A new software layer is forming between AI assistants and the go-to-market stack. Most of it appeared in the last two quarters, and most people who run go-to-market are just becoming aware of it. This is a map.
Current as of July 2026. A category this young moves fast; parts of this will be out of date by autumn, and we'd rather say so now than pretend otherwise.
Something happened
For about two years, "AI in go-to-market" meant a model writing copy. Subject lines, opening sentences, a personalized hook stapled onto a template. The model produced text, and then a person — or another piece of software — did something with it.
What changed is that AI models became able to operate the tools directly. Not draft the email: pull the list, fill in the missing contact details, check that the email addresses are real, write the records into the CRM, start the outreach sequence. The mechanism behind this is a protocol called MCP — the Model Context Protocol, a standard way for an AI model to connect to outside software, roughly what USB is for plugging in devices. Its practical effect is that a general-purpose assistant like Claude became a plausible interface to an entire go-to-market stack.
The cleanest way to understand what's forming as a result:
A person doing go-to-market work is an operator. They bring two things: judgment about what to do next, and logins to a dozen different systems. An AI agent already has the judgment. What it's now acquiring — through the layer this piece maps — is the logins, and the expertise to orchestrate many tools at once.
Software vendors moved quickly. Within roughly six months, most of the go-to-market software category shipped MCP connectors for their products. And then a second group of companies appeared — companies whose entire reason for existing is to sit between the model and those connectors: routing requests across data sources, holding credentials, imposing consistent shapes on the data, metering spend.
That second group is what this guide maps. It's new enough that it has no settled name — several are in circulation and none has won, which is itself the clearest evidence of how recent this is. We'll call it the GTM agent layer.
The timeline, because the compression is the story
You don't have to take "new" on faith. Here are the dated events, nearly all inside five months:
- February 2026 — Apollo ships a native MCP connector in beta: search, enrich, create and update contacts, add to sequences. Free on every plan, including the free tier.
- February 19, 2026 — HubSpot's Developer MCP Server reaches general availability, letting any software vendor build connectors into HubSpot's agent layer.
- April 13, 2026 — HubSpot's remote CRM MCP server reaches general availability with read and write access, respecting existing user permissions.
- June 1, 2026 — ZoomInfo makes GTM.AI generally available and describes it as "the headless GTM context layer" — the category's largest incumbent adopting the new vocabulary.
- June 30, 2026 — HubSpot signs an agreement to acquire Warmly, folding person-level intent data and agents into its own stack.
- July 7, 2026 — ColdIQ, a $7M-revenue outbound agency, launches a unified API for running go-to-market work through Claude Code. Its founder steps back from the agency to build it.
- July 9, 2026 — ZoomInfo ships a GTM.AI command-line tool: open source, installable with one command.
Around and between those dates: first-party MCP connectors from Clay, Common Room, Crustdata, Exa, Firecrawl, HeyReach, Hunter, Instantly, Octave and Salesforge, plus Clay's Agent Plugin, which bundles its API, command-line tool, connector and skills into one package for coding agents.
One observation, and only one: software categories do not usually assemble themselves in two quarters. When one does, it's worth understanding early — and it's also too early to know which of these shapes survive.
What it's actually like, and where the seams show
Here's the experience, concretely. You connect four MCP servers to Claude: a data provider, your CRM, your email sequencer, an email finder. You describe what you want in a sentence — fifty heads of revenue operations at Series B fintechs, real email addresses, into HubSpot, into the Q3 sequence. It works. Twenty minutes instead of an afternoon. The first time you see it, it's startling, and that reaction is correct: this capability did not exist eighteen months ago.
Run the same thing weekly against live customer data, though, and the experience gets uneven — in specific, repeatable places. Those places are worth cataloguing, not because they're damning but because the map in the next section is largely a map of them. Nearly every company in this space exists to smooth one of these seams.
Where the model has to hold too much
Tool volume. Every connector brings its own set of tools, and a few real ones put hundreds of options in front of the model before any work starts. For a sense of scale, one vendor in this space exposes more than 700 endpoints across more than 40 data sources behind a single surface. Grown up around this seam: routers that load tools on demand, and "verb" layers that offer a model a small vocabulary instead of a large one.
