Is trust the Achilles’ heel or the force behind agentic AI adoption?

The above conversation between my AI-agent assistant and me offers a glimpse of the future which may not be as distant as it seems. It surely is convenient and playful, but my first impression is “whoa…,” because so many trust-related questions pop into my head, all at once: How is it choosing for me? Is it influenced by ads or commission? Does it keep my personal information safe? Is it booking through official providers or a look-alike website? Who’s responsible if something goes wrong?

This little scenario involves AI acting autonomously on my behalf. It does so by simply knowing my general ‘intent’ or ‘goal’ and independently initiating a series of tasks to complete a multi-purchase process. This approach is part of what many refer to as agentic AI, particularly over the past year.

When I talk about “trust” here, I mean:

• privacy (share only what’s needed)
• security (protect my data and assets)
• reliability (consistently deliver the right result)
• transparency (explain the choices made)
• accountability (who’s responsible when things go wrong)
• governance (clear rules for what it can do)

Essentially, these are the basic conditions needed for AI to act on our behalf.

What follows are reflections shaped by our recent research and conversations with experts from AI tool and platform providers, as well as industry and ecosystem strategists, all working at the forefront of agentic AI.

Trust as a differentiator in the AI market

The AI market – and the agentic AI, for that matter – is surely crowded, with new actors popping up from every corner of the world daily. So, the golden question is: who can actually stand out in this crowd?

Right now, the major AI companies – model providers, chip vendors, and cloud platforms – hold a central position. But our research suggests that the real winners in this ecosystem will not be decided by raw compute alone. They’ll be the ones who build the trust layer around agents: the platforms and orchestrators that can set the rules, control what data and tools an agent can access, and keep an audit trail of what it actually did.

Let’s circle back to my Roman Holiday scenario. I’m no princess, but I still don’t want my AI agent to book me a tiny Roman apartment if I’d have to share it with a dodgy reporter. Even if he’s as charming as Gregory Peck! If the AI agent is going to handle my personal data, make payments, and act on my behalf, I need to know that I can trust its actions. This is exactly where the trust layer comes into play. An agent operating in a trustworthy environment, with clear safeguards and verifiable behavior becomes far more valuable for me than one that simply processes faster.

And it’s not just me feeling this way. Our interviews with agentic AI tool and platform providers repeated the same message: organizations prefer providers that offer clear agent controls and policies, transparent behavior, and even the option to run in air-gapped environments when needed. They often value these trust-centric factors over strictly superior compute performance.

This doesn’t mean the current major players in the market will lose importance. But it does suggest that they’re not automatically best placed to lead on trust in the agentic AI landscape. Other large actors, or even smaller specialist companies, could provide meaningful layers of trust around the AI agents.

Machine-readable trust

Let’s have a look at everything I’ve discussed so far, but this time through the eyes of my agentic AI assistant. I have repeatedly talked about how important trust is to me, and how I hope for a future where interactions with agentic AI become genuinely trustworthy. But how would a machine know what “trustworthy” means to me?

For AI agents to reliably act on my behalf and evaluate services, we’ll need policies and trust signals that are interpretable by machines and not just humans. Or, in marketing 101 terms, agents should be able to understand trust the same way humans feel about brand reputation.

AI agents understand trust through clear semantics, verifiable guarantees, policy metadata, and reputation scores. And because they are agents and not humans, the interactions to select providers and services, such as airlines, booking sites, and payment providers, must be policy-checked and fully auditable. There should be audit logs in place so you can see what happened, and only services that provide clear and consistent trust signals should appear in the agents’ selection flow. In sensitive areas like defense and energy, there must also be sovereign or on-prem options for cases where cloud services cannot be trusted.

That said, even if fully autonomous, agent-to-agent decision-making is the long-term vision. Our interviews showed we’re still at a stage where a human should be in the loop for consequential and risky actions. Enterprises aren’t willing to give AI full autonomy either. And this may remain true for critical tasks, even if machines become fully fluent in reading trust.

This brings us to another key question worth exploring.  

Can telcos become a trusted orchestrator?

The short answer is yes. Telcos have a strong starting point and are taking actions in the right direction. But they currently don’t have a central position in the agentic AI ecosystem.

Earlier, I talked about the value trusted platforms and orchestrators can bring to the agentic AI table. Telcos are one of the solid candidates for this role, as they already have many of the necessary building blocks.

Telcos are used to running regulated networks with strict requirements on availability and network identity. Our research shows that in sectors like defense and energy, customers often rely on telcos for secure connectivity, and, at times, private networks and on-premises deployments. As AI agents evolve into everyday services, they will depend increasingly on trusted context like location and identity: areas telcos can support reliably and at scale.

We have also seen the launch of Trusted Tech Alliance recently. It is a coalition of 16 leading technology providers, led by Ericsson and Microsoft, that aims to strengthen trust and security across the digital stack. If done well, it could become a very valuable initiative.

But at the moment, communications service providers and telecom application programming interface (API) platforms aren’t yet in a leading position in the agentic AI ecosystem. They are often viewed as part of the broader landscape of API and tool providers. This means the coming years will be decisive for telcos as they explore how to build on their strengths to increase their relevance in this evolving space.

Who knows! Trust may prove to be both the Achilles’ heel and the defining force of agentic AI. But I do know that it isn’t a luxury add-on or a premium feature to dress up agentic AI offerings. It’s a building block of any functioning agentic AI infrastructure. And the players who secure a foothold in that layer will gain a strategic position in the ecosystem. So, when people say, “trust is dead in the digital age,” all we can do is keep proving them wrong.

Read more

• Discover the four ways agentic AI will reshape telecom
• Learn how AI agents play an important role in telecom network architecture
• Learn how telecom networks are becoming the nervous system of physical AI