See pricing
Can AI Fix Professional Networking? | Inside Cosign
The a16z ShowAI

Can AI Fix Professional Networking? | Inside Cosign

Cosign's bet is that the most valuable career data is private: who would actually vouch for you, which companies have real talent density, and where mutual fit exists. This episode explores how awards, lists, and agent-readable reputation data could turn that hidden signal into a new professional network.

October 31, 2026 · 25 min listen · 12 min read · David Booth, Dhruv Gupta
0:00
–:––

Context

David Booth and Dhruv Gupta from a16z New Media use Cosign and the Cosign Awards to argue for a different kind of professional network. Their premise is that the most valuable career information is usually private: who would vouch for you, which companies have real talent density, and which relationships could turn into a job, collaborator, or co-founder. That matters to PMs because it is a case study in building a network product around trust signals instead of public profiles, and in preparing that network so AI agents can actually use it.

The Big Idea

Professional networking becomes far more useful when it is built on trusted reputation data, not just public resumes or follower counts. Cosign's longer-term bet is that once that trust graph exists, AI agents can do the noisy first pass of matching people, companies, and opportunities, then escalate only the connections with real mutual fit.

Key Insights

Trust is the scarce signal

Most career breaks in the episode are framed around one person willing to take a chance on someone else. Dhruv's own entry into tech came less from a formal application flow and more from Brent at MTS recognizing work he had done at Polymarket and deciding he could contribute. The product lesson is that a network becomes more valuable when it captures who would actually vouch for whom, because that signal is rarer and harder to fake than a polished profile.

Awards bootstrap the hidden graph

Cosign Awards is not just a media project. It is a data-collection mechanism for surfacing operators, companies, and products that are respected inside specific communities but may be under-recognized in public.

  • People categories are often role-based, such as engineering, product, design, AI research, comms, and growth.
  • Company categories include ideas like talent density, with employee voting treated as a signal that a company is a place talented people actually want to work.
  • The early nomination data already shows long-tail discovery. David cited 135 nominees in Rising Stars, 65 in public intellectuals, and 76 in podcast.

For PMs, the important move is turning hidden knowledge into structured input. A nomination or employee vote makes private reputation legible.

Lists add lightweight curation

David describes Cosign as having a middle layer between raw profiles and explicit endorsements: lists. That matters because not every useful signal needs a deep workflow or a heavy review process.

  • Cosign already has roughly 40,000 to 50,000 people in the graph.
  • On top of that, about 400 user-created lists act as lightweight curation.
  • David's example was Paki McCormick's list of interesting companies. The value is not that the list is comprehensive. The value is that someone with judgment spent time selecting it.

This is a useful PM pattern. Before building a full reputation system, create a low-friction way for credible users to organize and annotate the graph.

Search improves, relationships stay scarce

David makes an important distinction: better search over public data is coming anyway. Better models, better databases, and more agent tooling will make it easier to analyze the same public internet everyone else can access. That means search alone is a weak moat.

The harder problem is collecting the relationship data that public systems miss: who helped whose career, who would make an introduction, which people would work together anywhere, and which companies actually have the right cultural or mission fit. For a PM, the lesson is that differentiated data often comes from participation, not scraping.

Agents can pre-negotiate fit

The most forward-looking idea in the episode is not AI search. It is agent-to-agent matching. David imagines a job seeker's personal agent talking to a company's agent, or one person's agent talking to another person's agent, to test whether there is a real mutual fit before humans spend time.

The proposed flow is:

  • A personal agent knows the user's goals, skills, and preferences.
  • A company or hiring agent knows the role, team, and what good fit looks like.
  • The agents do a first pass on whether there is a genuine coincidence of intent.
  • Only then does the system elevate the opportunity to a hiring manager or make an introduction.

That is a strong PM pattern for any matching product. Use AI to reduce low-value interactions, not to automate the entire relationship.

