The Next Telco Transformation Is Not only the Network. It’s the Stuff All Around It
This week I was at Dreamforce, and the session that resonated most with me had nothing to do with CRM. It was Salesforce’s Chief Architect, Dave Ward, arguing that the biggest problem in telecom right now is not shifting packets. It’s the hundreds of corporate systems that wrap around the network, and what happens when you try to put AI agents on top of them.
Ward has spent decades in networking and telecom, including senior positions at Cisco and Lumen, so this is a network person telling network people that the center of gravity has moved. That framing alone is worth paying attention to. Here is my summary of his thesis, and why I think it matters more to operators than most of the AI announcements coming out of the show.
The trouble with a simple question
Ward opened with a question every telco sales leader has asked at one time or another:
Which of my existing customers can we upsell with a new product that the network can genuinely deliver, and who has paid their bills on time?
That should be an easy question. To answer it you need to pull data from CRM, billing, ERP, the product catalog, network inventory, assurance, telemetry and probably two or three other systems nobody remembers buying. It takes several teams, a pile of spreadsheets and weeks of effort.
Ward’s point is that AI agents don’t solve this. They inherit it. Drop an agent into that environment and it hits the same maze a human analyst does, only faster.
Why point-to-point integration fails agents
Over decades of acquisitions, organizational silos and technology generations, telcos and media companies have accumulated hundreds of applications and systems of record, connected by tightly coupled, point-to-point integrations.
Ward’s math is simple. Fully integrating just eight systems point to point takes 28 distinct integrations. Large operators have hundreds of systems. That number grows with every new product, channel, workflow or agent.
The chain he highlighted was old architecture leading to bespoke integrations, bespoke integrations driving high OpEx, high OpEx slowing transformation, and slow transformation leaving no funding for anything new. That loop is familiar to anyone who has sat through a budget review for a BSS/OSS modernization.
His alternative is a platform layer sitting between applications and systems of record, so that the applications and the data can evolve independently. The idea is not new, but AI makes it urgent.
AI interfaces are the new web browser
This was the line of the session for me. Ward said Salesforce is moving increasingly toward operating “headless.” Salespeople no longer have to live inside a Salesforce screen. They might work in Claude, in Copilot, or in whatever surface wins inside their enterprise next year.
So the strategic goal is not to own the user interface. It is to make enterprise data, tools and actions securely available to whichever approved interface a human or an agent chooses to use. That is a significant statement from a company that built its franchise on the front end.
The four things the architecture has to get right
Ward’s platform layer comes down to four capabilities.
Identity. Agents need identities, just as people do. The enterprise has to know who or what is acting, what permissions it holds, and which person, role or business function it represents. Ward expects that agents attached to a recognized business function and working alongside people will win adoption well before fully autonomous agents do.
Governance. You cannot send an agent into the estate and let it do whatever it thinks is right. The architecture has to answer, in real time: can this actor make this request, access this data, take this action, offer this product, approve this discount? Governance is built into the architecture, not bolted on after deployment.
Session and context. Every agent task needs a persistent session ID. If an agent has to make seven calls across CRM, ERP, billing and network systems to finish one job, those seven calls should be tracked as a single flow. Ward’s phrase was that agent observability should start to look like network flow observability.
Observability, evidence and auditability. Every activity produces a trail the enterprise can reconstruct: identity, request, data accessed, APIs called, decisions made, actions taken, outcome.
The most pointed phrase of the talk, and the idea that tied it all together, was this: “You can’t trust agents until we prove they can be trusted.” That shifts the conversation from deploying AI to building an architecture that can prove AI behaved correctly, with control points, audit points, policy enforcement and, where needed, kill switches.
Three use cases for telecom
Quote-to-cash. A rep asks an agent to build a configuration, check service availability, create a bill of materials, apply commercial rules and generate a quote. Today that touches dozens of systems. In Ward’s model the path is: customer request, agent, governance and action layer, systems of record, quote. No new integration per system.
Upsell and renewals. Ward treats renewal as an opportunity to grow revenue, not merely a task to retain it. An agent finds the accounts coming up for renewal, works out which ones should move off legacy products, which new services the network can actually deliver to them, which have healthy payment histories and which offers fit, then hands the account team a ready-made sales play. Notably, Ward argued that operators should start with use cases that make money, not just save it.