Shape mismatch. A "person" fetched from four different data sources is four different objects — different field names, different freshness, different ideas about where someone currently works. Left alone, the model reconciles them on the fly, in prose, a little differently every run. Grown up around this seam: aggregators and verb layers that impose one shape on many sources.
Improvised sequencing. Find, fill in, check, de-duplicate, write: that's a pipeline. With plain connectors, the pipeline gets reinvented from scratch on every run, because nothing holds it. Grown up around this seam: the workflow and "play" constructs now appearing across several categories.
Where the calls don't compose
No cross-vendor fallback. When one data source has no email address for someone, its connector returns nothing — it will never route you to a competitor. This is structural, not an oversight: no vendor's own tooling is going to hand your request to a rival. Grown up around this seam: the entire waterfall-aggregator category, which exists for precisely this.
"Nothing found" is ambiguous. This one is the most concrete and the least discussed. The plain-English version: when a lookup comes back empty, the agent usually can't tell which of four very different things just happened.
| What actually happened | What it implies |
|---|---|
| The source looked and found nothing | Move on to the next source |
| The source doesn't offer this capability at all | Stop asking this source, permanently |
| The source can't handle this particular input | Reshape the request and retry |
| The call simply failed | Retry as-is |
Each row calls for a different next move, and a plain connector surface generally doesn't distinguish them. An agent that can't tell these apart retries things that will never work and abandons things that would have.
Writes are hard to make safe. Pushing a contact into a CRM or starting someone on a sequence touches live business state, and there is no undo. A retried call can mean a duplicate record, or a prospect receiving the same campaign twice. No shared convention exists yet for making agent-driven writes safe — this one, notably, has no category to point to. It comes up again at the end of this piece.
Where accountability lives
Spend crosses billing systems. Every vendor prices in its own currency, and the units don't compare: one separates data credits from action credits, another charges several times more for a phone number than for an email address. One vendor's most-cited customer complaint is that automated runs consumed credits in ways that were hard to forecast — a specific case of the general problem, which is that an agent working across six billing systems has no overall ceiling. Grown up around this seam: per-result pricing and spend caps, now appearing in several categories.
Connections don't have customers. An MCP connection is one account. If you run work for a dozen clients — as every agency and fractional operator does — that's a dozen separate configurations, or worse, one configuration and a prayer that nothing crosses between clients. Partially addressed, genuinely unsolved; it appears again in "What's still moving."
Provenance ends at the vendor's edge. Which source answered this? How fresh was it? What did it cost? Several vendors answer these questions well inside their own walls — attributing sources field by field, annotating why each result matched, showing which providers were tried. But the trail stops at the vendor boundary. Across a multi-vendor run, nobody owns it. Grown up around this seam: usage ledgers and audit surfaces, mostly in the newer categories.
The map: ten categories
The single most useful framing: these categories are mostly not competitors. They're different altitudes. Most companies here own one layer cleanly and hand off to the layers above and below — and the interesting activity right now is in the handoffs.
The plain-language summary, before the table: at the bottom sit companies that sell facts; above them, companies that route across many fact-sellers; above those, the places where people or agents compose work; alongside, the systems that send messages and the systems that remember state; and at the top, the newest arrivals — thin layers built specifically so that agents can use everything below.
