Correction (Oct 19, 2026): The Q2 2026 figures in the funding section have been revised from $1,305M across six rounds to $1,355M across seven, after we found an unrecorded round while re-scoring the market maps: Bland’s $50M Series C of June 16, 2026, led by Dell Technologies Capital. Sierra‘s share of the quarter falls from 73% to 70%, and the ex-largest-deal figure rises from $355M to $405M. The Q3 figures and every conclusion in this report are unchanged. The miss is a known limitation we have flagged before — our research and the monthly routine’s are both roundup-level rather than per-vendor.

Editor’s note: This report publishes on October 19, nineteen days after the quarter closed. It covers July 1 – September 30, 2026, and every figure in it is checked against the funding tracker as recorded, not against our own arithmetic. Where a number is company-stated or single-sourced, it says so.

Q3 2026 was the quarter the platforms stopped selling seats and started giving agents away. ZoomInfo announced it would move customers from licences to consumption credits, then shipped its flagship multi-agent product with no new SKU attached. HubSpot retired the brand it had spent two years building and moved two agents to pay-per-result. Salesforce named a preferred frontier model after years of selling model-agnosticism. Meanwhile the venture market did almost nothing: $275M across five rounds, 42% of it to one company.

The thesis we have been running since Q2 — that the tools which augment judgment compound while the tools which promise to replace it stall — survived the quarter intact. What did not survive was our own predictive record. We made six specific company-level predictions in the Q2 report. Scored honestly, we went nought for four on the ones that could be resolved, and one of the six turned out to be unscoreable by construction. That scorecard is the most useful thing in this report, so it goes first.

The Q2 Scorecard: Six Predictions, Graded

Prediction 1 — One of 11x or Aisdr.com shuts down or gets acqui-hired by September 2026. Missed.

Neither happened. 11x remains operating; its founder-CEO had already stepped down to non-executive chairman in May 2025, and the company disputes the churn figures reported about it, citing 79% retention (company-stated). We should be careful about the inverse error too: an absence of news is not evidence of health, and 11x’s current state is simply not publicly verifiable. The category’s only genuine shutdown in the window was Relay.app, which closed free accounts on August 15 and ended paid access on September 14, with the team joining Google’s Chrome group — an orchestration tool, not an AI SDR. The consolidation pressure we identified was real; we pointed it at the wrong shelf.

Prediction 2 — HubSpot acquires Attio or Folk by Q4 2026. Missed, and the reasoning was inverted.

HubSpot did acquire — but it bought Warmly on June 30, in the intent and signal layer, not a CRM. Attio, meanwhile, behaved like an acquirer rather than a target: it raised a $52M Series B led by GV (taking it to $116M total) and bought Inconvo on July 10 for natural-language analytics. We read Attio as prey. It was hunting. Q4 technically runs to December 31, so this one is not formally closed out, but the premise is broken.

Prediction 3 — Smartlead crosses $50M ARR by December 2026. Unscoreable.

Smartlead is bootstrapped and does not publish ARR. There is no mechanism by which we — or anyone outside the company — could verify this, which means we should never have written it as a prediction. We are recording it as a methodology failure rather than quietly dropping it, and the Q4 predictions below each carry an explicit resolution criterion for exactly this reason.

Prediction 4 — RB2B raises a growth round at $300M+ by Q3 2026. Missed.

No round. RB2B’s record still shows seed-stage funding. The prediction itself named the alternative — “RB2B stays independent and profitable” — and that is what happened. When your own hedge is the outcome, you did not make a prediction; you described a distribution.

Prediction 5 — ZoomInfo gets taken private by Q4 2026. Not yet — but the premise strengthened considerably.

No deal has been announced. Everything underneath the thesis got worse for the company and better for the argument: Q2 revenue of $310.4M grew 1.2% year over year, net revenue retention slipped to 89%, the quarter carried a $650.5M goodwill impairment, and Mizuho cut the stock to Underperform. The market capitalisation we cited as the take-private price in Q2 was roughly $3.5B; it is now around $1.2B, down about 63% over twelve months. A take-private is cheaper and more logical than when we wrote it. Open until December 31.

Prediction 6 — Anthropic ships an agent skill marketplace by Q3 2026. Closest to a hit, with the mechanism different from what we described.

Agent Skills did move from a closed Anthropic-defined feature toward an open standard with a partner directory, and the Claude Developer Platform added Agent Skills support in August. But we have not been able to confirm from a primary Anthropic source that there is a third-party marketplace with a public submission process; the large skill ecosystems that exist are community-run and GitHub-based. Our sourcing here is secondary, so we are grading this as directionally right and mechanically wrong rather than claiming a win.

