The consultant-marketplace layer: expert networks are turning availability into an API

The first wave of agentic integration in expert networks stopped at the transcript. Through 2024 and 2026, Guidepoint stood up an MCP server exposing its transcript library to Claude and Perplexity; GLG piped expert transcripts into Bloomberg's generative AI surface; AlphaSense-Tegus rolled its library into LLM-accessible endpoints. The content asset became machine-readable. The consultant behind the content did not.

Our read is that the next primitive up the stack is the consultant graph itself: profession, geography, employer history, project history, blackout windows, disclosure state, hourly rate, and calendar availability, all queryable as structured fields. What used to sit behind a research manager on a phone becomes an inventory an agent can filter and book against. The move from transcript-as-API to consultant-as-API is a bigger architectural change than the first wave, because it reprices the human relationship layer that traditional networks have defended for two decades.

From content API to inventory API

MCP servers that expose transcripts answer a retrieval question: given a topic, hand back what experts have said. That is a natural fit for the way agentic research tools already work. Rogo and Hebbia consume transcript endpoints in the same shape they consume filings, sell-side notes, and earnings calls. The transcript is a document. Documents are what LLM tooling was built to read.

An expert, on the other hand, is not a document. An expert is a bookable resource with legal state attached. That distinction is what makes the second layer of integration structurally different from the first. To make a consultant queryable, the network has to publish not just what the consultant has said, but what the consultant is, what the consultant is allowed to talk about right now, and when the consultant is free.

The schema fields that fall out of this are recognizable to anyone who has staffed a call:

• Identity and provenance: current and prior employers, functional role, tenure in role, geography, language, seniority band.
• Coverage tags: named companies, named products, named regulatory regimes the consultant is credentialed on.
• Compliance state: ticker restrictions, employer NDA scope, disclosure history, blackout windows tied to prior employers, cooling-off periods, MNPI-adjacent flags.
• Availability: calendar windows in the consultant's local timezone, minimum notice, session length options.
• Commercial: hourly rate, minimum billing, cancellation terms.
• Consent scope: which of the above the consultant has authorized to be exposed to third-party agents, and at what granularity.

Each of these fields exists today inside the CRMs of the traditional networks. None of them are exposed as a single queryable object to a client-side agent. The gap between "we have this data" and "an agent can filter on this data across vendors" is the entire competitive question for the next 18 months.

What the AI-native entrants are actually doing

Three companies illustrate what a consultant-as-inventory posture looks like when it is designed in from the start rather than retrofitted onto a relationship-managed book.

NewtonX has productized its panel as a structured object. The Synthetic Personas launch exposes panel structure , role distributions, geography, seniority , as a queryable surface, which is a materially different posture from the industry default of describing panel depth in a sales deck. Once panel structure is a schema, sample composition becomes something an agent can specify and validate, not something a research manager negotiates in email.

Mercor operates a talent API originally built for AI training labs and has extended into structured expert data. The relevant primitive here is that Mercor was never a call-brokerage business first. It was a matching-and-routing API first, and expert calls sit on top of that infrastructure as one workflow among several. That inverts the traditional network's stack: instead of a sales-led services business exposing an API on the side, it is an API-led business that happens to sell expert time.

Bridgetown Research goes further and runs diligence workflows end-to-end with agents, which requires that the consultant graph be not just readable but bookable inside the same loop. If an agent is composing a diligence memo, it needs to be able to identify three former APAC procurement leads at a named OEM, check compliance clearance for the relevant ticker, confirm availability inside a 72-hour window, and book. Any handoff to a human research manager in the middle of that loop breaks the workflow. Bridgetown's architecture presupposes the primitive that the rest of the industry is still building.

Mercor's expansion into structured expert data for AI training, NewtonX productizing its panel, Bridgetown running end-to-end agentic diligence: these are not three unrelated product launches. They are the same architectural bet, made from three different starting points, that consultant availability is inventory and inventory belongs behind an API.

The traditional networks' defensive question

The nine established networks , AlphaSights, GLG, Guidepoint, Third Bridge, Dialectica, Coleman, ProSapient, Capvision, and VisasQ , face a genuinely difficult sequencing problem. Their franchise value historically compounds through the research manager: the human who knows which of the 1.5 million experts on the roster is actually the right one for a given client's actual question, who has spoken to that client before, and who is currently reachable. Exposing the consultant graph as an API commoditizes exactly that layer.

Guidepoint's MCP server launch is the clearest signal so far that the incumbents understand where the puck is going. It also stops at the transcript. We read that as a deliberate first step rather than a full posture: the transcript library is the least commercially sensitive asset to expose, because clients already receive transcripts under existing contracts and because transcript retrieval does not disintermediate the moderator-and-matcher relationship. Consultant availability does.

The defensive question resolves into three broad paths, and we think each of the nine will pick differently based on segment mix:

Expose on your own terms. Publish a graph MCP with full compliance state as a first-class field, price-per-hour as a queryable attribute, and OAuth-scoped consent from the consultant on which client agents can see which fields. The upside is that the network sets the schema the market converges on and captures agentic query volume as a new billable unit. The downside is that it accelerates the commoditization of the research manager.

Expose only mediated fields. Publish an availability API that returns candidate matches without disclosing raw consultant identity, keeping the actual matching and booking inside a human loop. This preserves the moderator layer but is unlikely to satisfy an agentic workflow that needs to run without handoffs.