| # | Category | What it is | Typical billing | What it leaves to others |
|---|---|---|---|---|
| 1 | Point data providers | Sell one class of fact — an email, a mobile number, a job posting, a technology detection | Per credit or per result | Everything outside their one fact |
| 2 | Waterfall aggregators | One connection over many sources, routed internally, usually charging only when a result is usable | Per usable result | Actions, CRM state, per-customer isolation |
| 3 | Orchestration workspaces | Spreadsheet-style interfaces where a person composes data and logic | Credits, often in two currencies | Repeatability without a human in the loop |
| 4 | Sequencers and sending infrastructure | Outreach campaign engines, and the mailboxes, domains and warmup beneath them | Per seat, per mailbox, per send | Data quality and targeting |
| 5 | Systems of record | CRMs and product-analytics platforms — where state lives | Per seat or per event | Acquiring the data that fills them |
| 6 | Context layers | A company's ideal customer, personas, offerings and signals, served to agents | Platform fee plus usage | Execution, and multi-source data |
| 7 | AI SDRs and closed-loop platforms | Sell booked meetings rather than infrastructure | Per seat or per outcome | Composability — closed by design |
| 8 | Agent plumbing | Gateways, tool routers and managed logins across many SaaS tools | Per tool call | Domain knowledge — no GTM-specific logic |
| 9 | Unified verb layers | Plain-language verbs — find, enrich, verify, sequence — routed across sources | Per usable result | Being anyone's system of record |
| 10 | Skills, plays and operator content | Prompts, procedures and playbooks, largely free and open | Usually nothing | Any data plane at all |
(A visual version of this map accompanies the piece. The full company inventory is in the appendix; the sections below name representatives.)
1. Point data providers
Where the data actually comes from. Each of these companies is excellent at one class of fact: Findymail and Prospeo find email addresses, ZeroBounce and MillionVerifier verify them, LeadMagic and Cognism find phone numbers, People Data Labs sells the underlying person-and-company graph, TheirStack reads hiring signals out of job postings, Wappalyzer detects what technology a website runs. There are roughly forty companies in this band in our inventory alone.
The detail that explains the categories above them: most of these companies have no agent connector of their own. The data an agent most needs tends to live behind plain APIs built for developers — which is precisely the gap the next category, and category 9, grew into.
2. Waterfall aggregators
One connection, many sources. You ask for an email address; the aggregator tries its sources in sequence — a "waterfall" — and returns the first usable answer. BetterContact, Databar.ai, FullEnrich and Waterfall.io are built on this pattern; Explorium and Crustdata offer agent-native data APIs over many upstream sources; Exa turns plain-language descriptions of a target market into verified lists.
This category converged fast on one commercial idea: charge only when the result is actually usable. It's now a norm rather than a differentiator — though vendors define "usable" slightly differently, which is one of those small unsettled details that tells you a category is young. Also characteristic of this space: several of these companies are simultaneously sources inside other companies' waterfalls and destinations in their own right. The layering has not settled.
3. Orchestration workspaces
Spreadsheet-like surfaces where a person composes enrichment, logic and outreach: Clay — the category's center of gravity, and per the 2026 State of GTM Engineering Report the tool used by 84% of GTM engineers — plus Bitscale, Cargo, Persana AI and SyncGTM. This is where the "GTM engineer" job title formed, and where most multi-source work actually happens today.
The relevant boundary: these are built around a person driving. All of them are now adding agent access — Clay's Agent Plugin is the clearest example — which makes this category the incumbent neighbor of the agent layer rather than a bystander.
4. Sequencers and sending infrastructure
The tools that actually send. This category is really three businesses that outsiders routinely conflate: the sequencer (campaign logic, reply handling — Instantly, lemlist, Salesforge, Smartlead for email; Expandi, HeyReach and Unipile for LinkedIn and multichannel), the mailbox and domain provisioning underneath it (Mailreef, Hypertide, Maildoso and a half-dozen others — yes, provisioning sending mailboxes is its own industry), and warmup and placement testing (Warmforge, Warmy) on top. If you've ever wondered why cold email involves so many vendors, this is why.
For the agent layer, this category matters as the place where actions become irreversible: a send is a send.
5. Systems of record
Where state lives: HubSpot, Attio, Pipedrive, folk and Salesforce on the CRM side; PostHog, Amplitude, Segment and RudderStack on the product-analytics side. They appear on this map for two reasons. First, they're the destination of most agent-driven work — the list ends up somewhere, and it's usually here. Second, the large ones are becoming platforms for the agent layer rather than just endpoints of it: HubSpot now operates its own read-write agent connector and a developer program for third parties to build into its agent ecosystem, and acquired an AI-native intent company this June.