What the scorecard actually says

Every structural call in the Q2 report held: the signal layer consolidated into platforms, pricing moved off seats, the replace-the-human vendors did not recover, and the augment-judgment vendors compounded. Every company-level call failed. That is not a coincidence — it is the difference between forecasting a system and forecasting an actor inside it. We will keep making named predictions, because unfalsifiable analysis is worthless, but this report’s Q4 section is deliberately weighted toward the structural end and every item now states how it will be scored.

The Quarter’s Structural Story: Seats Out, Agents Free

Three things happened in sequence, and together they are the quarter.

First, on August 5, ZoomInfo told investors it would move from seat-based licensing to hybrid consumption pricing — a lower platform fee plus pre-purchased credits — beginning with new business at the end of Q3 and migrating existing customers opportunistically through 2027. The model is explicitly optional rather than mandatory. The same afternoon, HubSpot’s leadership described the company as “changing products, pricing, and go-to-market model” around its AI agents, a week after retiring the Breeze brand into a new Agent Hub (public beta July 23) and launching Revenue Hub from $95/month plus transaction fees. We covered that afternoon in The Repricing, and nothing since has undermined it.

Second, the agent capability itself started arriving unmetered. Gong‘s August release added admin-managed MCP clients and a listing in the ChatGPT Apps directory — and, notably, monthly per-user AI credit limits, the first sign of AI usage becoming a billable dimension in conversation intelligence. Then on October 1, one day after announcing the acquisition that powered it, ZoomInfo shipped Agent Teams inside GTM Studio with no new SKU and no new contract. A company mid-transition to usage-based pricing gave its flagship agent product away as an entitlement.

Those two facts are in tension, and the tension is the most important open question in the category. If agents are the thing customers will pay consumption fees for, you do not hand them over as an entitlement. If agents are table stakes — a feature you must ship to defend a renewal — then the consumption revenue has to come from the underlying data calls instead, which is precisely what a “credit grid” meters. Read ZoomInfo’s two moves together and the strategy becomes legible: give away the orchestration, charge for the grounding. Whether that clears the revenue hole left by 89% net retention is the single number to watch when Q3 earnings land in early November.

Third, the model layer stopped being neutral ground. On August 26 Salesforce and Anthropic announced Claudeforce — Claude as a reasoning model inside Salesforce, and Salesforce inside Claude as a plugin with 37 prebuilt sales skills, with open beta slated for September. We have found no confirmation that the September beta opened; if you are planning a Q4 pilot on it, verify first. The strategic point stands regardless: the vendor that spent two years selling model-agnosticism publicly named a preferred frontier model.

Funding & M&A: Concentration, Not Collapse

Q3 2026 recorded five venture rounds totalling $275M and seven acquisitions. Taken alone those numbers read as a severe contraction. Taken in context they read as something more specific.

Excluding foundation-model rounds, which distort every aggregate they touch, GTM venture capital by quarter runs: Q2 2025 $352M (5 rounds) · Q3 2025 $603M (6) · Q4 2025 $239M (3) · Q1 2026 $285M (2) · Q2 2026 $1,355M (7) · Q3 2026 $275M (5). So the deal count has been broadly stable for a year and a half while cheque sizes swing wildly — and the swing is driven almost entirely by one deal per quarter. Sierra’s $950M in Q2 was 70% of that quarter’s total. Clay‘s $115M in Q3 was 42% of this one’s. Strip the largest round out of each and you get $405M and $160M respectively: a market funding four-to-five modest rounds a quarter, with an occasional outlier that rewrites the headline.

Against Q3 2025’s $603M, this quarter is down 54% year over year. That is the honest comparison, and it is a real cooling — but it is a cooling in cheque size at the top, not a drying-up of deal flow.

The rounds

Aligned $60M Series B (July 1) · Alta $25M Series A (July 8) · Sable $45M (July 16, no stage label in the primary source) · Encore AI $30M Series A (July 29) · Clay $115M Series D at $7.1B (September 9, Wellington Management leading). Clay’s valuation more than doubled its August 2025 mark of $3.1B; the company published neither a lifetime funding total nor whether $7.1B is pre- or post-money, so our ~$319M total is derived and labelled as such.

The acquisitions

Seven deals, and only two carried a disclosed price — both of which we know only because a public acquirer had to file. Common Room → Zoom, closed July 17 for $266.8M all-cash per Zoom’s 10-Q, $198.8M of it goodwill. Agency → Klaviyo, announced August 4 for up to $17M cash per Klaviyo’s 10-Q. The other five — Ciro → Reevo (July 8), Inconvo → Attio (July 10), Seam AI → Clarify (July 21), Highspot–Seismic (closed August 18) and DoubleO.ai → ZoomInfo (September 30) — were all undisclosed.

