Salesforce today published an opinion piece by Kishan Chetan, executive vice president and general manager of Agentforce and Service Cloud, arguing that personal AI agents acting for consumers will send more traffic to business contact centers than they remove - a forecast the company makes without a survey, a sample, a call-volume figure or a date by which the surge is expected.

In Short

Salesforce's head of customer service software wrote that the AI assistants people use to cancel subscriptions or chase deliveries will end up phoning companies far more often than humans do, because they never get bored on hold and never sleep. That matters to any business running a call center, since voice is the most expensive way to serve a customer and the extra calls would land on someone's budget. The piece offers no numbers to size the problem, and its proposed answer is for businesses to put their own AI agents, which Salesforce sells, in front of the incoming ones.

What Salesforce published

The article sits on the Salesforce Newsroom under the headline "Personal Agents Will Drive More Contact Center Traffic Than They Offset," with the subhead "Personal agents can start a customer's request, but the business still has to finish it." It is dated October 19, 2026, carries Chetan's byline, is labelled a four-minute read and is filed under the topics Agentic Enterprise, Agents, AI, and Customer Service and Support.

It is a short essay, roughly 800 words of body text, and the second half is product description. That structure matters for how much weight the forecast can carry. The headline makes a quantitative claim - more traffic than offset - yet the body contains no quantities. There is no estimate of how many consumers use personal agents, no share of service contacts currently started by an agent, no comparison of calls deflected against calls generated, and no reference to any Salesforce customer's call logs. A check of the PDF's embedded hyperlinks found eight distinct links in the body text, none of which points to a dataset or study. They lead to Salesforce product pages, a Salesforce help article, Salesforce's Gartner report page and an earlier Salesforce Newsroom essay.

That earlier essay, "Your Customer's Personal Agent Is Knocking. Is Your Business Ready to Answer?", was published on October 9, 2026, by Mark Wakelin, executive vice president of customer engineering and general manager of Agentforce. Its only statistic is a founder's claim, cited from a TechCrunch report, that the invite-only agent Instinct handles close to $1 billion in annual transactions. Chetan's piece is the second Salesforce executive essay on personal agents in ten days.

The argument

Chetan opens with what personal agents already do. "From cancelling subscriptions to booking restaurant reservations, consumers are starting to hand a wide variety of work to personal AI agents," he wrote. "So far, most of it doesn't require a phone call; The agent navigates a website, calls an API or works things out with a business's own agent."

The case turns on what such an agent does not know. "A personal agent might have access to a user's inbox and any information they've explicitly provided, but it likely doesn't have the full picture of their relationship with a business," according to Chetan. "Even cancelling a subscription could mean going back and forth with a company's agent or a human rep who has access to the business processes and workflows." He adds a second obstacle: "Personal agents also hit roadblocks like security checks designed to keep bots out, or questions only a human representative can answer."

The core of the piece is an argument about behaviour at scale rather than about any single interaction. "Personal agents don't mind waiting on hold or calling for order updates or reservation confirmations as often as they're asked to," Chetan wrote. "A person might check on a delayed order once. An agent asked to keep an eye on it can call back every few hours until the order ships. An agent shopping for a better phone plan can contact several providers in the time it takes a person to dial one." The timing changes as well. "Because it doesn't keep business hours, its calls can arrive at 2 a.m. as easily as at noon."

His conclusion is that this compounds. "Multiply that by every consumer who hands an errand to an agent and the volume quickly becomes astronomical," he wrote. Contact centers already handle seasonal spikes, Chetan notes, "but personal agents could turn peak volume into the new status quo. And since voice is already the most expensive engagement channel, organizations will want to be ready for a potential surge of agentic traffic."

The hedges in that passage are worth reading closely. Agents "could" turn peak volume into a constant; the surge is "potential." The headline is categorical. The body is conditional.

What the forecast rests on

The mechanism Chetan describes is plausible on its own terms. An automated client that polls a business every few hours, or contacts several providers in parallel, generates more contacts per consumer intent than a human does. Whether that outweighs the contacts personal agents remove - the routine status checks they resolve through a website or an API without anyone picking up a phone - is the empirical question the headline answers and the article does not test.

Several variables would decide it, and none is quantified. How many agents use voice at all, when most of the tasks Chetan lists can be done through web forms or APIs? How often do businesses block or throttle repeat automated callers? What share of agent-initiated contacts reach a human rather than a business's own automated system? The piece is silent on each. It also does not distinguish between contacts that reach a human representative, which carry the high cost Chetan refers to, and contacts absorbed by an automated layer, which cost far less per interaction.

There is also a commercial context that the article does not state but readers can see. Salesforce sells the software that the essay recommends as the response. The forecast and the remedy come from the same vendor, in the same document.

The remedy Salesforce proposes

"The best way to prepare is to meet personal agents at the front door with your own business agents," according to Chetan. The reasoning is that a personal agent "doesn't know the loyalty tier the customer qualifies for or what has to happen before a change goes through, so it improvises and often fails to get the job done," while the business "likely already has an established process in place." The business agent, in his words, brings "your defined processes and workflows, as well as valuable context grounded in your actual business data."

From there, the piece describes Service Cloud and its components. It states that Service Cloud works "whether you're using Agentforce Contact Center or another telephony provider," and that it "brings AI agents, human expertise and trusted data together on one platform, so whoever handles a request - human or agent - starts from the customer's complete record and case history." In practice that means the customer's CRM record, which is the asset Chetan argues a personal agent lacks.

