Zendoric
← Back to the day · October 31, 2026

Deloitte: 69% of Chilean companies use AI agents, but fewer than 1% govern them with auditing and traceability

🕒 Published on Zendoric: October 31, 2026 · 09:09

✨ AI-generated · how it's made

Deloitte's ConvergencIA study in Chile shows AI agents are already in 69% of organizations, but almost none have a formal control model. In our view, the problem is not adopting fast, but doing so without being able to audit what the agents do.

🎧 Listen to the analysis · synthetic voice

By Portal Innova · October 30, 2026.

According to the third edition of Deloitte's ConvergencIA study, more than 90% of companies in Chile use artificial intelligence to some degree. 69% report some use of AI agents, that is, systems capable of executing tasks or taking actions with varying levels of autonomy.

Of that 69%, 31% already have them in production or integrated into their operations, and 38% are in experimentation or pilots.

The gap appears in governance. According to the same study, 26% use agents without formal guidelines and 28% have only initial guidelines. Less than 1% have a formal model with monitoring, auditing and traceability.

The pattern repeats with AI in general. One in four companies applies no specific measures to manage the risks of its deployments, and only 13% conduct internal audits of these tools. Among the obstacles cited, 28% point to explainability and trust in results, and 23% to data privacy and security.

The study adds a nuance about where companies are focusing their efforts. According to Deloitte, they concentrate on measures that would not be structural, such as training to mitigate risks (36%) and human validation (28%). Jaime Caiceo, AI and Data lead partner at Deloitte Chile, also notes that the preference for low-code agents (tools requiring little programming that are quick and cheap to set up) reflects an intention to adopt the technology quickly and at low cost. Agents are used mainly for basic automation and partial process redesign.

Our take: the key finding is not that adoption is high, but that what is being deployed is the capacity to act without the layer that shows what each agent did, with what permissions and how to reverse it. An agent writing to real systems without traceability is a black box with credentials. Training people and having a human validate are reasonable measures, but they do not scale when agents act by the thousands.

As industry context, this is the same tension we see in the global agentic economy: value and the bottleneck are shifting from the model to the identity, permissions and auditing layer. Whoever builds it first, whether the company itself or a vendor, will gain a competitive advantage and not just compliance.

The study should be read with caution. These are self-reported figures from companies, in a single country and with whatever methodology Deloitte publishes, which the article does not detail. Even so, the figure of less than 1% with full governance is striking enough to be taken as a directional signal.

In the short term, this means avoidable incidents, fraud and automated decisions no one can explain, all of which will fuel distrust and regulatory pressure. In the long term, the outlook is brighter: automation that frees people from routine tasks will only be sustainable if it is auditable. Companies that treat governance as infrastructure rather than red tape will be the ones that get there with their customers' trust intact.

🔗 Related on Zendoric

Sources & references