6. Context layers
The newest incumbent-blessed idea: serve a company's situation — its ideal customer profile, personas, offerings, messaging, signals — to agents, headlessly, so that any agent working on its behalf starts informed. Octave built a standalone business on this; Common Room serves identity-resolved signals with enterprise governance; CaliberMind runs a governed query layer over marketing data; and ZoomInfo now describes its whole agent-facing offering as a context layer.
Four companies arriving at the same phrase within a year is worth noticing. In the operator terms this piece opened with: category 1 supplies facts, category 6 supplies judgment's raw material.
7. AI SDRs and closed-loop platforms
11x, Amplemarket, Artisan, Copy.ai GTM, Landbase, Regie.ai, Relevance AI, Unify. These sell the outcome — meetings booked, pipeline generated — rather than infrastructure. They're included on the map mostly so you can tell them apart from everything else: they consume the same data sources and sending infrastructure the rest of the map does, but they're closed by design. You don't compose them with other tools; you hand them the motion.
Neither better nor worse — a different bet, on outcomes over composability.
8. Agent plumbing
The general-purpose neighbor: Composio, Klavis, Pipedream (now Workday-owned), Scalekit, and the community ecosystem around MCP server directories. These solve agent-to-software connection generically — managed logins, tool routing, hundreds of app integrations — with no go-to-market knowledge at all. No waterfalls, no notion of a verified email, no idea what a sequence is.
One economic observation from this category travels well: pure routing has become very cheap per call. Whatever value accrues in the agent layer, it isn't accruing to the act of passing requests along.
9. Unified verb layers
The newest category, the smallest, and the one that most directly answers the seams in the previous section. The idea: instead of exposing an agent to hundreds of vendor-specific endpoints, expose a handful of plain-language verbs — find people, enrich a company, verify an email, add to a sequence, update the CRM — and resolve each verb across many sources behind the scenes, charging only for usable results.
Because it's this new, the honest way to describe it is by capability rather than by company. Four capabilities define the category: routing one verb across many sources; composing verbs into repeatable workflows; isolating work per customer (credentials, budgets, and results that never cross between clients); and carrying per-customer context into every call. As of this writing, no vendor has all four fully shipped. ColdIQ, launched out of a large outbound agency in early July, has verb routing live in production with workflow and context features announced as coming. GrowFlex — the company whose team wrote this guide; see the disclosure at the end — is building from the multi-customer-isolation end. Others, including some category-2 aggregators, are converging on the same territory from below.
Expect this category to look different by winter. That's not a hedge; it's the observable trajectory.
10. Skills, plays and operator content
The knowledge layer, and it's largely free: GTM Skills ships an open-source agentic GTM toolkit under an MIT license; the sales-skills project distributes agent skills for CRM, outbound and enrichment via a one-line install; and a wide band of agencies and operators — a dozen and counting in our inventory — publish playbooks, guides and small tools for agent-driven GTM. This is where the craft knowledge lives, and its price anchor is zero.
In operator terms: category 10 is the training manual. It's necessary, it's abundant, and on its own it operates nothing.
How to read the map: five axes
Categories tell you where something sits. But this space adds companies weekly, and a static list ages fast. What lasts longer is a set of questions — five axes that will place a tool this piece has never heard of.
Altitude. How close to raw data versus how close to the agent? Take one request — "find me an email address" — through three altitudes: ask Hunter and you're querying one source directly; ask FullEnrich and you're triggering a waterfall across fifteen; ask a verb layer and you're invoking find email with no opinion about sources at all. Same question, three altitudes, different tradeoffs in control, coverage and cost.
Who owns the logins. Do you bring your own accounts to each underlying source, or does the vendor pool its own accounts and resell access? This single distinction explains most of the pricing differences in the space, and most of what newcomers find confusing about it. Bring-your-own means your contracts, your rates, your data relationships — and your setup burden. Pooled means one bill and instant start — at a markup, on the vendor's terms.
Interface posture. Built for people, for developers, or for agents? A useful tell: is the API the product, or an upsell tier? Companies built agent-first tend to have thin or nonexistent user interfaces and excellent machine-facing documentation; companies retrofitting agent access onto a UI product tend to show the seams.
