I spent the first day at DTW Ignite 2026 in Copenhagen, and the message was not exactly subtle.
Everything is AI.
Everything is agentic.
Everything is possible.
At least if you listen to the big-stage energy for long enough.
TM Forum has leaned all the way into it this year. The event itself is being presented as an AI-native experience, with agentic AI used as co-hosts, analysts, intelligence layers and real-time guides across the agenda. The main tracks are exactly what you would expect in 2026: Composable IT and Ecosystems, Autonomous Networks, and Trustworthy AI and Data.
That is not a criticism. It is probably the right agenda.
But after a few hours, you also start to hear the difference between people selling the future and people who have to operate it on Monday morning.
You can hear the geography in the optimism
One funny thing from day one: close your eyes for a few minutes and you can often tell whether the speaker is coming from a European operator environment or from somewhere else.
Not by accent. By constraint.
The non-European and vendor-side talks often have a very clean shape:
AI will compress time. Agents will remove friction. Autonomous networks will unlock new business models. Digital platforms will turn complexity into opportunity. The telco will become faster, smarter, more adaptive, and more valuable.
I do not hate that message.
Some of it is true.
But the European operator tone tends to carry more weight in the shoes. Regulation. Legacy. Procurement. Customer expectations. Margin pressure. Data quality. Integration debt. Security. Sovereignty. Uncomfortable economics.
That difference is quite amusing when you notice it. It is also useful.
Because telecom does need ambition. But it also punishes fantasy.
The stage version of AI is clean
The best version of the AI story sounds obvious.
Telecom has complex networks, huge operational datasets, expensive support flows, old IT estates, difficult product catalogues, and a constant need to reduce manual work. That is exactly the kind of environment where AI should help.
Agents can monitor. Agents can summarize. Agents can recommend. Agents can generate configuration. Agents can investigate incidents. Agents can help customer service. Agents can support engineers. Agents can stitch together workflows across OSS, BSS, CRM, inventory, assurance, billing and field service.
Autonomous networks are not a silly idea either. There is real value in moving from manual operations to closed-loop automation, and then gradually toward higher levels of autonomy where the system can detect, reason, act and validate with less human intervention. The day-one agenda had exactly that flavor, including a session on the autonomous service provider and what has to be engineered in from day one.
That is the part I agree with.
The industry cannot hire enough people to manually operate all the complexity it has built. It cannot keep stacking process on top of process. It cannot keep asking humans to be the integration layer between systems that should already know how to talk to each other.
So yes, AI and agents belong in telecom.
The question is not whether they belong.
The question is whether they can survive the economics.
Telecom is a brutally awkward place to invest
Telco is a strange business.
Customers want connectivity all the time. They notice it immediately when it fails. They complain when the speed is bad, when coverage drops, when roaming is weird, when installation takes too long, when support is slow, when the router needs a reboot, when the mobile signal disappears in the wrong building.
Fair enough.
But most customers do not want to pay much more for it.
Connectivity has become critical infrastructure with commodity pricing expectations. That is a hard combination.
So when someone says AI will transform the telco, the useful follow-up is not “can we build a demo?”
Of course we can build a demo.
The useful follow-up is:
- Does it reduce cost in a measurable way?
- Does it improve reliability without creating new operational risk?
- Does it shorten delivery time for real products, not slideware products?
- Does it create revenue customers will actually pay for?
- Does it integrate into the existing estate without becoming another platform to babysit?
- Can it be governed, audited, secured and explained?
That list is less exciting than “agentic AI will unlock the future.”
It is also where the money is.
The hard part is not the model
One of my main reactions from day one is that the AI conversation is still too model-centered in places.
The model matters, obviously. But in telecom, the model is rarely the hardest part.
The hard part is the operating system around it.
Data ownership. Data quality. Event models. Inventory accuracy. Service models. API contracts. Process boundaries. Legacy system behavior. Monitoring. Human approval points. Rollback. Incident response. Vendor accountability. Regulatory exposure. Security review. Commercial ownership.
An agent that can reason beautifully is not very useful if it is reasoning over stale inventory, undocumented workflows, contradictory customer states, and APIs that only work because three people know which hidden field not to touch.
That is why the boring architecture work still matters.
Actually, it matters more now.
AI does not remove the need for clean integration. It increases the cost of not having it.
If we want agents to act inside telecom environments, we need systems with clear contracts, meaningful events, safe permissions, observable actions, and sane recovery paths. Otherwise we are not building autonomous operations. We are just giving a confident tool access to a messy estate.
That may be fast.
It is not mature.
Agentic AI needs a business case, not just a booth
The show floor energy is useful. Demos matter. Catalysts matter. Seeing what vendors and operators are trying matters.
DTW is good for that. The official day-one highlights point at opening keynotes, 60+ Catalyst projects and 120+ exhibition booths. That density is valuable because it shows where the industry vocabulary is moving.
But the next phase has to be less about proving that an agent can do something and more about proving that the agent should be allowed to do it repeatedly in production.
There is a big difference.
A demo agent can impress people by producing an answer.
A production agent has to earn trust over time.
It has to know when not to act. It has to leave an audit trail. It has to handle partial failure. It has to keep permissions tight. It has to escalate cleanly. It has to work with the existing operating model instead of pretending the operating model will magically disappear.
Most importantly, it has to pay for itself.
That payment can come as lower cost, fewer incidents, faster delivery, better customer experience, higher automation, or new revenue. But it has to show up somewhere real.
Otherwise the industry will repeat an old pattern: big transformation language, expensive platforms, complicated programs, and then a quiet return to manual workarounds when the promised economics do not arrive.
My first-day takeaway
I left day one more convinced that AI will matter in telecom.
I also left more convinced that the winning telcos will not be the ones with the loudest AI story.
They will be the ones that can connect the AI story to operations, architecture and economics.
The optimistic speakers are not wrong. There really is a lot of possibility here. Agentic workflows, autonomous networks, AI-assisted engineering, model-based operations, trustworthy data, composable IT. All of it points in the right direction.
But telco does not become easy because the vocabulary got better.
The networks still have to run.
The customers still expect everything to work.
The regulators still care.
The legacy systems still exist.
The investment still needs a return.
So yes, everything is AI now.
Fine.
The more interesting question is whether we can make it useful after the conference lights turn off.