Last week's issue asked who sets the limits on an agent. This week several vendors answered by building a person into the product. MessageGears released an MCP server that lets Claude or ChatGPT build audiences and campaigns, and says the MCP "can't launch a send" (Business Wire). Amperity introduced an agent called Pér whose release says users "review, refine, and approve the next step before action is taken" (Business Wire). HubSpot's deal progression beta can suggest CRM updates or, if you let it, auto-approve them (Orange Marketing recap).
Clay went the other direction. At its Sculpt conference on October 8 it announced an agent that plans, runs, and refines a play from an outcome you describe, and a Workflows canvas where you can trace runs record by record (Clay). Zeta Global described AthenaOS, which it says starts from a business objective and assembles the data, applications, and actions to reach it (Business Wire). Underneath both approaches sits a cost question. Kyle Poyar wrote that AI pricing is moving toward usage and outcomes, and that buyers want a bill they can predict (Growth Unhinged).
Every item below carries a source link. Where a figure or capability comes from a vendor describing its own product, customer, or survey, it is labeled a claim and attributed.
The big picture
Three shifts to take into your next planning meeting
- Some vendors are writing the approval step into the product. MessageGears says its MCP "can't launch a send" (MessageGears). Amperity says Pér keeps people in control by asking them to approve the next step before it acts (Amperity). Both are vendor descriptions, not tested results.
- Others give the agent more of the loop and rely on tracing. Clay's new agent plans, runs, and improves plays, and its Workflows beta lets you trace runs record by record (Clay). Zeta lists identity, governance, permissions, security, and transparency controls as part of AthenaOS, which is in limited internal beta (Zeta).
- Pricing is moving toward usage and outcomes. Poyar reports that 76% of AI-native companies use variable or hybrid pricing, against 49% of SaaS companies, and quotes HubSpot's CEO saying customers want "predictability in what it costs" (Growth Unhinged). We did not open the data behind his figures.
Approval built in
Two vendors put the approval step in the product
Lifecycle marketing
MessageGears releases an MCP for building campaigns from Claude or ChatGPT, and says it cannot launch a send
Vendor claimMessageGears released a set of warehouse-native AI capabilities. The MessageGears MCP lets users build audiences, campaigns, and multichannel content from Claude, ChatGPT, or other MCP-compatible assistants. It ships with 16 built-in skills and is available now to MessageGears customers. The release also describes programmatic API endpoints and 10 pre-built predictive models for segmentation, channel routing, and personalization.
On limits, the release says the MCP can prepare a campaign and run a go/no-go checklist, but "it can't launch a send." The same release says the programmatic endpoints let brands compose and launch campaigns from their own AI agents, data pipelines, or orchestration tools. So the restriction applies to the MCP, not to every way of connecting an agent. The release names Chewy, GoDaddy, Indeed, and Sherwin-Williams as customers and gives no pricing, adoption, or performance figures.
Customer data
Amperity introduces Pér, an agent that prepares audiences and actions for approval
Vendor claimAmperity introduced Pér, an AI agent that takes a brand from a question about customers, campaigns, or audiences to an answer and a recommendation. Per the release, it can investigate what changed, identify affected customers, explain the evidence behind a recommendation, and prepare an audience or action for approval. It works within a brand's existing identity-resolved customer data, permissions, and enabled workflows.
Under the heading "Keep people in control," the release says users "review, refine, and approve the next step before action is taken." Amperity says more than 400 brands use its platform. The release gives no availability date, pricing, or performance figure. A webinar is set for October 15 at 10 a.m. Pacific.
Agents that do more
Three ways vendors hand work to an agent
GTM tooling
Clay announces an agent that plans and runs plays, a Workflows canvas, and a voice agent for inbound leads
Vendor claimClay's recap of its Sculpt conference reports three announcements and says the event drew 2,000 attendees from 30 countries. The first is an agent-first way to build: you describe an outcome, and an agent plans, runs, and improves the play. It draws on your ICP, personas, and positioning in Knowledge Hub and recommends ICP and segment changes based on campaign results. It is available now to Sculpt attendees and, per Clay, rolling out to early access users "in the coming weeks."
The second is Clay Workflows, a visual canvas where steps can branch, loop, run code, or call an agent. Plays can be described in natural language, traced record by record, and re-run in bulk, with results written back to Audiences. It is in beta for all Clay users and Clay says it launches "this fall." The third is an Inbound SDR Agent in preview, a voice agent that qualifies inbound leads and demos the product in a sandbox. It is not released, and companies can join a waitlist. The post does not mention pricing. Its example of an ICP suggestion, five of six booked meetings with telehealth companies, is labeled illustrative.