Billing unit. Per seat, per credit, per usable result, per connected account, or flat. The unit tells you what the vendor believes it's selling: seats mean people, credits mean consumption, usable results mean outcomes, connected accounts mean infrastructure. When a vendor's billing unit and its marketing disagree, trust the billing unit.
Open or closed loop. Does it compose with other tools, or does it own the whole motion and sell the outcome? Neither is wrong. But mistaking one for the other is expensive in both directions.
A worked example, because the method matters more than the list: Apollo is the tool that most confuses people trying to categorize this space, and the axes resolve it. It sits at three altitudes at once — its own database (category 1), a workspace (category 3), a sequencer (category 4). You bring your own account, and its data is also resold inside other vendors' waterfalls. It's UI-first with a real API and a free agent connector. It bills per credit, with phone numbers costing a multiple of emails. And it's open-loop. No single category holds it; five axes place it precisely.
What's still moving
Anyone who tells you this space is settled is selling something. Five things are genuinely unresolved.
Running many customers through one setup. Agencies and fractional operators — a large fraction of the people doing this work — run many clients through one toolset, and connector-shaped access has no real notion of whose account is whose. Several vendors have partial answers: per-client workspaces, multi-workspace patterns, white-label tiers. Nothing has settled, and it's the open problem with the most money waiting behind it.
Safe writes. There is no shared convention for what happens when an agent retries a write against a live CRM or a running campaign. Every vendor solves it locally, differently. Until some convention emerges — "writing the same thing twice is safe and reports success" is the obvious candidate — agent-driven writes to production systems will stay the place where careful teams put a human back in the loop.
Caching and data licensing. If a vendor fetches a fact for one customer, may it re-serve that fact to another? That's a licensing question, not a technical one — and for personal data it carries privacy obligations (deletion, opt-out, subject-access) that most vendors in this space are not currently built to meet. Expect this to matter more as the space grows into it.
Supply risk on scraped sources. A meaningful share of the data in this space originates from scraping professional networks. In 2025, a well-known provider in that class was shut down by litigation, and enforcement continues. This is stated as a fact about the environment, not a criticism of any vendor: anyone building on scraped professional data is building on contested ground, and buyers should know which of their sources sit in that class.
How much of this the models absorb. The honest question hanging over the whole map: as models get better at holding context and orchestrating tools, how much of this layer do they simply absorb? Some of it, plausibly — the seams in the "model has to hold too much" family are partly artifacts of today's limits. But the other two families are different in kind. No improvement in the model changes the fact that one vendor's connector won't hand your request to a rival, that spend crosses billing systems nobody jointly owns, or that a connection has no notion of whose customer is whose. Those follow from who owns which tool. We decline to predict where the line lands; we're confident there is one.
What to watch
Four things whose movement will tell you this space is resolving. Whether the naming converges — right now every vendor describes the layer in its own vocabulary, and shared language is usually the first sign a category is real. Whether the seams get standardized — an emerging convention for "nothing found," or for safe retries, would change the texture of everything above it. Whether the incumbents absorb the layer or it stays independent — the CRM platforms and the data giants are both moving, and this decade's GTM stack was largely shaped by which layers the last set of incumbents swallowed. And whether multi-customer operation gets a real answer, because the agencies are already here and improvising.
This is a snapshot of July 2026, taken while the subject was moving. Parts of it will be wrong by autumn. That's what a two-quarter-old category looks like — and it's exactly why a map is worth having now, while the territory is still small enough to fit on one.
Appendix: the inventory
More than a hundred companies, placed by primary business, with links. Alphabetical within each group, never ranked. Representative and current as of July 2026, not exhaustive — this space adds companies weekly. Descriptions are one line and deliberately plain; where a claim is a vendor's own, we've either attributed it or left it out.