Better filtering beats bigger inboxes

Dhruv argues that today's networking tools fail partly because the inbox experience is basically the same for everyone. High-value people, junior users, and even teenagers all receive outreach through nearly the same blunt interface. The result is noise, not discovery.

Cosign's thesis is that a network should preserve signal by filtering aggressively before a message or opportunity reaches a person. For PMs, this is a reminder that a marketplace or network product is often won by attention management, not by maximizing raw connection volume.

Mental Models & Frameworks

Surface area for luck

David reframes the purpose of networking as increasing your surface area for luck. You cannot plan the exact moment you meet a future co-founder or the person who changes your career, but you can build systems that make those collisions more likely. Use this model when deciding whether a feature helps people discover high-value opportunities or merely adds more noise.

Three-layer graph model

Cosign's product logic has three layers:

  • Profiles: the base graph of people, companies, and relationships.
  • Lists: lightweight curation that helps users navigate the graph through someone else's judgment.
  • Cosigns: deliberate endorsements that say a person is worth trusting, hiring, introducing, or tracking.

This is a useful framework for PMs building network products. Raw entities create coverage. Curation creates usability. Explicit endorsements create trust.

Coincidence of intent

David uses the phrase coincidence of intent to describe the moment when two people have aligned goals plus complementary skills or needs. The point is not to connect everyone. The point is to detect when a connection is actually worth making. Use this model when designing recommendations, introductions, or agent workflows. A smaller number of high-fit matches is often better than more outreach.

Agent-first networking flow

Dhruv and David sketch a future workflow for agent-native networking:

  • The user contributes profile data, lists, and signals about what they want.
  • The network exposes that structured graph through an agent interface, including MCP access.
  • A personal agent searches and reasons over the graph on the user's behalf.
  • The system escalates to a human only when the agent finds a strong match.

This framework is useful when thinking about where AI belongs in a network product. AI is not the product by itself. AI is the layer that navigates a high-quality graph.

Decision Principles

Principle: Build on trust

  • When: choosing which user actions or data fields to treat as core signals in a hiring, networking, or marketplace product.
  • Why: public profile data is easy to copy. A nomination, introduction, employee vote, or explicit endorsement is harder to fake because it puts the endorser's reputation at risk.

Principle: Curate before automation

  • When: adding AI search, recommendations, or agent workflows to a product with messy data.
  • Why: agents only get meaningfully better when the underlying graph has structure and judgment baked into it. Cosign's lists and awards are ways of creating that structure before full automation.

Principle: Filter before introducing

  • When: a product creates too many low-quality messages, applications, or connection requests.
  • Why: Cosign's agent vision uses AI for first-pass screening, then escalates only the matches with real mutual fit. That protects scarce human attention on both sides.

Trade-offs & Nuance

Openness can dilute signal

Allowing anyone to nominate or vote increases discovery and helps uncover under-recognized talent. The trade-off is that open participation can weaken trust if every signal counts equally. Cosign tries to balance this by combining community input with judges and with stronger signals like explicit cosigns.

Public data lacks intent

Public internet data is useful for search and broad discovery, and David expects that layer to keep improving fast. The limitation is that public data rarely tells you who would actually back someone, hire them, or collaborate with them. A PM should see this as a trade-off between easy scale and harder, higher-value data collection.

Agents save time, not judgment

The episode does not argue that agents should replace human choice. The proposal is narrower: agents should handle the noisy first pass. People still matter at the moment of introduction, hiring, or collaboration. This distinction matters because a fully automated networking product could optimize for activity while degrading trust.

Utility depends on onboarding

David is explicit that Cosign is still early and has rough edges. Dhruv says smoother onboarding is one of the biggest pieces of feedback. For PMs, that is a useful reminder that a graph product cannot rely only on ambitious future vision. If onboarding is weak, the graph grows slowly and the later AI layer has less to work with.

Common Mistakes

Mistake: Chasing visibility over trust

A big public profile can help, but the episode argues that career outcomes often come from a smaller and deeper signal: one credible person willing to make an introduction or take a chance. Products that optimize only for visibility risk amplifying popularity while missing actual fit.

