INFRASTRUCTURE
Agentic AI: It’s a boss, but not as we know it
by Arthur Goldstuck: Founder and CEO of World Wide Worx, author, speaker, trainer.
Agentic AI has taken the world by storm this year, and now we must welcome the Agent-in-Chief.
Your AI assistant wants a raise. It’s working overtime managing five other AI assistants, each more annoying than the last. One books your meetings at 2am, another insists you really need a wellness webinar. The assistant has had enough – and nominates itself for promotion.
It turns out, the machines are running into the same problem humans do: too many workers, not enough leadership. As the new concept of AI agents sees these virtual assistants multiply across the workplace, they will need coordination, structure, and a chain of command.
One of the world’s leading customer relations software makers, Salesforce, has given that role a title: the Agent-in-Chief.
While the scenario sounds absurd, it mirrors a growing reality inside the most AI-driven companies. Mick Costigan, who leads the Salesforce Futures team within the CEO’s office, regards it as a structural inevitability.
“We’re seeing not just standalone assistants, but fleets of agents surfacing in enterprise environments,” he told Gadget in a telephonic interview from San Francisco. “The next step is to coordinate them.”
That coordination is no longer theoretical. Agents can now schedule meetings, answer service queries, generate sales leads, and manage logistics. What’s missing is a layer that knows which alerts to prioritise, which tasks require judgment, and when to prompt the user. According to Costigan, this is the role of the Agent-in-Chief: a central digital entity, trained on an individual’s working patterns, that directs the flow of information and decisions.
“It’s more like a Chief of Staff: someone who knows when to bring issues to your attention, what context to include, and how you prefer to make decisions.”
Clearly, this is not a virtual secretary. It’s a high-level agent that manages the rest of the agents, handling complexity that lone humans struggle to absorb.
Salesforce’s Agentforce platform, launched to help companies build and deploy intelligent agents inside its ecosystem, provides one possible approach. Drawing on the company’s Data Cloud – which integrates structured and unstructured data from sources like Snowflake, BigQuery, and Slack – agents can access a broad view of customers, employees, and processes. With the right guardrails in place, they are beginning to reason, plan, and act.
“The Agent-in-Chief will be grounded in much more personal information: when you’re most productive, how you respond to different types of input, even biometric data if you choose to connect it. It’s the next frontier in enterprise intelligence.”
Salesforce’s approach shares some similarities to Google’s new Agentspace, but is rooted in its long history with customer data and enterprise access control. Costigan agrees that interoperability is essential.
“Our customers use both platforms. They expect these systems to work together.”
He envisions a structured ecosystem of agents, each with a specific domain: sales, marketing, logistics, service. At the top sits the Agent in Chief, orchestrating action across functions. As this structure matures, it begins to mirror a corporate hierarchy. Agents operate at multiple levels – some handling repetitive tasks, others synthesising information or managing strategic workflows.
This evolution changes how tasks are performed, how decisions are surfaced, and how workflows unfold. Rather than displace human leadership, the Agent in Chief reshapes its workload.
“Think of it like how the body works,” said Costigan. “Digestion, blood flow, temperature control – all happen autonomously. You only focus on what really needs your attention.”
The model introduces new demands. The Agent-in-Chief will require more computing resources, broader access to data, and deeper integration into enterprise systems. It must also respect access rights – both to protect sensitive data and to ensure agents operate within clearly defined boundaries. Because Agentforce is built on the existing Salesforce platform, it inherits long-standing permissions structures.
Researchers are now exploring the hierarchy itself. Some agent systems involve oversight loops, where one agent checks the output of others. Others experiment with decision layers similar to executive management. Salesforce doesn’t prescribe a single configuration but expects structures to evolve.
That structure won’t look the same in every business. A roadside motel might automate most functions. A luxury hotel may prefer more human discretion. The same tools will serve different strategic goals, depending on how companies define value, control, and trust.
“We think humans will still be needed for quite a while,” said Costigan. Originally published on Gadget
Useful resources:
Gadget is South Africa’s leading online magazine on consumer technology, office hi-tech and general gadgetry that makes our lives both easier and more complicated.