1. Point data providers
Email finding — Datagma (email and phone lookup, common inside waterfalls) · Dropcontact (algorithmic email finding, GDPR-centered, strong EU coverage) · Findymail (verified-email finding, charges only for deliverable results) · Forager (email data source appearing in aggregator waterfalls) · Hunter (long-running domain search, email finding and verification) · Icypeas (low-cost email finding and verification) · Prospeo (email finding, including LinkedIn-URL-to-email lookup) · Wiza (LinkedIn-oriented contact extraction)
Email verification — MillionVerifier (bulk verification at the category's price floor) · NeverBounce (verification, ZoomInfo-owned) · ZeroBounce (real-time verification suite)
Phone — Cognism (verified mobile numbers with compliance screening, strong in Europe) · ContactOut (personal emails and mobiles, recruiting heritage) · Kaspr (lightweight LinkedIn-oriented contact lookup) · LeadMagic (pay-per-result contact finding with a mobile emphasis) · Lusha (direct-dial and contact data, seat-based)
People and company data — BlitzAPI (flat-rate LinkedIn-sourced contact data) · CompanyEnrich (low-cost company enrichment, charged on success) · EnrichLayer (LinkedIn profile and company lookup by URL) · Extruct AI (company research with per-field source attribution) · Linked API (programmatic LinkedIn account operations) · People Data Labs (bulk person-and-company data graph; the build-your-own primitive) · ScrapIn / Reverse Contact (real-time LinkedIn profile data) · Sumble (tech-stack and project intelligence mined from job postings)
Signals and intent — 6sense (account-level intent, ABM heritage) · Bombora (publisher co-op topic intent, enterprise) · Clearbit, now Breeze Intelligence (enrichment, part of HubSpot) · Demandbase (ABM platform with intent data) · G2 (buyer intent from software-review activity) · LinkUp (job postings indexed directly from employer sites) · PredictLeads (multi-signal company intelligence — jobs, news, financing, technographics — built for agents) · Signado (LinkedIn engagement signals gathered on its own infrastructure, never the customer's account) · TheirStack (job-posting search for hiring signals and job-derived technographics)
Technology detection — BuiltWith (deepest historical technology index) · Wappalyzer (live website technology detection)
Web search and extraction — Firecrawl (web crawling into LLM-ready formats) · Jina (URL-to-markdown reading and embeddings) · Serper (low-cost Google results API) · Tavily (search-and-answer API for AI agents)
Ads and SEO — DataForSEO (pay-as-you-go SEO, SERP and ads data) · Meta Ad Library (free archive of active Meta ads) · SimilarWeb (traffic intelligence) · SpyFu (competitor PPC and SEO history)
2. Waterfall aggregators
BetterContact (email and mobile waterfall over 20+ sources; catch-all emails not charged) · Crustdata (real-time people and company data with webhook alerts; also a source in others' waterfalls) · Databar.ai (100+ providers behind one subscription; charges only for successful requests) · Exa (neural web search; turns a plain-language customer profile into verified lists; also a source for others) · Explorium (agent-native data API split into fetch, enrich and events) · FullEnrich (email and phone waterfall across 15+ providers) · Waterfall.io (single-API waterfall across dozens of vendors)
3. Orchestration workspaces
Bitscale (spreadsheet-style enrichment with credit rollover) · Cargo (revenue orchestration built on agents, plays, reusable tools and unified data models) · Clay (the category's center of gravity: 150+ providers, plus an Agent Plugin bundling API, CLI, connector and skills) · Persana AI (prospecting tables with waterfall enrichment and autopilot agents) · SyncGTM (waterfall enrichment plus signals and workflows, agency-oriented)
4. Sequencers and sending infrastructure
Email sequencers — Instantly (high-volume sequencing with a large self-serve base and built-in lead database) · lemlist (multichannel sequencing, strong personalization, EU presence) · Salesforge (sequencing inside a vertically integrated stack — its own inboxes, warmup and one MCP over all of it) · Smartlead (API-first sequencing, unlimited mailboxes, white-label client workspaces)
LinkedIn and multichannel — Expandi (LinkedIn automation on the customer's own account, with safety limits) · GTM API (new unified API over LinkedIn, inboxes and messengers) · HeyReach (LinkedIn sequencing built for agencies, with a campaign API and MCP) · Unipile (unified messaging API across LinkedIn, email, WhatsApp and calendars, built for embedding)
