Seventeen days in one line
The outside assistants stopped reading the record and started writing in it. Outreach lets an agent build a whole sequence, ChurnZero and Airspeed ship write access switched off until someone turns it on, and Demodesk, Arrows and Amplemarket leave every draft waiting for a person, so the agent got the pen and the rep kept the stamps. Elsewhere Salesforge and HeyReach put cloned voices into LinkedIn voice notes, Salesloft took the recording bot out of the Zoom participant list, and RB2B stopped identifying website visitors in Germany.
This issue covers November 1 to 17, with 62 changes across 62 vendors, one each. More than a quarter of them put a product within reach of an outside AI assistant.
Thanksgiving is next week in the United States, and the outside assistants have been moved up from the kids’ table. They may pass the dishes now. In most houses somebody else still carves.
The agent got the pen
The October 18 issue watched about thirty vendors plug into AI assistants, the better ones with a ceiling so an agent could read the pipeline without rewriting it. The last issue gave the assistants a budget. This period they started writing.
Outreach added eleven tools to its MCP Server, covering voice call logs, Kaia coaching scorecards, template search, a task list and full sequence construction. An agent in Claude or another MCP client can now build a sequence end to end. The tools follow standard role based access, need the Amplify add on and roll out from November 12 to December 8.
That is the line between an assistant that reports on outreach and one that authors it. What an agent can alter under an ordinary rep role is now a product question, and a fair one to put to any vendor selling against it.
ChurnZero opened writes with the most care. AI tools connected through ChurnZero Connect, and integrations built on the REST API, can now create and update records such as tasks, notes, success plans and account fields. Admins enable Create and Update separately for each record type, nobody has write access by default, and message drafts wait for a person to send them.
MarketBetter went wide instead. Agents can operate its full API, which the company puts at over 1,100 operations, through three generic tools that discover, inspect and call endpoints. Access runs through a remote MCP server with OAuth 2.1 sign in or a bring your own key skill, and every request runs under the connected user’s role. MarketBetter says it has not benchmarked the token cost, which is a candid thing to print and a number a trial will have to supply.
monday CRM now has an official plugin in Claude’s plugin directory that reads from and writes to CRM boards, so a meeting recap can be posted to the matching deal without opening the CRM. Amplemarket’s MCP lets an assistant create and edit email drafts in Unibox and return a link so the rep reviews and sends them. It can also update value propositions, tone of voice and sequence instructions, with locked fields still requiring the Manage AI messaging settings permission.
The second half of that is the bigger change. An outside assistant can now edit the settings that shape every AI written message, so whoever holds that permission is in effect the copy chief.
Oppora added 21 agent tools to its MCP server. Agents can read lead and company records, enrich from saved lists, find phone numbers in bulk, save draft replies for review and add Do Not Contact entries, for now only on the direct MCP connection. Airspeed agents can search and read Jira issues, projects and boards, and with permission comment on tickets or update their fields. Write access is off by default, each person turns it on in their own Jira connection, and admins can block it for the whole workspace.
Avoma built the pen in house. Agent Builder is a no code workspace for custom agents that draw on calls, emails, deals and CRM data. Agents run on triggers such as a meeting ending or a deal update and can update CRM fields, alert Slack or Teams or send a Gmail digest. The number of agents depends on plan, and a test mode lets a team check one before relying on it.
Our readWhether an agent can write is settled. The variable now is the unit the permission comes in. ChurnZero grants it per record type and per verb. Airspeed grants it per person, with an admin veto. Outreach and MarketBetter inherit the user’s role, which is tidy until someone remembers how generous that role already was. A role written for someone who works at the speed of a person is now held by something that does not.
Somebody still presses send
Writing to the record is one thing. Writing to a customer is another, and this period most products kept that step with a person.
Demodesk now saves the follow up email it writes after a recorded meeting with external attendees as a draft in the user’s Gmail or Outlook mailbox. Nothing is sent automatically. The feature is on by default once a mailbox is connected with the Create drafts permission.
