From the newsletter (sign-up if you want it sooner): Telecom operators are embracing AI through tightly governed, domain-specific deployments, focusing on bounded autonomy rather than full automation. The emerging model wraps probabilistic AI in deterministic controls, enabling safe delegation while preserving the reliability and accountability networks demand.
More from Sean, who follows up his DTW Ignite narrative from yesterday with another today. The same idea is at the heart of both pieces, as well as his recent Cisco and Qualcomm reports: how to safely delegate to AI inside systems that were not designed for such probabilistic behaviour. As yesterday: networks are engineered for repeatability, auditability, and control, while AI produces variable outputs shaped by context and likelihood.
That mismatch defines where autonomy can even exist. AWS’s distinction is decisive: in analytics, AI error is tolerable; in configuration, “you can’t be wrong”. Which is kind of funny, because this is what the six-nines enterprise sector has been telling the five-nines telco sector for years. Either way, the telco industry’s task is not to make networks probabilistic, but to wrap AI in constraints that make its outputs safe to act on.
Both articles converge on this idea of bounded autonomy – “deterministic envelopes around probabilistic systems” in the first; governed execution layers in the second. The architecture is the same. TM Forum frames this through its Open Digital Architecture extensions and Level 4 autonomy models, where AI is embedded into operations but constrained by policy, observability, control.
The point is not augmentation for its own sake, but enforceable delegation. There is good discussion here from telcos about how this materialises – from Boost Mobile, Rakuten Mobile, Deutsche Telekom. This all feels pragmatic, and therefore quite mature: autonomy is not a system-wide state but a sequence of constrained deployments. Each use case defines what data an agent can see, what action it can take, when it must escalate, and what outcome validates success.
Even so-called ‘Level 4’ autonomy is partial, applied to specific domains like energy optimisation rather than end-to-end networks. The agentic network will arrive as a governance pattern in telecoms, repeated across domains until autonomy is earned, and not just assumed. It starts to feel like telcos might be the ones to make the best fist of AI transformation; maybe that will be the foundation for them to help other industries with the same.
