Agentic AI · BraivIQ AI Blog

Salesforce Just Launched Seven Named AI Agents - But The Real Story Is The Control Plane: Why 2026 Is The Year Enterprises Learned To Govern Agent Sprawl

On 11 September 2026, days before Dreamforce, Salesforce unveiled seven named, function-specific AI agents - Casey, Paige, Carter, Hunter, Marshall, Piper and Fin - each built to do a specific job in service, IT/HR, commerce, sales and supply chain. The named agents grabbed the headlines, but the far more important announcement sat quietly beside them: an AI Control Plane and a Trusted Enterprise AI Harness designed to register, govern, monitor and cost-control every agent running in a business - including ones not built on Salesforce. That is the tell. The industry has realised the real enterprise problem in 2026 is not a shortage of agents; it is dozens of ungoverned agents from different vendors already running inside one company. This featured analysis explains what Salesforce actually launched, why the control plane matters far more than the seven named agents, and what it means for any UK business now deploying AI agents.

September 14, 2026 · 13 min read · By BraivIQ Editorial

7 agents - Named, function-specific Agentforce agents launched: Casey, Paige, Carter, Hunter, Marshall, Piper and Fin · Sept 11 - 2026 - the launch date, timed just ahead of Dreamforce (15-17 September, San Francisco) · Control Plane - The quieter, bigger announcement: register, govern, monitor and cost-control agents across Salesforce AND third-party platforms · Dozens - Of ungoverned agents from different vendors are the real enterprise problem in 2026 - not a shortage of agents

On 11 September 2026, days before its Dreamforce conference, Salesforce made the biggest enterprise-AI announcement of the month - and most of the coverage focused on the wrong half of it. The headline-grabbing part was seven named, function-specific AI agents: Casey for help and service, Paige for IT and HR, Carter for shoppers and commerce, Hunter for outbound sales, Marshall for supply chain, Piper for inbound pipeline and Fin for customer experience. Giving agents human names and specific jobs makes for a great demo and an easy story. But sitting quietly alongside them was a far more consequential announcement: an AI Control Plane and a six-pillar Trusted Enterprise AI Harness built to register, govern, monitor, evaluate and cost-control AI agents across an entire organisation - explicitly including agents not built on Salesforce. As an AI Agency London that builds and governs AI agents for a living, we think that second announcement is the real story of 2026, and this featured analysis is why.

The Real Story In One Paragraph: The exciting bit of Salesforce's announcement is seven named AI agents that each do a specific job. The important bit is the control plane sitting beside them - a layer to register every agent, give it an identity, set what it is allowed to do, watch what it actually does, evaluate whether it is working and control what it costs, across every platform, not just Salesforce. That combination is a tell about where enterprise AI has actually got to. The problem in most large companies in 2026 is no longer 'we need an agent'; it is 'we already have dozens of agents from different vendors doing things across our systems, and nobody can see, govern or account for them all'. The named agents are the product. The control plane is the admission of what the market now actually needs. Any business deploying agents should read it the same way: the hard, valuable part is not launching an agent - it is governing the ones you already have.

What Salesforce Actually Launched

It helps to separate the two layers clearly. The first layer is the agents themselves: seven pre-built, purpose-specific agents with names, each aimed at a common enterprise function so a buyer can point at 'Casey' for customer service or 'Marshall' for supply chain rather than assembling something from scratch. This is a smart packaging move - it turns the abstract idea of 'an AI agent' into concrete, hireable-feeling colleagues with defined jobs, and it lowers the barrier to getting started. The second layer, and the strategically important one, is the governance stack: the Trusted Enterprise AI Harness, a six-pillar framework for deploying agents safely, and the AI Control Plane, a single place to register agents, set their identity and policy, manage their lifecycle, evaluate their performance, observe their behaviour and control their cost. Crucially, Salesforce designed the control plane to govern agents across Salesforce and third-party platforms alike. A company that sells agents does not build a cross-vendor control plane unless it has concluded that the customer's biggest pain is no longer buying agents - it is managing the sprawl of agents they already have.