Mailbox and domain provisioning — Hypertide (automated inbox provisioning at the category's price floor) · InboxKit (US-IP Google and Microsoft accounts) · Infraforge (private sending infrastructure with API and white-label) · Maildoso (volume-tiered mailboxes) · Mailforge (shared cold-email infrastructure) · Mailreef (dedicated SMTP servers with a provisioning API) · Primeforge (Google and Microsoft mailboxes on US IPs) · Zapmail (Google and Microsoft mailbox provisioning)
Warmup and placement — Warmforge (warmup plus inbox-placement testing) · Warmy (warmup and deliverability dashboards)
Meetings and scheduling — Cal.com (open-source, API-first scheduling) · Recall.ai (universal meeting-bot API)
5. Systems of record
CRM — Attio (API-first modern CRM) · folk (lightweight relationship CRM) · HubSpot (the SMB and mid-market default; now runs its own read-write agent connector and developer program) · Pipedrive (pipeline-focused CRM) · Salesforce (the enterprise system of record)
Product analytics and customer data — Amplitude (product analytics with cohort export) · PostHog (developer-native product analytics, SQL-over-API) · RudderStack (warehouse-native open-source customer data platform) · Segment (enterprise customer data platform)
6. Context layers
CaliberMind (governed query layer over a unified marketing data warehouse) · Common Room (identity-resolved signals with enterprise governance, via API, MCP and CLI) · Octave (context engine — ideal customer, personas, offerings — served to agents) · ZoomInfo GTM.AI (the incumbent's headless context layer: API, MCP and CLI over its data graph)
7. AI SDRs and closed-loop platforms
11x (autonomous AI SDR, enterprise focus) · Amplemarket (engagement plus data plus AI copilot) · Artisan (AI SDR with its own database and deliverability stack) · Copy.ai GTM (content and workflow agents) · Landbase (natural-language audience building and campaign launch) · Regie.ai (AI prospecting content) · Relevance AI (build-your-own agent workforce) · Unify (warm outbound on intent signals, with agent access and per-verb pricing)
8. Agent plumbing
Composio (tool router and managed auth over hundreds of app toolkits) · Klavis (connector infrastructure for agent tools) · Pipedream (thousands of apps exposed as agent tools with managed auth; Workday-acquired) · Scalekit (auth infrastructure for agent tools) · plus the community directory ecosystem around MCP servers
9. Unified verb layers
ColdIQ (unified GTM API and skills out of a $7M outbound agency; verb routing live, workflow and context features announced) · GrowFlex (multi-tenant verb API routing across providers, built from the per-customer-isolation end; ours — see disclosure)
Several category-2 aggregators are converging on this territory from below; expect this list to be longer, and possibly differently shaped, by winter.
10. Skills, plays and operator content
GTM Skills (open-source agentic GTM toolkit, MIT-licensed) · sales-skills (agent skills for CRM, outbound and enrichment, one-line install) · and a wide band of agencies and operators publishing playbooks, guides and small tools for agent-driven GTM, including Cannonball GTM, DevCommX, Foundersgtm, GTM Engineering, GTM Works, Novoslo, OneAway, Reachly, SaaSHero, SalesCaptain, The Signal Club, Understory and Veerview.
Spanning categories, and cautionary tales
Apollo (a data source, a workspace and a sequencer at once — the clearest demonstration that altitude matters more than category) · Warmly (AI-native intent platform, acquired by HubSpot in June 2026 — the first acquisition of this space by an incumbent) · Seamless.ai (contact data, now surfaced inside HubSpot's prospecting agents) · Proxycurl (LinkedIn data provider shut down by litigation in 2025 — the environment's standing cautionary tale; deliberately unlinked).
A note on sources: this space is young enough that much of what's written about it is written by its vendors, whose content operations often dominate search results for their own categories. We have excluded vendor-published accuracy benchmarks throughout, described pricing by shape rather than by number (numbers in this space rarely survive a quarter), and dated every claim we could.
Disclosure: this guide was written by the team building GrowFlex, one of the companies listed in category 9. We've tried to describe everyone — ourselves included — in the same register. Corrections are welcome and will be incorporated in future revisions.
Current as of July 2026.