Arrows lets anyone with access to its spaces connect Claude or ChatGPT and work across all of their deals from the AI app, including one change across many deals, such as drafting a launch email for each. Arrows states that nothing goes out on its own, since drafts wait in Arrows for approval. The connection signs in as the user, so a manager’s assistant sees the whole team’s deals.
FoxReach made the pause a rule. AI assistants connected to it now ask the user before any step that sends email, changes existing records or spends credits. The same release introduced cold.md, a free and open Claude Code plugin that finds and verifies leads, drafts in the user’s voice and sends through FoxReach from one file in a project.
GetResponse published AI Agent Skills on GitHub, packaged knowledge of its API for newsletters, autoresponders, contacts and contact search, for any account with API access. GetResponse states that the agent shows a summary and waits for confirmation before any send, import, enable or delete. The agent works with a full API key, so that confirmation is doing a lot of the work.
Nooks introduced its Revenue Agent Platform, which the company says runs Deal, Outbound, Expansion and Coaching Agents today, with Inbound Agents due in Q1 2027. It also announced Node, an AI assistant for account executives that is open by waitlist only. Nooks states that reps approve every send.
Two releases went the other way. Sendr V4.1 added triggers and actions to its Sequencer, so a sequence can start itself from an event such as a new CRM contact or a Typeform submission. Aircall’s Outbound Campaigns for AI Voice Agents can now run always on, calling new contacts as they arrive by import or API with no end date or overall contact cap. Aircall also extended Simulations to outbound agents, so a team can test scenarios against success criteria before a campaign goes live.
In both, the human decision moved upstream. Nobody approves the message or the call. Somebody approved the entry condition, possibly a while ago, which makes the trigger worth the same review as the copy.
Our readDraft and wait is becoming the default grammar of the category. Demodesk, Arrows, Amplemarket, Oppora and ChurnZero stop at a draft, and FoxReach and GetResponse stop at a question. It is a sensible place to stand, and it is a claim a buyer can test on every send path in an afternoon. The harder cases are the ones where no send step exists to test, because the outreach starts when a record appears.
The connector came with the plan
The packaging was a story of its own. Agent access is mostly being given away.
SalesHood MCP lets sellers query approved library content and add assets or AI created pages to Digital Sales Rooms from Claude, ChatGPT, Gemini, Microsoft Copilot, Glean and Perplexity. SalesHood states that it is available now on every plan, that each request runs as the signed in user and that the seller approves every share.
Albacross shipped an MCP server that connects Claude, ChatGPT and other MCP assistants to its website visitor intent data. An assistant can ask which companies visited and what they read, then tag accounts, push them to the CRM or share them in Slack or Teams. Albacross includes it in every paid plan and trial.
ActiveCampaign pointed the plug the other way. Active Intelligence now connects to outside tools through prebuilt connectors, Google Ads, LinkedIn Ads and Stripe among them, and to any system that runs an MCP server. ActiveCampaign states that all plans can connect as many MCP servers as they need, with each team member connecting under their own account.
Calendly moved its Scheduling API and webhooks onto all plans, Free included. The POST /invitees endpoint, which lets an outside system book a meeting directly, no longer requires a paid plan, though Free usage has its own rate limits and some webhook events still need a plan that includes the feature.
Salesgenie’s MCP connector lets users search its business and consumer data from inside Claude and ChatGPT. Searching is free, a purchase runs through a Salesgenie checkout link at standard pricing, and paid subscribers can upload a customer file for matching and enrichment.
6sense took its Sales Intelligence MCP to general availability on November 3, so any MCP compatible agent can check which accounts are in market, and their buying stage, before it acts. 6sense states that the November release also brings keyword intent APIs to general availability and technographics APIs to beta. Which plans include MCP and API access, and how calls are metered, is the part still to ask.
Distribution moved too. GetAccept is now an official plugin in ChatGPT, which the company says OpenAI reviewed and approved. ChatGPT can see activity in Deal Rooms, contracts and recorded meetings and act on it, for example by sending a signing reminder or logging a call. A reviewed listing does not change what a connector does. It changes whether IT has to be talked into it, which in practice is most of the launch.