It is tempting to think the networking problem is just a better query box over public data. David's argument is that search quality will keep commoditizing. The harder and more defensible problem is collecting data that never becomes public unless users deliberately contribute it.

Mistake: Giving everyone one inbox

Dhruv's critique of current networks is that they force very different users through almost the same outreach funnel. That creates message overload for some users and poor response rates for everyone. A better design filters and ranks attention before the inbox stage.

Practical Application

Audit your trust signals

List the places where users already vouch for each other inside your product or workflow. Examples include referrals, repeat collaborators, employee advocacy, invited introductions, or public nominations. If those actions matter in real decisions but are not first-class product data, you are likely missing your highest-value signal.

Productize lightweight curation

Before building a complicated reputation score, add a simple way for credible users to create structured curation. A list of companies worth joining, a shortlist of operators to watch, or a tag for people a user would work with again can reveal judgment without requiring a heavy review workflow.

Define agent escalation rules

If you plan to add agents to a matching or recruiting flow, decide in advance what the agent can do alone and when a human must step in. Cosign's proposed role for agents is to search, compare goals, and test fit early. The human takes over when the connection is strong enough to deserve attention.

Celebrate operators, not founders

Recognition mechanics can do more than create buzz. Cosign Awards focuses on operators, products, and companies because that is where hidden information lives. If you run a community or ecosystem product, reward the contributors whose work is obvious to insiders but invisible to outsiders.

Smooth onboarding first

A network product only compounds if new users can quickly understand what to do and why contributing improves the graph. Dhruv says onboarding is a major focus because better downstream matching depends on better upstream participation. Fix the first-run experience before adding more surface area.

Questions to Consider

  • Where in your product does a real decision depend on private trust, such as who would make an introduction or hire someone, even though the product mainly captures public profile data?
  • If an AI agent searched your product today, what structured signals would it find besides text fields and resumes, and would those signals be good enough to judge fit?
  • Where is your team increasing connection volume when the better move would be filtering for mutual intent before a message reaches a human inbox?
  • What lightweight contribution mechanic, such as nominations, lists, or employee votes, could turn hidden knowledge inside your ecosystem into searchable product data?

Bottom Line

Cosign's core insight is that professional networking gets much better when the product captures trusted reputation signals instead of stopping at public profiles. The AI angle only works because of that foundation: agents can filter, search, and negotiate fit only after a high-quality relationship graph exists.

Tools & Products

Tool / ProductWhat it doesWhy it was mentioned
CosignA professional network built around profiles, lists, cosigns, and agent accessIt is the core product discussed, and the episode explains the data and interaction model behind it.
Cosign AwardsA nomination and voting system for operators, companies, and productsIt acts as both recognition and a way to surface hidden reputation signals that can feed the broader graph.
LinkedIn and HandshakeMainstream networking and job application platformsDhruv uses them as the contrast case for cold applying and blunt, noisy outreach flows.
Muse, Grok, and InstinctPersonal AI agents or interfaces that can search on a user's behalfDhruv says Cosign wants to integrate with this emerging agent layer so users can query the graph from their preferred assistant.
MCPA protocol for giving models and agents structured access to tools and dataDhruv argues agent-readable access to the relationship graph is a key differentiator that traditional networks are unlikely to expose.

People to Follow

David Booth

David Booth is part of a16z New Media and one of the main voices behind Cosign's product thesis. He is especially useful to follow if you care about network products, reputation systems, and how AI agents may change discovery and matching.

Dhruv Gupta

Dhruv Gupta is part of a16z New Media and has been closely involved in building Cosign. He brings a practical builder view on agent-native networking, onboarding, and why trusted referrals often beat cold applications.

Notable Quotes

"I hope it's a place where you can increase your surface area for luck." (David Booth)

"How do you make two people who have the same hopes, dreams and complementary skills come into each other's orbits sooner?" (David Booth)

AI PM course

Everyone hears the same episodes.
Few can do what they describe.

Start for free