. Emails, meetings, and social connections can show that two people have interacted. They do not reliably reveal whether they genuinely know each other, how much trust exists, what they would be comfortable asking of each other, or whether the relationship is relevant to a specific goal.
General agents are not optimized to understand how professional relationships evolve, judge what is appropriate to ask, and learn from whether those judgments lead to customers, hires, investments, or partnerships. That is the capability noticed is focused on building.
Closing that gap requires three capabilities: identity, relationship judgment, and attention ranking.
What noticed needs to master
Helping someone win a customer requires understanding who they know, what those relationships mean, and who could help with that specific goal.
There are 3 core capabilities:
- Identity — who is this person? Resolve the same person across email, calendar, social profiles, and company data. Enrich their profile and keep it current.
- Relationship judgment — what does this relationship mean? Do these people genuinely know each other? How much trust exists? What would be appropriate to ask? Shared meetings can suggest a connection without revealing whether either person would make an introduction.
- Attention ranking — who matters now? Given a professional goal and relationship context, identify and rank the people who deserve attention, including relationships we have neglected and people who should receive less of our time.
We’re working to become the best in the world at identifying people, understanding their relationships, and knowing who deserves their attention.
Better at building professional relationships
We’re testing three ways to help people build better professional relationships: better context, better use of existing models, and specialized training.
- Better context means bringing together the relevant evidence about a person and relationship across sources, including what interaction history alone cannot reveal.
- Better harness means improving how the system selects tools and assembles context to answer relationship questions. We need to test whether those choices produce better judgments and recommendations.
- Better intelligence means training specialized models on proprietary relationship judgments and outcomes, then measuring whether they outperform the alternatives on relevant benchmarks.
Our research will test how far existing models can go with better context and tools, and whether specialized training produces better relationship judgments.
Learning from human judgment
A person can tell us things their inbox cannot: “This is a close friend,” “I would not ask them for this,” or “This is the wrong person.”
We then track which recommendations people act on and whether those actions lead to replies, meetings, or deals.
Together, these corrections and outcomes connect relationship state → decision → action → outcome. This is the proprietary data that can improve our context, ranking, and models.
To learn from real relationship decisions, we’re building noticed alongside a small group of Research Partners.
Building with Research Partners
noticed isn’t publicly available yet. We’re building it alongside Research Partners: early-stage B2B founders and their teams who use noticed to find sales opportunities in their networks.
Partners work with a dedicated expert to identify relevant prospects, decide how to reach them, and review the results. Their feedback gives us real relationship decisions and outcomes to learn from.
This work keeps us focused on three priorities:
- Build relationship intelligence. We are building and testing specialized intelligence, not a general agent or a SaaS business. Our product is how we test that intelligence, collect feedback, and prove its value.
- Capture human judgment. Design every product interaction to capture corrections, decisions, and outcomes. Turn that feedback into proprietary data that improves our understanding of professional relationships.
- Build benchmarks. Measure progress in identity resolution and relationship search and ranking, with the goal of becoming the best in the world at these core capabilities.
If you want to find opportunities in your team’s network and help shape this research, become a Research Partner.