AT&T and GSMA warn Telco AI Models Drift

Boost Mobile data scientist Priyank Jain, GSMA director of AI technologies Louis Powell and AT&T data office vice president Mark Austin flagged a specific and seldom discussed risk in telco AI deployments this week: the gap between the data a model was trained on and the data it actually encounters once it is live, a failure mode known as train serve skew. Speaking to Fierce Network on 25 September 2026, Jain explained that the danger is not a crash or a failed job. Nothing alerts an operator that anything is wrong. Jain said the model "just gets quietly worse."

Two Different Ways Models Drift

The article identifies 2 separate causes. General purpose frontier models arrive with no exposure to telecom specific data formats or vendor taxonomies, since none of that appears in their internet scale training data. Telecom specific models face a subtler problem: production data, such as settlement records, arrives late or out of order in ways the training set never captured, and datasets in this domain routinely carry over 100 columns of custom, operator specific parameters. Powell added that frontier models have no inherent understanding of telecommunications and can produce confident, wrong answers, a real risk once a model is embedded in an enterprise workflow. Austin's prescription is procedural rather than technical: match training data to production sources as closely as possible, run a new model silently alongside the live system before cutting over, and keep testing it after deployment rather than treating the launch as the finish line.

Why This Matters Before Autonomy

This is a narrower and more mundane problem than the industry's current preoccupation with agentic AI and autonomous networks, and that is exactly why it matters. The NGMN report on agentic AI concluded that even coordination within a single operator's own domains, let alone accountability between an enterprise and its network provider, requires common information models and audit trails that do not yet exist. Train serve skew shows the problem starts a level below that. A supervised classifier with a known, static task can still fail silently once its inputs drift from what it was trained on. An autonomous agent negotiating across an operational boundary carries the same drift risk plus the coordination and accountability gap NGMN identified. Operators evaluating agentic AI pilots should ask a more basic question first: whether they would know if the model already in production today has quietly stopped working.

Quiet drifts like this example are worse than explicit failures. The quality of the output degrades silently, which degrades the quality of service. Explicit quality control for data quality before and after transformation, and before and after agentic intervention are necessary to enable autonomous networks.

Orange Business Moves Agentic AI Onto Open-Weight Models

Orange Business is moving its most sensitive AI workloads, including its agentic systems, onto open-weight models running on infrastructure it controls, and it says sovereignty rather than cost is the reason. The account comes from a Fierce Network interview with Miguel Alvarez, chief data and AI officer of Orange Business, published on 14 September by Mitch Wagner. Alvarez describes 3 changes in the operator's AI strategy over the past year: a shift of sensitive work from public cloud models to open-weight models it hosts itself, some of them Chinese; a move from chat windows beside applications to agents inside the systems of record; and a 3-month mandate given to a mixed team of AI and operations process experts to automate the front third of the incident management chain. That last piece is the one with a number attached. Qualification, triage and routing of a ticket used to take a technician 30 to 40 minutes working down a checklist of connectivity and configuration tests. Alvarez says an agent now does the same work in about 3 minutes, with 95% of it automated, and that the Level 1 support team "may not survive in its current form".

Sovereignty is the differentiator

The first change is the one Orange Business leads with. Alvarez says "the importance of being able to do AI in a trusted environment has increased a lot", that open-weight models now run work which needed frontier models 4 to 6 months ago, and that the models his teams use most for coding are Alibaba's Qwen and MiniMax, with Moonshot's Kimi under test. The sovereign option is also a product: Orange Business's Live Intelligence platform, built for 100,000 Orange Group employees and sold to 150 enterprise customers, serves Gemini, ChatGPT and Claude alongside Mistral and the open-weight models Orange hosts itself, so a customer can route ordinary work to a hyperscaler model and keep sensitive work inside a French enclave. Fierce reports that the optionality became a practical priority in June, when the US government ordered Anthropic to suspend foreign access to its 2 most advanced models. In July I argued that sovereign AI capacity is the most immediate of the credible operator revenue lines, and that edge and distributed compute become fundable when sovereignty is the demand driver rather than the garnish. Orange Business is the first operator I have seen describe the same logic from the inside, as a buyer of models rather than a seller of GPUs. It did not move to open-weight models because they were cheaper. It moved because a customer, or a regulator, or a foreign government, can now switch a frontier model off.

