Gong

The ingestion layer is the product rather than a feature. The Revenue Graph connects calls, emails, meetings and CRM data into one layer, so conferencing, email, calendar and CRM are all first-class inputs rather than optional connectors.

The Microsoft collaboration adds Dynamics 365 and the wider Microsoft application estate, with interactions captured, structured, synced and enriched with AI-generated summaries, key topics and next steps so records populate without manual entry, plus Marketplace distribution. MCP support extends tool interoperability beyond the named set.

An independent review reports no API access for parallel dialers, incomplete third-party outreach coverage and some CRM syncs requiring workarounds. That is a third-party characterization. Gong does not publish its connector estate as a directory a buyer can check.

Sourcegong.io/press Mission Big Dipper and Microsoft collaboration releasesread 2026-09-04

An explicit agentic execution layer, shipped rather than announced. The Gong Revenue Harness governs, orchestrates and connects AI agents across the full revenue cycle, described by the vendor as the evolution of its Agent Studio and Model Context Protocol investments, and Custom Agents reached general availability in June 2026, automating follow-ups, pipeline edits, enablement triggers and forecast corrections across connected systems.

The pitch is that a revenue operations leader describes a workflow in natural language and the agent runs against the Revenue Graph, so composition does not require engineering. Several capabilities in the same announcement were roadmap rather than shipped, listed as expected in July 2026, among them Gong Assistant in Account Console.

Sourcegong.io/press Mission Big Dipper releaseread 2026-09-04

An explicit grounding layer that is the product's whole architectural claim. The Revenue Graph connects calls, emails, meetings and CRM data into one trusted source that agents operate against, so an agent answers from the interaction record rather than from a model's general knowledge, and the vendor's stated advantage is the data and real-world context underneath the AI rather than the AI itself. Forecast reads more than three hundred distinct signals per conversation.

Proprietary generative and predictive models are described as trained on customer interaction data at scale and tuned to revenue terminology, so the grounding is domain-specific as well as customer-specific. Capture is automatic and complete across customer-facing calls, meetings, demos and emails, which is what makes the corpus exhaustive rather than sampled. The Revenue Graph is a record store that grounds the agents rather than agent memory. No citation or source attribution on generated output is documented, so a summary or a forecast correction cannot be traced to the specific conversation behind it.

Sourcegong.io/press Mission Big Dipper release and gong.io product pagesread 2026-09-04

A governance layer is claimed and no gate is documented. Gong positions the Revenue Harness as supplying the three things it says generic AI tools lack in production, revenue context, governance and human-in-the-loop control, and describes the Harness as governing, orchestrating and connecting agents across the revenue cycle. That is a stated design commitment.

The phrase human-in-the-loop control is positioning rather than a mechanism: nothing documents an approval queue, review step, confidence threshold that withholds an action, blocked-action policy or escalation rule before a Custom Agent edits a pipeline, corrects a forecast or updates a CRM record. Dry Run is not a gate either: it is a rep role-play rehearsal inside the coaching product, not a facility for exercising an agent before it touches live data, and it was listed as expected in July 2026. The governance layer is an explicit product component rather than a marketing adjective alone, and agents are deployed through Agent Studio rather than running by default.

Sourcegong.io/press Mission Big Dipper releaseread 2026-09-04

Gong's own security documentation lists SOC 2 Type 2, ISO 27001, ISO 27017 for cloud service controls, ISO 27018 for personally identifiable information in public clouds, and ISO 27701 for privacy information management, with the ISO certificate downloadable from the trust center at trust.gong.io, which also carries a dedicated Artificial Intelligence section. Gong is certified to the EU-US Data Privacy Framework and aligned with GDPR and CCPA.

Its assurance practice includes a vulnerability management program validated through the SOC 2 and ISO certifications, an ongoing bug bounty, and a vulnerability disclosure program at vdp.gong.io. On access, the trust page names multi-factor authentication and least-privilege access controls, with configurable controls over how data is collected, accessed and retained, and voice identification is disabled by default and requires user consent when enabled. ISO 42001, which some third-party profiles list, is not claimed on Gong's own pages.

Sourcehelp.gong.io security summary, gong.io/platform/trust and trust.gong.ioread 2026-09-04

Reporting across the revenue picture, and no record of the agents. Scorecards, deal health, pipeline risk and rep performance reporting give managers a single view, conversation capture and logging are inherent to the platform so every call and email is retained and searchable, and the Revenue Harness is positioned as the governance layer over agent execution. Governance framing is not a trail.

Nothing documents a log of which Custom Agent ran, what it changed in a pipeline or forecast, what it read from the Revenue Graph before acting, or a configuration change history, and no retention or export control over such a record is described. The gap is material rather than technical here because Custom Agents reached general availability in June 2026 and write into systems of record.

Sourcegong.io/platform/trust and gong.io/press Mission Big Dipper releaseread 2026-09-04

The Revenue Graph is the interaction record, not agent memory. It persists every call, email, meeting and CRM record agents operate against, so an agent picking up an account reads what happened before. That is the data platform the agents ground on, and a record store is not memory: no memory construct the agents write and read back as their own state, no per agent recall across runs and no retention or forgetting control at the agent level is documented.