Our readSalesHood, Albacross, ActiveCampaign and Calendly asked for no plan upgrade, and Salesgenie lets the assistant browse for free and charges at checkout. When access costs nothing extra, the thing being sold is whatever the agent uses once it is inside: credits, enrichment, records, a list. Pricing the connector as an add on is still on the menu, and this period it was the exception.
Harder to tell from a person
A second set of releases made automated outreach look, sound and arrive more like something a person did by hand.
Salesforge added a Send voice note step for LinkedIn 1st degree connections in multichannel sequences. A note can be recorded once, or generated by AI from a script so each lead hears a personal version in the sender’s voice, using a library voice or a clone, with a text message as the fallback.
HeyReach, whose cloned voice notes this Brief first covered in September, released Voice Notes V2. A workspace can connect its own ElevenLabs account, so senders can use voices from that library, clones included, in LinkedIn voice note steps. Steps can be built from Claude, the HeyReach MCP or the API, and each voice step carries a text fallback.
A voice note used to be the one message in a LinkedIn inbox that proved somebody had spent thirty seconds on you. The recipient now hears what sounds like a personal recording. Whether they are told otherwise is a policy question, and it belongs to the sender.
Salesloft continued a thread from October, when it began importing Microsoft Teams’ own recordings. Zoom recording in Conversation Intelligence now runs through Zoom Real-Time Media Streams instead of a meeting bot, so no bot appears in the participant list. Users who recorded with the bot move to RTMS by default once a Zoom admin approves the app, and Zoom shows participants a disclosure message.
The notice has moved from a visible guest to a platform message. Whether that message meets a consent policy is a question for whoever wrote the policy, and the bot will still turn up on some externally hosted meetings.
QuickProsp v1.32.0 packages cold emails to match what Gmail’s own compose window sends, so a receiving server sees what looks like an ordinary personal email. Gmail no longer shows an Unsubscribe link next to the sender name on those emails unless the visible unsubscribe line is turned on for that campaign. That default touches bulk sender rules and anti spam law, and it is set campaign by campaign, so the decision lands on whoever builds the campaign.
Close added Local Presence, which picks the outbound caller ID for each call and text, choosing a number the contact already recognizes or the one nearest to their area code. It works with Workflows, voice agents and the Predictive Dialer on the Essentials, Growth and Scale plans, and it can only pick from numbers the account holds. It is the local presence thread from the last issue, now inside the CRM.
Leadsforge’s Chrome extension can now pull every lead from a Sales Navigator or LinkedIn people search into a new list that syncs to a Salesforge workspace. The import runs in the background, can find emails and phone numbers, shows a credit estimate first and can enroll leads in a sequence. It runs through the user’s own LinkedIn session, so that is also the account that answers to LinkedIn’s terms on automated collection.
Our readEach of these has a fair use. A local number, a familiar voice and an email that looks typed are what good outreach has always tried to be. What changed is that the resemblance is now a step in a sequence builder or a default in a release, and the disclosure is a setting somebody has to go and find. In most of these products the plainer version is one toggle away, and the toggle belongs to the buyer.
Less data, on purpose
A smaller group of releases did something rarer in this category. They collected, kept or showed less.
RB2B no longer identifies website visitors located in Germany. In the same update the service became inaccessible to users from China, Indonesia, Iran, North Korea and Russia, and signup now requires a company email domain. A team that counted on person level identification of German traffic has lost that coverage. The questions that follow are which other countries are excluded and how future changes get announced.
Clay added two public API endpoints that hard delete a person’s records from Audiences. A request to POST /audiences/privacy-delete with an email or LinkedIn slug queues the deletion, and GET /audiences/privacy-delete/{job_id} reports progress. Job status is kept for only five days, so the record that an erasure finished has to live somewhere else. The same roundup lets admins allowlist MCP clients by domain and cap the credits a workflow can spend, which picks up the allowance thread from the last issue.