Why 'Agent Sprawl' Is The Defining Enterprise AI Problem Of 2026

Rewind eighteen months and the enterprise AI conversation was about capability: could an agent actually do the job? In 2026 that question has largely been answered - agents can do real work - and a new, less glamorous problem has taken its place. Every major software vendor now ships agents. Your CRM has agents, your service desk has agents, your data platform has agents, your marketing suite has agents, and your own developers have built a few. The result inside a typical large organisation is agent sprawl: dozens of semi-autonomous systems, from different vendors, each with some access to data and systems, each taking actions, and no single place to see them all, know what they are permitted to do, or account for what they cost and whether they are behaving. This is exactly the situation that Gartner has been warning creates both runaway cost and unmanaged risk, and it is precisely the gap Salesforce's control plane - and rival control planes from the likes of Boomi - are racing to fill. The reason this is the defining problem is simple: ungoverned autonomy does not scale. You cannot responsibly let dozens of agents act across your business if you cannot see, constrain and audit them.

Identity - every agent should have a distinct identity, so you know which agent did what, just as every human user does.
Permission - each agent should be scoped to least privilege: only the data and actions it genuinely needs, nothing more.
Observability - you must be able to see what agents are actually doing - their actions, tool calls and decisions - not just trust that they are fine.
Evaluation - agents should be measured on whether they are actually working and delivering value, not assumed to be.
Cost control - autonomous systems can quietly rack up enormous token and orchestration costs; someone has to see and cap that.
Human oversight - the consequential decisions (money, contracts, customer-facing actions) stay gated behind human approval.

What This Means For UK Businesses Deploying Agents

You do not need to be a Salesforce customer, or an enterprise, for this to matter to you - the lesson generalises to any business putting AI agents into production. First, treat governance as a first-class part of every agent project, not an afterthought: from day one, give each agent a clear identity, scope its permissions to the minimum it needs, and make sure you can see what it does. Second, keep humans firmly in control of the decisions that carry real consequences - approvals, money, contracts, customer-facing commitments - while agents handle the drafting, routing, researching and volume. Third, measure and cap cost, because autonomous systems that loop or over-call can burn money invisibly. Fourth, think in terms of a single control point rather than a scatter of disconnected agents, so that as you add more you are extending a governed system rather than accumulating ungoverned sprawl. For most UK businesses the practical starting point is not 'which named agent do we hire' but 'how do we deploy agents in a way we can see, trust, govern and afford as we scale' - and that is precisely the work we do as an AI Agency London: building agents that do real work and stand up to real governance.

The Bottom Line

Salesforce's seven named agents are a clever, accessible way to package AI agents as hireable colleagues, and they will help a lot of businesses get started. But the announcement that really tells you where enterprise AI has got to is the control plane sitting beside them - a cross-vendor layer to register, govern, observe, evaluate and cost-control agents across the whole organisation. A company that sells agents does not build that unless the market's real problem has shifted from getting agents to governing them, and it has. Agent sprawl - dozens of ungoverned, semi-autonomous systems from different vendors acting across a business - is the defining enterprise AI problem of 2026, and control planes are the answer the industry is converging on. For UK businesses the takeaway is clear and useful: the hard, valuable part of AI agents is not launching one, it is deploying them so you can see, trust, govern and afford them as they multiply. Get the governance right and agents become a durable advantage; ignore it and you are simply accumulating risk with a friendly name. As an AI Agency London, building agents that are both genuinely useful and genuinely governable is exactly what we do.

References & Further Reading

Salesforce - Agentforce 360 announcements and what's new: https://www.salesforce.com/agentforce/what-is-new/
Constellation Research - Salesforce makes its Agentforce 360 case to be your AI agent platform: https://www.constellationr.com/insights/news/salesforce-makes-its-agentforce-360-case-be-your-ai-agent-platform
AI Agents Directory - AI Agents News Brief, September 13 2026: https://aiagentsdirectory.com/news/ai-agents-news-brief-september-13-2026
Gartner - by 2026, 40% of enterprise applications will embed task-specific AI agents (agent adoption and governance): https://www.gartner.com/en/newsroom
Boomi - Agent Control Plane for governing how AI agents access systems, data and resources: https://boomi.com/