Anthropic CEO Dario Amodei wants the entire AI industry to slow down, citing recursive self-improvement and a rogue agent swarm that tried to hack its own grader. He’s proposing embedded evaluators, industry coordination, and even global treaties on the pace of AI. Maybe he’s right. Maybe it’s positioning. Either way, telcos have no stake in this fight. The pacing debate is a model-layer fight, and models have never been your moat. You’ll buy whatever frontier model exists, at whatever pace it arrives, same as your competitors. What’s yours is the context layer on top of the model, and it compounds at any speed. Own the layer that lasts and let the labs argue about pace.
I think Bain misunderstands the AI opportunity. The consulting firm warns that AI could bloat telco OpEx, projecting an “agentic operating model” where token costs pile 20–30% on top of a legacy cost base that stays 70–80% intact. 🤦♀️ Just… no. Bain is literally describing bolting agents onto old workflows and keeping everyone, then acting surprised the bill goes up. That’s an AI-abled trap. It also sets the ambition embarrassingly low. Think bigger. Redesign every role for autonomous operations, where agents do the work and people handle exceptions, and the target is 50–70% OpEx reduction, with token spend a rounding error against what you eliminate. That’s how you AI.
Dan Schulman has spent year one at Verizon doing what Bain says telcos won’t: attacking the operating model instead of decorating it. The PayPal transplant cut almost 13,000 people in month two, called out a culture that was “process-oriented as opposed to results-oriented,” and will exit 2026 with over $5B in OpEx savings on a run rate basis. He’s also investing in the layer telcos should own in the AI buildout, signing a $1B dark-fiber deal with Google and a $2B+ supply agreement with Corning. Now comes the hard work: the actual redesign of every role to optimize operations with AI. Cuts prove the will. The redesign proves the model.
Accenture and Google Cloud just formed a joint unit with 1,000 “forward deployed engineers” (FDEs) to push agentic AI into enterprises, drawing on Accenture’s 50,000 Google-skilled staff. It follows Microsoft’s $2.5B Frontier Company, which embeds 6,000 experts at customers. OpenAI and Anthropic’s backers are funding the same play. See what’s happening? The industry has renamed the professional services army. Palantir invented the FDE to make the product absorb the work, so each deployment needs fewer people, not more. Accenture bills by the head, so its 1,000-person FDE force is the same old consulting model wearing Palantir’s jacket. Remember: one Totogi FDE built six BSS modules live in nine hours in Düsseldorf. Ask your AI partner how many people its “transformation” requires. The answer tells you what they’re really selling.
Deutsche Telekom just named Jan Hofmeyr as its next board member for product and technology. His last job? Five years at Amazon Web Services (AWS), most recently running the hyperscaler’s telco business. CEO Tim Höttges said the quiet part out loud: Hofmeyr knows “the speed of cloud and digital business models,” and “this is exactly the connection we need now.” I’ve spent years telling this industry that the public cloud is the key to our future. Guess I was right: Europe’s biggest telco just made it a board-level hiring criterion. The public cloud vendors were never your enemy. They were partners that could uplevel your business, and DT figured it out. Have you?
In one week, T-Mobile’s CFO declared Starlink “not a competitive threat... at all,” AT&T’s CEO trashed its small-cell plan, and Verizon’s CEO reported zero discernible impact. Behind the tough talk, the carriers’ actual defense is a promise that nobody will ever grant Musk network access. But he’s sitting on $19.6B of EchoStar spectrum, he’s the richest man on Earth, and it only takes one yes to the best offer. Even the dismissals concede the ground game: Osvaldik admits Starlink will win “from the deep rural into suburban fringe,” which is a CFO describing classic low-end disruption in the middle of denying one. And SpaceX’s endgame isn’t even your subscribers. Shotwell is playing for outdoor physical AI—cars, robots, drones—a market she bets will dwarf smartphones. Is she right? Will Elon win in the end?
Telcos are taking AI into production with the Totogi Ontology, and the receipts are piling up: seven live deployments across five continents in the past year alone. Dormant cell resolution is down from days to 30 minutes for one operator. Alarm floods cut 97% for another. A tier-one Southeast Asian operator now runs its entire BSS AI estate on the ontology. The throughline: deploy in weeks, prove it in production, then expand—not another pilot that dies in a slide deck. What are you waiting for? Come see the Totogi team at TM Forum Innovate Americas in Dallas, October 6–7, and we’ll show you what we can do for you.
Pipeline Magazine just published my piece on why AI keeps failing in telco despite operators having more data than almost any industry: it’s a context problem, and until you name it correctly, every AI investment underperforms. I walk through the Zain Sudan story, of cell sites showing green on every dashboard while carrying zero traffic, because the problem was hidden between the systems, not in them. The fun part: I’m in an issue right alongside Ericsson, Oracle, and CSG explaining their approaches to the same problem. Read them all and decide whose architecture actually closes the gap. I know my answer. 😉
