Agentic AI represents the next stage of artificial intelligence—and a significant shift in its capabilities. Unlike traditional ML models and generative AI, agentic systems operate autonomously, using tools to achieve outcomes without human guidance. The self-sufficient nature of this technology has companies across various industries excited… and nervous.

A number of our clients are exploring AI agents, either as product offerings or to enhance their own business operations. As such, we’ve had a lot of conversations about agentic systems; and one thing we’ve noticed in our research is the interesting overlap between agent preparation requirements and brand marketing. (No, really.)

Over the years, we’ve used a set of questions in brand and positioning workshops to help businesses articulate intent when it feels too abstract to capture. As it turns out, the same questions work surprisingly well for agentic systems—because they force companies to make firm decisions on outcomes, constraints, and trade-offs before a single workflow is designed.

Used alongside typical agent guidelines, these questions can help businesses further refine their expectations of (and requirements for) their agents. These questions aren’t technical, but they do a good job at exploring intent, guiding behaviour, and identifying both surface and underlying preferences—and it’s that clarity that separates agentic systems that will scale safely from those that will create new problems faster than they solve old ones.

THINKING ABOUT SUCCESS

Marketing parallel: “Who are we for, and what promises are we making?”

In brand and positioning work, defining success starts with identifying the core promise and the audience it serves. With agentic systems, we’ve noticed the same exercise applies—but the stakes feel higher, because the agent will optimise ruthlessly for whatever metric you give it. Picking the wrong one (or leaving it vague) seems to create immediate downstream risk.

In our conversations with clients building agent products, we’ve noticed that the most useful exercise hasn’t listing every KPI; it’s identifying the single north star that aligns customer value with business outcomes, then stress-testing what happens when that metric is pursued in isolation.

Worth considering:

• If the agent can only optimise for one metric, what would best represent success—and what could be the consequence of that decision?
• What does “good” look like to the customer vs the business? And which one wins when they conflict?
• What would make you proud to put the agent’s decisions on the front page of the Financial Times?

NOTICING WHERE THE LINES SHOULD BE DRAWN

Marketing parallel: “What’s on-brand vs off-brand? Where are our non-negotiables?”

Brand guidelines exist to prevent tone and messaging drift across channels and markets. In our brand work with clients who are building agent products, we’ve seen the same logic apply—except instead of tone and colour palettes, the guardrails become operational red lines.

The pattern emerging in conversations is this: surface the actions that might be technically compliant or commercially attractive, but that would erode trust or brand credibility if customers saw them. Think of it as defining the agent’s ethical and reputational boundaries before it ever encounters an edge case.

Consider:

• What actions would feel like a betrayal of the brand promise—even if it’s technically compliant?
• Which customer segments or scenarios should the agent never touch without human oversight, and why?
• What channels, tones, or frequencies would feel intrusive or inappropriate for your brand, even if they convert?

OBSERVING TRADE-OFFS IN ACTION

Marketing parallel: “How do we behave under pressure? What do we value when our goals conflict?”

No operational system runs without tension: speed vs safety, conversion vs trust, efficiency vs empathy. In brand strategy, clients typically define how the brand behaves when stakes are high. With agentic systems, the pattern we’re seeing is that these trade-offs can’t be left to chance; the agent will make a call one way or the other.

In our conversations, the useful question is whether that call is one the business could defend in the boardroom. It’s less about eliminating conflict and more about making priorities explicit so the agent doesn’t invent its own.

Useful questions include:

• When speed and safety conflict, which one wins? And at what threshold (if any) does that flip?
• What’s your stance on false positives vs false negatives in this context?
• If the agent had to explain its decision to a regulator, a journalist, or a frustrated customer, could it do so convincingly?

BRINGING THE HUMAN BACK INTO THE LOOP

Marketing parallel: “When do we know a customer needs a human—and what does that experience feel like?”

Even the most autonomous system will encounter scenarios it can’t (or shouldn’t) handle alone. In CX and messaging work, clients often note that the handoff moment is critical: a clunky escalation feels like failure, while a seamless one feels like care. The same pattern appears with agentic systems.

In our conversations, the emerging insight is that you need to define not just when the agent steps back, but how that transition should feel to preserve trust and continuity. This is where the “please repeat your account number” frustration tends to hide—and where confidence in the entire system can quietly erode.

Useful questions include:

• What signals should trigger a human handoff? How should that transition feel to the customer?
• What context must an agent pass to the human so the customer doesn’t have to repeat themselves?
• When should the agent recommend human review even if it’s confident—and why?

NOTICING CONSISTENCY ACROSS TOUCHPOINTS

Marketing parallel: “Does this feel like us across every touchpoint?”

Brand consistency isn’t about removing creativity; it’s about ensuring coherence. One of the key tenets of marketing is that customers should be able to recognise the brand whether they’re reading an email, walking into a branch, or stumbling across an ad online—and the same principle should cover interacting with agents.

With agentic systems, we’re observing that any inconsistencies compound faster, because the system operates at scale and speed. Anchoring an agent’s behaviour to existing brand attributes seems to help ensure it feels like a natural extension of the business rather than a disconnected automation layer.

Useful questions include:

• If a customer interacted with the agent for six months, would they describe your brand consistently?

• What three adjectives should customers use to describe the agent’s behaviour—and what would violate those?

• Where does the agent sit on your brand’s spectrum of formal/casual, proactive/reactive, concise/detailed?

BRINGING LEFT-BRAIN AND RIGHT-BRAIN TOGETHER

When agentic systems fail, it isn’t due to a lack of intelligence—it’s because they’re given vague or conflicting marching orders. As organisations begin to hand more decision-making authority to autonomous systems, the work of clarifying what “good” looks like (and what is never acceptable) becomes foundational.

Marketing teams have spent decades grappling with these questions in public-facing contexts where trust, coherence, and reputation are on the line. In our client conversations, the emerging view is that borrowing that discipline doesn’t turn brand into a control function, but it does offer a proven way to surface priorities and critical fault points before autonomy scales.

These questions aren’t about marketing taking ownership of AI. They’re about noticing that the same clarity needed for brand strategy is also what keeps agentic systems from going off-piste.