Marketing platforms
Zeta Global describes AthenaOS, an operating system built around a business objective, in limited internal beta
Vendor claimZeta says AthenaOS combines a company's own data with Zeta's customer insights, starts from a stated business objective, and assembles the data, applications, and actions needed to reach it. Components named in the release include Zeta Data Cloud, the Athena conversational interface, Athena MCP for use from ChatGPT, Claude, and Gemini, and the Athena Inference Model, which Zeta says it is developing with Fireworks using NVIDIA Nemotron. The release also lists an enterprise foundation covering identity, governance, permissions, security, and transparency controls.
Timing per the release: limited internal beta now, public beta later in 2026, and general availability planned for Q1 2027. It contains no metrics. Jeff Brousseau of Synchrony Financial is quoted saying "Athena has already helped our teams work more efficiently," describing agents used for quality assurance, link validation, and translation. That is a customer quote supplied in the vendor's release.
CRM
HubSpot deal progression beta can suggest or auto-approve CRM updates and build a shared buyer plan
According to the recap, HubSpot's deal progression feature uses emails, call transcripts, notes, and deal context to suggest or auto-approve CRM updates, surface next best actions, and draft follow-up emails. A new Catch-Up tab on the deal record shows next best actions, suggested CRM updates, and a re-engagement prompt when a deal has had no activity for 14 or more days. It can also track a sales methodology such as MEDDIC, BANT, or SPICED and show which criteria are met or missing.
Deal Plans generates a branded, shared buyer portal from the deal record to track milestones, owners, and next steps, and buyers do not need a HubSpot login. The recap says setup requires opting into the public beta, enabling conversation recording, and configuring CRM update preferences, and that admins must activate sales methodology and deal plans before they appear.
Pricing
What an agent costs is part of the control question
Pricing
Kyle Poyar: AI pricing is moving toward usage and outcomes, and buyers want a bill they can predict
Poyar writes that AI monetization is still unsettled and that pricing is moving from fixed access toward usage and outcomes. He reports that 76% of AI-native companies use variable or hybrid pricing, against 49% of SaaS companies. Asked which model investors prefer, 15% chose flat-fee or seat-based; the rest chose hybrid (35%), outcome-based (26%), or usage-based (24%). We did not open the data behind these figures, so read them as his reporting.
He cites examples we did not verify: Fin charges $0.99 per outcome, and Gumloop passes through tokens and compute at cost plus an 8% orchestration fee. On buyer needs he quotes HubSpot CEO Yamini Rangan from the Q2 2026 earnings call: "Customers adopting AI want proof of value before they commit and predictability in what it costs." For making a bill predictable he suggests a universal credit model, hard budget caps, annual usage drawdowns, or a platform fee plus usage. He also argues that changing pricing model touches billing, sales compensation, and forecasting, so it is a hard pivot.
Plumbing worth knowing about
Agent checkout arrives at the store and the wallet
- Newegg turned on checkout inside Google's AI Mode and the Gemini app. Per the release, U.S. shoppers can buy eligible products there using Universal Commerce Protocol checkout and Google Pay, and "Newegg is the merchant of record for every order." The release gives no sales or conversion figures (Business Wire; October 8 per the release URL).
- Crossmint launched an Agent Commerce Toolkit for agent payments. Per the release, agents can pay with tokenized Visa Intelligent Commerce or Mastercard Agent Pay credentials, described as scoped virtual cards with expiration dates and spend limits, and check out through Shop Pay, Universal Commerce Protocol, or browser automation. The release gives no pricing or success rates (PR Newswire, October 8).
The week ahead
Two dated events and several betas to watch
- Braze Decisioning Studio Go is expected to reach general availability on October 14, and Agentic Standards sometime in October. Both dates come from Braze's own release and can move (Braze).
- Amperity holds a webinar on Pér on October 15 at 10 a.m. Pacific. The date and time come from the release (Business Wire).
- Several betas are scheduled but undated. Clay says its agent reaches early access users "in the coming weeks" and Workflows launches "this fall" (Clay). Zeta plans a public beta of AthenaOS later in 2026 (Business Wire).
Put it to work
What to do with this, this week
- Write down which steps an agent may prepare and which a person must launch. MessageGears and Amperity both describe that line in their products. Check whether the tools you use draw it for you (MessageGears, Amperity).
- If your CRM can auto-approve agent updates, start with suggest only. Review a sample each week before you widen it (Orange Marketing recap).
- For any tool that runs on its own, ask whether you can trace its work record by record. Clay's Workflows beta describes that ability (Clay).
- Before you sign for usage or outcome pricing, ask for a cap and a way to forecast the bill. Name who can spend against it (Growth Unhinged).
Every claim above carries a source link. Figures attributed to vendors, to companies describing their own customers, or to surveys a vendor ran or published are their claims, not independently verified facts, and are labeled as such. Details of beta products may change. Business Wire release dates are taken from the release URL. Coverage window: October 5 to October 12, 2026. Compiled October 12, 2026.