Sourcegong.io/press Revenue Harness launchread 2026-10-01

Multi-tenant cloud with no published region choice. Gong's trust page, platform pages, security blog and help center security summary enumerate certifications, encryption, multi-factor authentication and least-privilege controls in detail without naming a hosting region or a residency option, and none offers region selection, an EU or other in-region option, private cloud, VPC, a single-tenant tier or an on-premises route.

Purchasing through Microsoft Marketplace under Azure consumption commitments is a procurement route, not a deployment one. What the vendor does document on cross-border movement is legal cover, certification under the EU-US Data Privacy Framework alongside GDPR and CCPA alignment, and legal cover is not residency.

Sourcegong.io/platform/trust and help.gong.io security summaryread 2026-09-04

Specialized agents that ship built, plus a generator that produces the configuration rather than asking for it. Prebuilt agents cover pre-call research, post-call follow-up drafting, pipeline review summaries, forecast corrections and CRM updates, and Custom Agents reached general availability in June 2026 so a customer composes their own on top.

The distinctive element is AI Builder for Scorecards, generally available in the same release, which generates complete coaching-ready scorecards automatically from the exact criteria found in a customer's top-performing closed-won calls, eliminating hours of manual evaluation design. Each scorecard is derived from the customer's own data rather than a generic template. No browsable catalog or marketplace of agents is published, so the set is what Gong ships plus what the customer builds.

Sourcegong.io/press Mission Big Dipper releaseread 2026-09-04

The conversation itself is the trigger surface, and capture is automatic rather than opt-in per meeting. Every customer-facing call, meeting, demo and email is captured and analyzed without anyone starting a recording, so work reaches the platform as a by-product of selling rather than as an act of administration.

Agents fire on post-call events, pipeline changes and enablement triggers, and AI Coach delivers a personalized debrief immediately after an AI Trainer session rather than waiting on a manager, which is an event-driven handoff rather than a scheduled report. Channels out reach the systems teams already work in: CRM writes, Dynamics 365 and Microsoft applications through the Marketplace integration, and the Gong Assistant surfaces. No inbound conversational channel is documented, so a rep asks through Gong's own surfaces rather than by messaging the agent where they already are, such as in Slack.

Sourcegong.io/press Mission Big Dipper release and gong.io product pagesread 2026-09-04

The customer does not choose the model and Gong does not name what runs. Gong's own material describes proprietary generative and predictive models tuned for revenue workflows and trained on customer interaction data, and its trust page, platform pages, security blog and Mission Big Dipper release carry no model list, routing policy, per-workspace selection or bring-your-own-model option.

Sourcegong.io/platform/trust and gong.io/press Mission Big Dipper releaseread 2026-09-04

Two documented external surfaces, both live. Agent Studio lets customers build custom agents rather than only consume shipped ones, and Model Context Protocol support is named first-party as a foundation of the Revenue Harness, with Gong stating in its Microsoft collaboration announcement that MCP support is now live and that the platform is available on Microsoft Marketplace today.

Interoperability runs into the Microsoft ecosystem specifically, with customer interaction data captured, structured and synced into Dynamics 365 and across Microsoft applications, enriched with AI-generated summaries, key topics and next steps. An independent review reports no API access for parallel dialers and some CRM syncs requiring workarounds, a third-party characterization.

Sourcegong.io/press Mission Big Dipper and Microsoft collaboration releasesread 2026-09-04

Everything documented here evaluates the salespeople, not the agents. AI Trainer runs practice sessions and AI Coach delivers a personalized debrief immediately afterwards rather than waiting on a manager; AI Builder for Scorecards generates coaching-ready scorecards automatically from the criteria found in the customer's own top-performing closed-won calls, which removes hours of manual rubric design.

Those are customer-held mechanisms with criteria and readable results, aimed at rep performance as a business outcome. Dry Run is not agent testing. Gong's own release places Dry Run inside Gong Enable, the coaching product, among three features aimed at rep readiness, and describes it as spinning up a role-play rehearsal from a calendar invite that replicates the customer persona using real account context and conversation history.

It rehearses a person against a simulated buyer; it does not test an agent. The same release lists it as expected in July 2026 rather than generally available, and no first-party confirmation that it shipped is published. Nothing else documents an evaluation harness, scored test set, benchmark or regression check over agent behavior, which matters because Custom Agents write pipeline edits and forecast corrections into systems of record.

Sourcegong.io/press Mission Big Dipper releaseread 2026-09-04

Everything the agents touch arrives through the Revenue Graph or an authenticated connector. Capture runs through conferencing platforms, email and calendar as an integrated participant and processor; agents act through the Revenue Graph, CRM integrations, Dynamics 365 and connected systems; and MCP carries tool interoperability. The Mission Big Dipper release, the Microsoft collaboration material, the trust pages and the help center describe no browser control, page navigation, portal autofill, form filling, headless session or screen automation.

Sourcegong.io/press Mission Big Dipper release and help.gong.ioread 2026-09-04