Written by
Patricia Diaz-Hymes
Principal Product Marketing Manager
On Oct 1, Decagon announced Personal Agent Gateway, a product that identifies personal AI agents contacting a company on a customer's behalf, routes their requests and controls their permissions. It uses "various signals" to flag likely personal agents in live chat and voice. The business defines the scopes an agent can request, and the customer approves what the agent actually gets: in Decagon's airline example, an agent asks to view and rebook flights and the customer approves viewing only. Decagon also introduced PACT (Personal Agent Consent & Trust), built on OAuth 2.0 and the Agent2Agent protocol, and says it is working with personal agent providers and enterprise customers to refine the specification. In Decagon's words: "The customer chooses who acts for them. The business sets the terms."
What it means for a GTM team: support, success, and sales teams may start receiving requests that come from a customer's software rather than the customer. Someone has to decide which requests an agent can make on its own (a usage lookup, an invoice copy), which need a human, and which are off the table (a renewal quote, a cancellation, a discount). A useful exercise is to draft that list with your CS lead and a PMM and see where they disagree. Whatever an agent can read, from the pricing page to help-center articles, is likely what it will be working from.
Salesforce agrees to buy Listen Labs, an AI customer research platform
On Sept 29, Salesforce signed a definitive agreement to acquire Listen Labs. Terms were not disclosed, and the deal is expected to close in Q4 of Salesforce's fiscal 2027, pending regulatory approval. Listen Labs' agents design studies, recruit participants, run interviews and synthesize findings; the platform cites 50M+ participants, interviews in 120+ languages, and digital twins that simulate how customers might respond to a new product or marketing message before launch. Salesforce says the capabilities will extend across Marketing Cloud, Service Cloud and its AI portfolio.
What it means for a GTM team: research and message testing may sit inside the same stack that already holds campaign and customer data, and Salesforce describes the cycle as months compressed to days. For PMM, the practical questions are which messaging hypotheses you would run through it first, and how you would compare a simulated result against what customers say in real calls.
ZoomInfo ships Agent Teams inside GTM Studio
ZoomInfo launched Agent Teams, a multi-agent orchestration layer in GTM Studio built on technology from DoubleO.ai, now part of ZoomInfo (page updated Oct 1). It is not a separate product: "There's no new contract, no new SKU, and no separate purchase," with usage billed as AI Credits against the existing contract. Teams describe a goal in plain English and Agent Teams generates the workflow; templates cover motions like inbound lead activation and closed-won account handoffs. Each agent evaluates the next best action from the current state of the record rather than a fixed path.
What it means for a GTM team: if a play is described in plain language and acts on the current state of the record, the quality of the account fields, persona definitions and handoff criteria in the CRM sets the quality of the play. Teams already on ZoomInfo may also want to ask RevOps how credit consumption will be tracked.
Braze moves campaign QA and campaign control into agents
At its Sept 29 event, Braze announced four AI releases. Agentic Standards (beta now, GA in October) automates campaign and Canvas QA across naming, tagging, copy, personalization and compliance. Operator Connect (beta) lets marketers reach Braze from Claude, ChatGPT, Microsoft Copilot and MCP-compatible tools. Decisioning Studio Go, a self-service AI decisioning agent, is in beta with GA expected Oct 14, and conversational agents for WhatsApp, SMS, RCS and web are in beta.
What it means for a GTM team: an automated QA pass checks against whatever standards have been written down, so it is worth finding out whether your naming conventions, messaging guardrails and compliance language exist in a form a tool can apply. Lifecycle marketers may also soon run campaigns from a chat window, which shifts review from the finished campaign to the instructions behind it.
How it connects: Decagon covers the customer side, where an agent may arrive on a customer's behalf. ZoomInfo and Braze cover the vendor side, where agents run plays and check campaigns. Listen Labs sits between them, testing the messages both sides will read. In each case the agent works from written definitions of who the audience is, what can be claimed and what is allowed.
My take
At one B2B martech company, a client was sold functionality that sat just beyond what the product could do, matched to what a competitor was claiming. Nobody lied; the people running the sale and filling out the RFP believed it was close, and it was further away than the team could afford. Product and engineering stayed up through a holiday to close the gap.
That was GTM debt paid in a person's time, and it only worked because a person was there to pay it. A customer's agent reading your pricing page, your help center and your answers in chat has no one in the loop who knows the stretch is a stretch, and it can compare what you claim against whatever it can check. Plenty of teams carry two versions of the story, the one that gets said in the room and the one the product can back up, and so far people have been quietly absorbing the distance between them.
The gateway Decagon built lets a business set the terms for what an agent may do. The same exercise applies to what you let an agent believe about you: write down why you win, who you win with, and what you won't pretend to do, clearly enough that a stranger's software could read it and a new hire could repeat it. The last item is the one I'd start with, because it's the one almost nobody has put on paper.
