Salesforce Named Its Outbound AI Agent "Hunter." Here's the Governance Checklist Lean Teams Need Before Trying the Same Thing.

On September 11, 2026, ahead of Dreamforce (September 15-17), Salesforce introduced Agentforce 360 with seven named, "job-ready" AI agents: Casey (internal help and service requests), Paige (IT and HR support), Carter (shopper and commerce interactions), Marshall (supply chain operations), Piper (inbound sales pipeline), Fin (customer experience) - and Hunter, an autonomous outbound sales agent. (Kurums, September 2026)

Hunter is the one worth stopping on. It is Salesforce's name for an agent that does the work a human SDR does - researching accounts, writing outreach, sequencing follow-up - autonomously. And as of launch, Hunter is running in pilot only, through November 2026. (Kurums, September 2026)

That pilot label matters more than the launch headline. Salesforce has more enterprise trust infrastructure, more compliance staff and more incentive to get this right than almost any vendor in the category - and it is still not letting Hunter run unsupervised in the wild. If Salesforce is hedging on autonomous outbound, a lean team evaluating any AI SDR tool should be asking exactly why, and building the same hedge into its own rollout.

What Salesforce built alongside the agents

The more interesting part of the Agentforce 360 launch isn't the agents - it's what Salesforce shipped to supervise them. The "AI Control Plane" is operational infrastructure that registers every agent (Salesforce's own and third-party), sets identity and access policy per agent, manages each one's lifecycle, evaluates its performance, observes its behavior in production, and tracks its costs - explicitly positioned as governance that spans Salesforce and non-Salesforce AI systems, not a walled-garden feature. (Kurums, September 2026)

Salesforce built that because it has seen where this goes wrong. Its own stated risk framing is blunt: the biggest adoption danger is "agent sprawl without governance - deploying multiple task-specific agents faster than security, compliance and IT teams can oversee what each one can access and do." (Kurums, September 2026) Industry forecasts cited alongside the launch put the pace of that sprawl in context: task-specific AI agents are expected in close to 40% of enterprise applications by the end of 2026, up from under 5% in 2025. (Kurums, September 2026)

A lean team is not deploying seven named agents across an enterprise. But the sprawl risk shows up at smaller scale too, faster than you'd expect: one outbound agent, one enrichment agent, one meeting-scheduling agent, each configured by a different person, none of them logged anywhere a founder or head of GTM could review in five minutes.

Why "cold email is dead" already told half this story

Nukipa covered the demand side of this shift in July: AI SDR volume flooded inboxes, mailbox providers cracked down, and 2022-era cold email started getting rejected outright^[1], forcing a budget reallocation away from raw outbound volume. (^[1]) Hunter is the supply-side mirror of that story - a major vendor formalizing exactly the kind of agent that produced the volume problem in the first place, but wrapping it in enterprise governance that most of the tools flooding those inboxes never had.

That is the actual lesson from Agentforce 360, and it's not "avoid AI SDR agents." It's "the deliverability and reputation damage came from agents running with no oversight layer, and the fix everyone is now converging on is oversight, not abstinence."

Gartner's prediction that over 40% of agentic AI projects will be canceled by 2027^[2] - covered here separately - names the same root cause: runaway cost, unclear value and weak risk controls, not agents that fundamentally don't work. (^[2]) Hunter's pilot status and the Control Plane built around it are Salesforce's answer to that exact prediction, built before the cancellations happen instead of after.

The governance checklist a lean team actually needs

You will not build an AI Control Plane. You do not need Salesforce's version of this. You need five things that plane is a heavyweight answer to, in a form a lean team can implement in a sprint:

1. A written scope for what the agent may do without a human. Not "handles outbound" - specifically: which accounts, which channels, whether it can send without review, whether it can reply to inbound responses autonomously or must hand off.

2. An identity and access limit, enforced technically, not just in a policy doc. If the agent has a CRM login, that login should not have write access to fields it doesn't need, and it should be revocable in one click, by one person, without a support ticket.

3. A cap on autonomous actions per day, reviewed weekly. Volume is exactly what broke cold email in the first place. An agent that can send unlimited messages the moment it's configured is a liability, not a feature - cap it low, and raise the cap only after you've reviewed a week of its output.

4. A log a non-technical person can read. Not raw API logs - a plain-English record of what the agent sent, to whom, and why it chose that account, reviewable by whoever owns GTM without engineering help.

5. A defined kill switch and a named owner who can pull it. One person, not a committee, empowered to turn the agent off the moment something looks wrong, with no approval chain in between.

That list is the lean-team version of Salesforce's Control Plane and Hunter's pilot status combined: scope, access, volume caps, visibility, and a fast off switch. Build those five before you turn any outbound agent loose - whether it's Hunter, a competitor's agent, or something assembled from a stack of point tools.

The point

The headline from Agentforce 360 is "Salesforce shipped seven AI agents." The actual signal is narrower and more useful: the vendor with the most to lose from getting this wrong still won't let its own autonomous outbound agent run without a pilot label and a purpose-built governance layer watching it. That is not a reason to avoid AI SDR agents. It is a reason to build the oversight first and the automation second - in whatever order your vendor's roadmap prefers.

Nukipa's approach keeps that order intact: GTM automation that runs against your own governed account and pipeline data, with every action traceable to a reason a human could sign off on. If you want to see how that oversight works before you delegate outbound to anything, test Nukipa.