The named components are:

  • Casey, described as "your help agent," which "runs across voice, web chat, SMS, WhatsApp and Apple Messages, so a personal agent gets the same answers and workflows on any surface."
  • Agent Script, which "defines what the agent can and can't do."
  • Testing Center, where "simulated test calls" put an agent "through realistic conversations before it takes a live one."
  • Service Rep Assistant, which supports human representatives with "AI-generated, step-by-step action plans for every case, built from real-time customer data and company knowledge."

On escalation, the article states that "Casey hands off the customer with an AI-generated summary of the conversation, along with the full transcript and a recommended next action."

The piece closes by citing Salesforce's placement as "a Leader in Gartner's 2026 Magic Quadrant for Conversational AI Platforms," linking to a Salesforce page hosting the report. No pricing, availability date, customer deployment or performance figure appears for any of the products named.

What is missing from the product section

The essay's problem statement is about personal agents specifically, yet the product section does not explain how any of the named tools identifies an incoming caller as an agent rather than a person, verifies that the agent is authorised to act for the account holder, or limits repeat calls from the same agent. Those are the operational questions its own argument raises. A business agent that answers every call, around the clock, on five channels, absorbs volume. It does not by itself reduce it, and the article does not say what it costs to run.

Why this matters for marketers and service teams

The contact center has rarely been part of the conversation about agentic AI in advertising, which has centred on how agents find products, read pages and complete purchases. Chetan's essay pulls the after-sale side into view. If personal agents shop, they will also chase refunds, query deliveries and cancel subscriptions, and those contacts land on service budgets that marketing teams seldom see.

PPC Land has tracked the arrival of these agents on the buying side through the autumn. Meta put an AI agent that buys things inside WhatsApp, US only in September, running on its own cloud machine and paying with single-use cards. Within weeks, HUMAN Security estimated that Muse generated about 72% of daily AI-agent requests in its early observations, and Amazon blocked Meta's Muse agent from ordering on Amazon.com, arguing the agent operates without identifying itself. Profound's research found that Muse and Instinct browse without naming themselves, and DataDome counted Muse at 33% of agentic browser requests on its network in September.

Those findings bear directly on Chetan's argument. The "security checks designed to keep bots out" that he lists as a roadblock are the same defences - from CAPTCHA challenges to outright blocks - that retailers are tightening because they cannot tell a legitimate agent from a scraper. Earlier this year, HUMAN reported that only half a percentage point separated benign from malicious automation in its data. An agent stopped at a website's bot filter has one obvious fallback: the phone line. Identity schemes such as Experian's Agent Trust framework, which binds verified consumers to their agents, address the same gap from the other side.

The agent-to-customer-service scenario is also not new. In July 2025, PPC Land reported a Perplexity Comet user describing how the browser's agent took over a FedEx customer support chat, tracked a package and spoke with a live agent. Voice infrastructure has grown underneath that behaviour: Vapi said in May that its voice AI platform had handled more than 1 billion calls.

Salesforce's own record on agent reliability

Chetan's claim that a personal agent "improvises and often fails to get the job done" sits alongside Salesforce's own research on business agents. In June 2025, Salesforce AI Research's CRMArena-Pro benchmark found that leading agents succeeded in 58% of single-turn and 35% of multi-turn business tasks. The essay assumes business agents, grounded in company data, will perform materially better than personal agents in the same conversations. It cites no measurement of that gap.

The essay is also part of a broader product push. At its TDX conference in April, Salesforce opened its platform to external AI agents through APIs and MCP tools and open-sourced Agent Script. In September it released seven prebuilt agents for sales, service, commerce and back-office work, two of them obtained through acquisitions. On October 13, PPC Land covered a Newsroom interview in which Salesforce's chief data officer said Koa, its CRM reasoning model, was trained on 27 years of CRM workflow simulations. And on October 18, Demandbase said its intent data would be available inside Salesforce's Hunter agent without a separate contract.

Forrester's warning from October 2025 frames the risk on the business side of the line. It predicted that one-third of brands would damage customer trust with AI self-service in 2026 by deploying chatbots and virtual agents before they were ready. Chetan's essay asks businesses to deploy more of exactly that kind of agent, faster, to meet a volume of traffic it does not measure.

What would settle it

The forecast is testable. A contact center operator could report what share of inbound contacts are agent-initiated, how many repeat contacts come from the same automated caller, and whether total volume per customer has moved since consumer agents became widely available in September. Salesforce runs service software for many such operators and has not published those figures here. Until someone does, the claim in the headline - more traffic than offset - remains a vendor's argument rather than a finding. The arithmetic of agents that redial every few hours is real enough; the size of the bill is not yet known.

Timeline

Summary

Who: Kishan Chetan, executive vice president and general manager of Agentforce and Service Cloud at Salesforce, writing on the Salesforce Newsroom.

What: An opinion piece arguing that personal AI agents acting for consumers will generate more contact center traffic than they remove, because they redial, contact several providers at once and call at any hour, followed by a description of Service Cloud, Casey, Agent Script, Testing Center and Service Rep Assistant as the response. The piece cites no survey, sample, call-volume data or customer evidence.

When: Published today, October 19, 2026, ten days after a related Salesforce essay by Mark Wakelin.

Where: The Salesforce Newsroom, under the Agents and Customer Service and Support topics. The argument concerns contact centers in general and names no market or region.

Why: Consumer personal agents such as Meta's Muse arrived in force in September 2026 and are already colliding with retailer bot defences. If they fall back to phone and chat channels, the cost lands on service operations, and Salesforce, which sells contact center software, is positioning its business agents as the answer before any data on the scale of the problem has been published.