Field operations. A technician gets a trouble ticket on a phone, but the knowledge needed to fix it lives in the customer record, network telemetry, historical performance, transport quality, video quality, topology and service configuration. BSS and OSS have always treated these as two separate worlds. Ward’s model defines the need for a “golden customer record,” where operational data becomes real-time context attached to the customer. That means a technician’s ticket should carry shared customer and network context, rather than requiring separate lookups across silos. The trouble ticket becomes just one more object on that record, and the technician gets a single view instead of six logins.
The bigger shift: front office and back office are converging
This is the most important idea in the session. Back-office operational data must be available to front-office teams and agents. CRM can no longer be isolated from operational systems, and BSS can no longer be isolated from OSS, because agents need to work across all of them.
Ward was careful about what he was not saying. He is not proposing that everything be dumped into one giant database. The data stays where it is. The platform is federated, and access to it is governed. That distinction is essential, because “consolidate everything” has been the graveyard of telco transformation programs for two decades.
His advice: pick one workflow with an obvious economic benefit, such as quote-to-cash or renewals. Build it. Test it. Watch it. Measure it. Then reuse the same identity, governance, federation, action and evidence components on the next process. Reuse is what changes the economics. Instead of paying for bespoke integration every time, each new agentic workflow rides on the platform already in place.
The whole argument comes down to a single transition:
Old model: Human → App → Point integration → System of record
Agentic model: Human or agent → Any interface → Governed action platform → Federated enterprise data and systems
My take: why this matters to the market
I write about carrier routing and switching for a living, so I am used to hearing vendors describe the network as the center of the world. Ward’s point is that for the next few years the value in telco will not be captured at the network layer. I think he is right: whoever solves the systems layer owns it. Here is why.
The money is in the sales process, and operators know it. Most agentic AI in telecom so far has landed in customer care, because deflecting calls is easy to measure. Ward is instead talking about quote-to-cash and renewals. In my conversations with operators, the biggest untapped revenue pool they have is the gap between what the network can deliver and what sales actually offers. If an agent can close that gap using network inventory and assurance data the sales organization has never had access to, the ROI dwarfs anything in the contact center.
This demands a discipline telcos already have. That is the irony Ward arrives at, and the part I find most interesting. Flow observability, policy enforcement, identity, session tracking, audit trails, kill switches: operators have been doing all of this for network traffic for decades. The agentic enterprise simply needs those concepts applied to agent flows instead of packet flows. If telcos recognize that the skills transfer, they may be better placed to run regulated agents at scale than most enterprises.
Federation is the claim to watch. Every operator I speak to has a data consolidation program that is years late and over budget. Ward’s insistence that data stays where it lives, with the platform governing access, is the only version of this that stands a chance of deploying inside a real telco. If Salesforce and its competitors can show that federation works against real OSS inventory and telemetry, and not just CRM and billing, it changes the buying conversation.
The risk is that the integration bill moves rather than disappears. The honest response is that a governed action platform is still a platform somebody has to integrate with hundreds of systems. The economics only work if the cross-workflow reuse is real. Operators should ask to see the second and third workflows built on the same components before they believe the cost curve. And the vendor offering the platform obviously has an interest in becoming the layer everything passes through, so lock-in is a fair question to raise.
The impact on the BSS/OSS market could be significant. If the front-office and back-office wall comes down the way Ward suggests, the conventional boundaries between CRM providers, BSS suppliers and OSS vendors blur quickly. That is a threat to incumbents on all three sides and an opportunity for whoever proves out the governance layer first. I expect this to be one of the defining contests of the next planning cycle.
At bottom, Ward’s talk was a network architect warning the industry that the next great shift is not algorithmic but architectural. The models will keep improving. The question for operators is whether their systems let an agent do anything valuable with them. Most do not yet. Start with one revenue workflow, build the trust architecture around it, and you will have an advantage that lasts.
Ray Mota ACG Research
rmota@acgcc.com
Originally published on Ray Mota's Substack. Subscribe to get new analysis by email.
