At a recent TechCentral executive roundtable co-hosted with Clickatell, the conversation on agentic AI moved past tooling to something more important: consequence.

Agentic systems don’t just assist humans. They take actions – routing, deciding, recommending resolving and increasingly transacting – inside customer journeys that touch revenue, risk and reputation. When those actions run at scale, small choices stop being small.

Contact Clickatell to learn more

One line captured the mood in the room: pull a lever in one part of the business and you move everything else. A routing rule, a scoring threshold, an automated response might look harmless in isolation. In production, operating at speed, it can quietly reshape outcomes across teams and channels.

That’s why the old question – should we adopt AI? – already feels dated. Adoption is happening. The sharper question is: what outcome are we delegating, and what are the limits of that delegation?

Several leaders returned to the same reality: AI multiplies scale in both directions. If automation helps a team do 10 times more work, it can also multiply the blast radius – fraud exposure, customer harm, regulatory breach and reputational damage. Scale doesn’t only accelerate wins; it compounds mistakes.

Governance is an operating model

This is where governance stops being a compliance checkbox and becomes an operating model. The most useful questions weren’t about features. They were about ownership.

What are the agent’s KPIs – and do they match the business outcome we actually care about?
Who defines success and trade-offs (speed vs accuracy, automation vs escalation, growth vs risk)?
Who is accountable when the system makes the call and the impact shows up somewhere else?

A standout theme was the growing role of AI-powered chat as a point-of-sale channel. Chat is no longer only a service layer. For many organisations it is becoming a frontline commercial surface – where customers discover, decide and transact. When a system can recommend, route, resolve and take payment, “who owns this?” becomes a board-level question, not an IT detail.

And it’s often the ownership question that slows progress. Many organisations have pilots and pockets of experimentation, but struggle to move into production because the blockers aren’t technical. They’re organisational: competing priorities, unclear decision rights, risk teams brought in late, and teams optimising for different definitions of “success”. Until those fundamentals are addressed, agentic AI stays a topic of conversation rather than a capability the business can trust.

Then there’s the backend reality. Large organisations still run on legacy platforms built for a different era. Many promising AI initiatives fail not because the model is weak, but because the environment can’t support it – fragmented data, brittle workflows, limited integration and processes that were never designed for real-time autonomy. The challenge is integrating intelligence into systems that already operate at scale, already carry consequence and can’t simply be ripped out and replaced.

There was also a point the room acknowledged quietly: people are already using these tools, often informally. The risk isn’t that organisations are behind on adoption. The risk is that usage is happening without structure – without shared governance, clear guardrails or consistent handling of customer data and operational decisions.

One of the sharper competitive observations was this: somewhere, an AI-first company is being built without the technical debt and process sprawl that slows incumbents down. They’re designing for compliance, auditability and scale from day one. That doesn’t just introduce new technology – it sets a new operating baseline.

Trust is a design requirement

Trust came up repeatedly, but not as a marketing idea. As a design requirement. In some use cases, automated decisions can outperform humans. The question is whether organisations understand which use cases are safe to delegate, how to monitor them and what customers will accept. Ethics and governance aren’t side conversations. They shape how agentic systems are designed, deployed, audited and corrected when they get it wrong – especially in regulated sectors where customer protection is non-negotiable.

The takeaway was simple, and slightly uncomfortable: agentic AI doesn’t fix unclear thinking. It amplifies it. Clarity comes first – purpose, ownership, limits and proof. The leaders who win won’t be the ones moving fastest in pilot mode. They’ll be the ones who can move from assisted chat to outcome-driven, agentic journeys with control, credibility and measurable impact.

Agentic pilots are easy. Production is hard. Contact Clickatell to discuss what’s holding your organisation back and how to establish the ownership, guardrails and trust needed to move forward.