That ratio is the useful finding. The Common Room comp — roughly five times the target’s lifetime venture funding, three-quarters of the price booked as goodwill — is the only genuine public anchor founders in this category have. Everything else is a handshake nobody has to report. If you are building a single-job AI GTM tool, the tuck-in is still the likeliest exit, and you now have exactly one number to negotiate against.

Category Movements

Data & Enrichment

The category’s centre of gravity moved decisively. Clay is now valued at $7.1B while Apollo‘s $1.6B mark dates to August 2023 and ZoomInfo’s market capitalisation sits near $1.2B. All three shipped agent surfaces: Apollo’s MCP server has been free on every plan since February, with an AI Assistant in March and a ChatGPT app in April; ZoomInfo shipped a CLI (July 9), a self-scored agent benchmark (July 10) and GTM.AI general availability (July 17); Apollo’s ApolloNEXT keynote on September 30 announced an Intelligence Layer (available now) plus Builder Studio and Messaging OS (both “coming soon” — several write-ups have reported these as shipped; they are not). Agent access is no longer a differentiator in this category. Price stability and data depth are.

Intent & Signal

Effectively absorbed. Warmly → HubSpot (June 30), Common Room → Zoom (closed July 17), Seam AI → Clarify (July 21). Of the credible independents we tracked entering 2026, Unify is the last one standing at scale, which makes independence a genuine selection criterion and Unify the most obvious next target. RB2B remains independent at the SMB end.

Sales Enablement

Collapsed from two credible independents to one. Seismic completed its merger with Highspot on August 18 — signed February 12, no value disclosed, Seismic the surviving brand under CEO Rob Tarkoff, Permira the controlling shareholder, 2,500 customers and 3.5 million users (company-stated). Buyers lose the competitive tension that made this category negotiable. The base rate on integration promises here is Clari’s: it bought Groove in August 2023 targeting six to nine months, and the products were still separate in 2026.

Revenue Intelligence

Gong spent the quarter making itself reachable from wherever the seller already is — a ChatGPT Apps listing, admin-managed MCP clients, then Mission Callisto on September 30 with Gong Enrich piping third-party contact data into its Revenue Graph (Gong’s help centre still describes Enrich as rolling out, so treat availability as staged). The launch partner list includes Apollo and ZoomInfo alongside Findymail, Firmable, Kernel, LeadIQ, Lusha, RocketReach, Wiza and ZeroBounce — competitors as suppliers inside a rival’s product. Clari + Salesloft, meanwhile, integrated Clari Copilot into Salesloft in June and then partnered with allGood on August 5 rather than build AI email itself. We re-score that merger at its twelve-month mark in December; renting a core capability is a data point for it.

Support & CX Agents

The quarter’s consolidation was downstream of the frontier. Salesforce agreed to buy Fin (formerly Intercom) for $3.6B on June 15, closing expected early 2027. SoundHound completed its acquisition of LivePerson on September 4 — announced April 21, roughly $43M in equity value, implying about $250M enterprise value after discounted debt. A company once valued in the billions exited for less than a Series B. Sierra ($15.8B), Decagon ($4.5B) and Crescendo sit above that line; the pre-LLM conversational cohort is being cleared out beneath it.

Orchestration

ZoomInfo bought DoubleO.ai on September 30 and shipped Agent Teams on its architecture the next day, and folded Workbounce (acquired November 2024) into its GTM Context Graph to add internal unstructured knowledge to external verified data. n8n, valued at $5.2B after SAP’s May secondary, remains the open-source anchor. The category is bifurcating: platform-owned orchestration bundled free, versus independent orchestration you host.

AI SDR

Still the quietest shelf relative to its 2024 noise. Alta’s $25M Series A and Sable’s $45M were the quarter’s only AI-SDR-adjacent rounds, and both fund bounded, product-grounded jobs rather than rep replacement — Alta an orchestration “company brain,” Sable an AI sales engineer that runs live demos. Neither 11x nor Artisan published pricing during the quarter; Artisan’s pricing page now routes to a sales conversation, which removed the transparency advantage it used to hold.

The Foundation Model Layer

Anthropic shipped three models in seven weeks — Fable 5 (June 9, $10/$50 per million tokens), Sonnet 5 (June 30, $2/$10, now permanent pricing) and Opus 5 (July 24, $5/$25, the same price as its predecessor) — alongside Haiku 4.5 at $1/$5. OpenAI shipped the GPT-5.6 family in July and cut prices on July 30: Sol $5/$30, Terra $2/$12, Luna $0.20/$1.20. Input pricing is now identical at two tiers, with Claude roughly 17% cheaper on output and GPT-5.6 Luna having no Claude equivalent at the volume end. Our Claude vs GPT comparison works through the routing implications, and the Q3 capability tracker covers the lineup in detail.