Revenue Grid added two controls over what its CRM Sync sends to Salesforce. Rule-Based Email Sync captures only emails that match keyword and phrase rules. Confidential Event Sync shows a meeting’s title, description and location as Confidential when it includes anyone on a confidential list. Both are turned on through the Revenue Grid team.
Aloware v11.42.0 added Hide phone numbers from agents, an admin setting that shows agents only the last four digits of a contact’s number across Talk, the Power Dialer and conversations. Agents can still call and text, Admins and Supervisors keep full visibility, and the setting is off by default. For a team with contractors or offshore agents, it limits how much contact data one person can copy.
The other direction shipped too. lemlist added buyer intent signals at the contact level. Powered by Zeer AI, the signal identifies individuals from their LinkedIn posts and comments and from visits across a network of B2B websites. Each one costs 100 credits and carries an AI generated activity summary and a buying stage. Person level intent is a sourcing question before it is a feature, and the answer matters most for anyone writing to contacts in the EU.
Our readRB2B stopped identifying people in one country on the same day lemlist started identifying them by their LinkedIn comments. Both are reasonable readings of the same uncertainty about person level data, and a buyer can easily end up with both in one stack. The controls that shipped here, delete by API, capture by rule and mask by role, are the ones a privacy review asks for by name. Most of them are off until someone asks.
Everybody left their lane
Category lines moved, and several products now sit on a different shortlist than they did a month ago.
Fireflies.ai relaunched Fred, its notetaker, as an AI teammate with an Email Assistant, a Slack Assistant, Artifacts that turn conversations into docs, decks and reports, Knowledge Search and Daily Briefs. Fireflies says Fred now connects to more than 240 tools, with paid plans from $10 per user per month billed annually and subject to usage credits. FredBots, Scheduler and Clips are still behind waitlists.
A notetaker is now pitching for the general work assistant budget, with reach into email, Slack and the CRM. Nooks made the same kind of move from the other side, from an outbound dialing workspace toward agents that own deal and expansion work.
Dock reached from buyer facing deal rooms into rep training. Dock Learning adds Pathways, AI Tutors, AI Roleplay and an Assignment hub to the Playbooks and Courses released earlier this year. AI Roleplay simulates sales calls and scores them against editable scorecards for frameworks including MEDDIC, SPIN and Challenger.
11x launched Scout List, which turns a plain English audience description into a search plan the user can review, then builds a qualified lead or account list from database records, the customer’s CRM, identified website visitors and live web research. Each result shows the rule and evidence used. A finished list can be exported or used as a campaign audience that adds new qualified leads automatically.
Breakout put 35 buying signals across 10 categories into one account view, matched to the customer’s CRM and target account lists. The set spans website visitor identification, Bombora and G2 intent, Clay enrichment columns, AI account research, job moves, funding, technographics and warm intro paths. That makes signal coverage a direct comparison with the dedicated intent platforms, and whether the Bombora and G2 data is included in the price is the first thing to pin down.
Granola turned meeting notes into standing context. Agent Briefs give connected AI agents such as Claude and ChatGPT a snapshot of the user’s work, written by Granola from their meetings and rewritten every night. A brief covers priorities, projects, people, jargon and communication style, and users can read and correct it in the Connect tab. It is a useful idea. It is also a nightly profile of people and projects sent to each connected agent, which is the part an admin will want to look at first.
Our readA notetaker sells a teammate, a dialer sells agents, a deal room sells training and an outbound agent sells list building. Each move makes sense alone. Together they mean the category name on a homepage now describes where a product started, and the shortlist for almost any GTM purchase got longer this month.
Market notes
Cold Navigator has closed. Its website now carries only a notice that the service stopped operating as of November 6, with no migration path, export period or successor product. Former customers should assume campaign data is no longer reachable.