Do not expose, and compete on service depth. Bet that the client segments who actually pay for expert calls , long-only funds, credit desks, PE diligence teams, corporate strategy , value the moderator's judgment more than they value cross-vendor discovery. This is a defensible position for the top of the market and a weak one for the middle.

Our read is that the industry ends up with a bifurcated stack. The premium franchises hold the moderator layer for high-touch mandates and expose a graph for lower-touch, higher-volume agentic workflows. The AI-native entrants pick off the middle by offering programmatic access at a price point the traditional networks cannot match without cannibalizing their own sales motion.

Compliance as a first-class field

The hardest schema question is not calendar or price. It is compliance state.

The expert-network compliance layer exists to keep clients out of MNPI territory. Blackout windows tied to prior employers, cooling-off periods after departure from a public company, ticker restrictions, employer NDAs with named-company carveouts, disclosure obligations under the client's own compliance regime , each of these is a state that changes over time and that historically has been checked by a human at the moment of booking.

Making compliance queryable means encoding those states in a way an agent can filter on and, more importantly, in a way that survives audit. If a buy-side analyst books an expert through an agent and later has to defend that call to a compliance officer, the schema needs to have carried the compliance state at the moment of booking, not just at the moment of query. That is a temporal-audit-log problem, not a data-modeling problem, and it is a real reason why the traditional networks , who carry the compliance liability today , are unlikely to hand the graph out uncaveated.

We think this is where the standards conversation will actually happen. Transcript MCP was easy to standardize because the object is a document. Consultant MCP is harder because the object is a person with a legal state that changes. Any schema that stabilizes will need at minimum: compliance state versioned with a timestamp, consent scope from the consultant as an explicit and revocable field, and an audit trail the buy-side compliance function can pull in a subpoena.

The network that publishes that schema first and gets it endorsed by the buy-side compliance layer sets the industry standard. The network that waits inherits someone else's schema.

What changes for the buy-side workflow

The near-term consequence for a buy-side research desk is prosaic and large. Today, a research analyst who wants an expert call opens the client portal at one network, scans a few profiles, requests availability, waits, gets a proposed time, and books. If the first network cannot cover the specific angle, the analyst opens a second portal and repeats. Cross-network price comparison happens quarterly at the procurement level, not per-call at the analyst's desk.

With consultant graphs exposed as MCP, a single agent can fan out across vendors in one query. "Find me three former APAC procurement leads at [named OEM], cleared for [ticker], available in the next 72 hours, with hourly rate under [threshold], from any of AlphaSights, GLG, Guidepoint, Third Bridge, or NewtonX." The agent returns a ranked list with price, availability, and compliance state. The analyst approves. The agent books.

The emergence of pooled MCP procurement across buy-side desks is the leading indicator that this is happening. Once a hedge fund's IT function has centralized MCP credentials for four vendors behind a single internal endpoint, the marginal cost of adding a fifth vendor to the fan-out query is zero. The competitive pressure on any single network then runs through the agent, not through the sales cycle.

The vendor whose graph is deepest and whose compliance state is best-modeled wins the query, per query, in real time.

That is a materially different competitive surface than the annual RFP and the standing subscription. It rewards the network with the most experts, the freshest availability, the cleanest compliance encoding, and the most competitive price on the specific query. It punishes networks whose franchise depends on the client not knowing what a comparable expert costs at a competitor.

Three scenarios for how the layer settles

Scenario one: incumbent-led standardization. One or two of the traditional networks , Guidepoint is the natural candidate given its MCP posture , publish a consultant graph schema with compliance state as a first-class field, and the rest of the industry converges on it defensively. The incumbents retain the compliance-liability layer, which becomes their moat. The AI-native entrants integrate as additional supply on the same schema. This is the outcome most favorable to the traditional networks and most likely if the buy-side compliance function drives standard adoption.

Scenario two: entrant-led disintermediation. Mercor, NewtonX, or a Bridgetown-adjacent player publishes an open schema that the buy-side pools converge on directly, because the AI-natives are willing to expose price-per-hour as a queryable field and the incumbents are not. The traditional networks either capitulate and expose their graphs on the entrant's schema, or defend the top of the market and cede the middle. This is the outcome most likely if buy-side procurement drives standard adoption ahead of buy-side compliance.

Scenario three: fragmented graph, agent-side normalization. No standard emerges. Each network exposes its own schema. Buy-side agents normalize across schemas at query time, which pushes complexity to the client but preserves each network's control over its own data model. This is the outcome most likely if the compliance-audit problem takes longer to solve than any single vendor is willing to wait, and it is probably the base case for the next 18 to 24 months.

All three scenarios share one property: the consultant graph gets exposed. The disagreement is over who writes the schema and who captures the query volume.

What we would ask an expert next

A research analyst evaluating this thesis should be putting these questions to the industry, not to us:

1. Which network has the most complete internal encoding of consultant compliance state today, and how much of that encoding is machine-readable versus captured in free-text notes by research managers?
2. Among the AI-native entrants, whose consent-scope model has been reviewed by consultant-side counsel, and are consultants actually opting in to structured availability exposure at scale, or are the panels still primarily human-brokered?
3. What is the current price differential per hour for a comparable expert profile across the top five traditional networks, and how much of that differential survives cross-vendor discovery through a single agent?
4. Which buy-side compliance functions have already approved MCP-mediated expert booking, and what audit-trail requirements did they impose as a condition?
5. When Bridgetown-style end-to-end agentic diligence workflows book an expert, who carries the compliance liability , the client, the network, or the agent operator?