Two non-capability events matter more for GTM teams than any benchmark. An eighteen-day export-control suspension took Fable 5 off the market globally between June 12 and 30; model availability is now a regulatory variable, not merely an uptime one. And from August 2, Anthropic began embedding imperceptible watermarks in text from models launched on or after that date, with C2PA metadata on supported files — driven by EU AI Act Article 50 but applied worldwide, with pre-August models covered by a transition period ending December 2. If your team drafts outbound with Claude, that content now carries a machine-readable marker. It proves processing, not authorship, and its absence proves nothing.

On the protocol layer, the MCP 2026-07-28 specification went final on schedule: stateless, with enterprise-managed authorization promoted to stable. Every platform move described above runs through it. We have found no successor revision as of this writing.

Buyer Recommendations by Stage

Pre-Seed / Bootstrap

Nothing in this quarter changes the answer: Apollo plus a sending layer, with free tiers used for evaluation rather than operation — note that Apollo’s free plan is now 900 credits per seat per year, not the five-figure monthly allowance older comparisons quote. Clay’s entry tier at $54/month on annual billing is reachable earlier than its $7.1B valuation suggests.

Seed

Buy the augment-judgment tools and avoid anything selling autonomy. The one new instruction this quarter: if a vendor is mid-pricing-transition, do not sign a multi-year deal on the outgoing model without price-equivalence in writing.

Series A–B

This is where the repricing bites. If you run ZoomInfo, model your consumption before your renewal — under credit pricing your agent architecture is your invoice, and an enrichment loop that re-queries per step costs real money. Take the free agent entitlements (Agent Teams, Agent Hub beta) and pilot them, but assume the metering moves to the data calls underneath.

Enterprise

Two structural losses to price in: enablement is now a single-vendor category, and the intent layer is almost entirely platform-owned. Both reduce your negotiating leverage. The standing 2026 contract rules — change-of-control, data portability, annual maximums, and now pricing-model stability with an exit right when the meter changes — are no longer prudent extras.

Predictions for Q4 2026

Five predictions. Each states how it will be scored, which is the correction to our Q3 methodology failure.

1. ZoomInfo’s Q3 earnings show net revenue retention at or below 89%. Resolves: the reported NRR figure in ZoomInfo’s Q3 2026 release, expected early November. The consumption transition starts with new business, so it cannot help retention this quarter, and the trend has been monotonic.

2. No AI GTM venture round above $150M is announced in Q4 2026. Resolves: our own tracker, foundation-model rounds excluded. Three of the last four quarters had exactly one outlier deal; Clay just absorbed this cycle’s appetite.

3. At least two more intent-and-signal or enablement vendors are acquired by platforms by December 31. Resolves: dated, double-sourced acquisition announcements recorded on the tracker. Seven deals in Q3 and a category with few independents left.

4. Unify is acquired or announces a round at a materially higher valuation by Q1 2027. Resolves: a dated announcement either way; scored as a miss if neither occurs. Being the last independent in an absorbed category is not a stable position.

5. No vendor in this category announces a pure per-agent or per-outcome price list — published, not quoted — by December 31. Resolves: a public pricing page. Agents are being given away as entitlements precisely because nobody has solved how to price them standalone.

The Independent Analyst’s View

The augment-versus-replace thesis is now boring, which is the best thing that can happen to an argument. The interesting question has moved one layer down: nobody has worked out how to charge for an agent. ZoomInfo gives its agent orchestration away and meters the data underneath. HubSpot moved two agents to pay-per-result and bundled the assistant at no charge. Gong capped AI credits per user rather than pricing agents as a product. Salesforce sells Agentforce per resolution, then redefined the ARR metric that measures it. Five serious companies, five different answers, and not one published per-agent price list between them.

That is what a category looks like before a pricing standard emerges, and it has a direct consequence for buyers: every agent you are offered this quarter is cheaper than it will be. Entitlements are a land-grab phase, not a permanent state. Take them, build on the ones you can replace, and read every contract on the assumption that the meter moves — because on the evidence of this quarter, it will.

Methodology + Sources

Funding and M&A figures are as recorded on the GTMLens funding tracker, which requires an exact date and either a primary source or two independent sources before an event is recorded; undisclosed deal values are recorded as undisclosed rather than estimated. Quarterly capital comparisons exclude foundation-model rounds, which would otherwise dominate every aggregate. Public-company figures come from the companies’ own releases and filings — ZoomInfo and Klaviyo 10-Qs, Zoom’s 10-Q for the Common Room consideration, and earnings releases for revenue, retention and guidance. Model pricing comes from the vendors’ own announcements. Vendor growth claims, customer counts and retention figures sourced to a company are labelled as such and are not independently verified. Where our sourcing is secondary, the text says so. GTMLens accepts no vendor money, paid placements or affiliate commissions; see our editorial policy.