Mautic published the last security release for the 5.x series, which ends free community security support for Mautic 5. Free security support for Mautic 6 ended on September 30, so self hosted installs on either version now face an upgrade to Mautic 7 or the cost of paid Extended Long Term Support.
Signado retired Deep Dive, its on demand scan of company events. Warm leads from LinkedIn conversations and Hiring sources work as before. Six days later it reset its credit packs to four sizes, from 1,000 credits for $35 to 10,000 for $249, and the 25,000 and 40,000 credit packs are gone.
Two migrations now have dates. Sendr will retire its older Automation Builder on December 1, and running automations must move to the Sequencer before then. Consensus released a native HubSpot app that connects through OAuth 2.0 and syncs Demo Views and New Lead events to matching Contacts, Companies and Deals. Consensus states that this is the HubSpot app it will support going forward, with the earlier integration planned for deprecation no later than April 2027.
Pricing kept moving toward the unit. GetProspect now charges phone number lookups and email verifications to a workspace wallet, a US dollar balance shared by the team, while email finding still uses plan credits. A phone number costs $0.12 when found and verification costs $0.80 per 1,000 emails. With 20% of each plan payment added to the wallet, a $49 plan adds only $10, so a team that calls a lot should expect to top up.
Swarmhit changed seat packaging in Smart API 1.26.0 so that a seat covers one LinkedIn account or one WhatsApp number, while mailboxes still take none. Linking a new WhatsApp number now needs a free seat, and a workspace over its plan’s seats has its newest accounts suspended until a seat is free, which can pause a live campaign.
Map My Customers released a Field Sales Performance Study of 172 outside sales reps at Jasper Engines & Transmissions, run with a visiting researcher at NYU. According to the company, top performing reps averaged $83,500 in quarterly revenue growth against $48,200 for lower performing peers, a 73% gap it ties to daily tool usage. It is a vendor sponsored study of one customer and shows correlation only. It is also a named customer and a sample size, which is more than field sales software usually prints.
What these weeks say about the space
Sixty two changes, one per vendor, and the theme is the pen. In August the news was that a CRM could be reached from a chat window. In October it was the permission slip and then the allowance. In November the outside assistant writes to the record, builds the sequence and edits the messaging settings, and the release notes spend most of their words on who lets it.
The send button is the line that held. Nearly every product that handed over write access kept a person between the draft and the customer. The ones that removed the person moved the decision into a trigger.
Access itself is turning into table stakes, included in the plan and listed in a directory. That moves the competition to what an agent finds once it is connected: which objects it can touch, in which direction, under whose role, and what the documentation says happens when a write fails.
So the guest list is settled, and the seating chart is what matters now. The count below is how many vendors have written theirs down.
Which GTM vendors document what their integrations actually do?
About seven in ten. Of the 414 vendors the GTM Tech Index grades on integration depth, 297 document it at a standard a buyer can check, 72 percent. Of those, 50 reach the top of the scale, which means documented bidirectional sync against named objects, a public API with real reference material and a marketplace listing that matches the claims.
Another 103 present their integrations as logos, with depth, direction and limits left unstated. That is 25 percent of the index. A further 14 make claims that could not be verified in any marketplace, doc set or API reference.
The spread by category is wide. Conversation Intelligence is the best documented, at 91 percent, and LinkedIn & Social Selling the thinnest, at 57 percent.
The question got sharper this period because the reader of the documentation changed. A person can find out during implementation that a sync only runs one way. An agent with write access finds out by writing.
The connectors in this issue that named their objects, their defaults and their limits gave a buyer something to test before the first write. A switch for Create and another for Update on each record type is a sentence a security review can check. A logo is not.
A grade on this axis records what a vendor has published, so a low one means a thin public record and not a broken integration. Every vendor at the top of the scale, category by category, is on the integration depth page.
The GTM Tech Index Brief is published by GTM Tech Index, an independent reference for evaluating the software revenue teams use to find, win, and keep customers. No vendor pays for inclusion, placement, or rating. Compare any indexed vendors by capability at Compare and read the evaluation standards at Methodology.