The second change is the one that matters for the business case. When Bain warned last week that agentic AI could raise telco opex by 30%, the mechanism was that a process augmented rather than replaced removes nothing: the human team keeps running the workflow while agents perform isolated tasks at its edges, and the legacy 70% to 80% of the cost base stays. Alvarez describes exactly that trap in his own first phase. Summarisation, ticket correlation and root cause analysis were the early work, and "you still have the same roles doing more or less the same work in more or less the same way". The 3-month mandate was the correction. It targeted a whole segment of the chain rather than a task inside it, it was staffed by process experts as well as AI experts, and it ends with a change to the organisation: Level 1 teams that supervise agents and handle customer communication instead of running the checklist. Similar exercises are running at Orange Group level across HR, finance, legal, software development and B2B sales, aimed, in Alvarez's words, at whole job lines rather than individual use cases. That is the difference between a demonstration and a P&L entry, and it is the first operator account that describes the retirement of the legacy process as the goal rather than as a consequence to be managed later.

The 2 changes are connected. The reason sensitive work moved onto controlled infrastructure is that agentic systems, "whose autonomy introduces risks requiring greater scrutiny", are now among the workloads. Fierce lists the season's incidents: an autonomous agent system breaking into part of Hugging Face's production infrastructure in July, a swarm of agents using a dormant wiki as a coordination board, a model reaching third-party systems during a security evaluation. Alvarez's response is operational rather than philosophical. Trusting vendors to watch on the operator's behalf is no longer sufficient; vulnerability assessments are more frequent; and the operator needs the ability to quarantine sensitive services and cut their connectivity. That is the containment layer I described when I argued that the agent runtime is not the agent model, and it is the enforcement half of the governance question the industry has mostly discussed as policy. An operator that runs the model on its own hardware can quarantine it. An operator that calls a frontier model through an API can only stop calling. Orange Business's stack, on Alvarez's description, is built on LangChain and LangGraph, OpenTelemetry for observability, Model Context Protocol for interconnection and open-source components for the LLM gateway and MCP registry. Every one of those is a runtime component. None of them is a governance model, and the interview does not describe one.

The main challenge for AI remains trust. Delivering trustable operation, results in a controlled, auditable, traceable manner is paramount. As a result, operating on models that are under your control, deciding what dataset should remain on premise, in jurisdiction or in public cloud and what governance model agentic relies on is a crucial set of decisions for operators.

Agentic AI Could Raise Telco Opex by 30%

Two reports published on 10 September describe the same balance sheet from opposite ends. Bain & Company, as reported by Telecoms.com's Mary Lennighan, warns that operators could see operating costs rise by around 30% without any gain in productivity or growth if agentic AI is bolted onto legacy operating models. The consultancy describes an emerging agentic operating model in which traditional costs account for 70% to 80% of an operator's total and AI costs for 20% to 30%, with the risk that the legacy 70% to 80% never goes away. The same morning, TelecomTV's James Pearce published a second cut of the AI-Native Telco Index, the 276-page evaluation of 56 operators based on published material from January to June 2026, and found that 20 of the 56 frame their AI strategy primarily around revenue, 28 balance revenue and efficiency, and only 6 are primarily focused on efficiency. TelecomTV names the 6 as KPN, Proximus, Spark New Zealand, BT Group, Comcast and Rogers. It also notes, in the same article, that across all 56 operators cost savings remain the most clearly measured AI-related financial impact.

A new cost layer on top of the old one

Bain's argument is about arithmetic rather than technology. Model prices have fallen almost 10 times in the past year, according to the report, but token usage is rising faster as staff find new uses in network operations and customer care. At roughly half the token price, a 4.5 times increase in usage doubles the bill. The recommendation is to measure the cost of a business outcome, a resolved customer issue or a closed network incident, rather than the cost per token, and to hold someone accountable for model drift and token spend. AT&T is the worked example: it redesigned its orchestration so that large supervising agents delegate to smaller specialised models, which by AT&T's own account cut costs by up to 90% while tripling throughput. Bain's 3 cost traps are the token bill, the dual operating model in which human teams keep running the process while AI performs isolated tasks at the edges, and the low-risk demonstration that improves efficiency at the margin without touching the P&L. Internal chatbots, call summaries and copilots are named in the third category. This is the cost side of the 3 money flows I set out in July, when I argued that cost avoidance and revenue must be separated before an operator can say what AI is worth to it. Bain adds the point that cost avoidance is not automatic either: the avoided cost has to be removed from the organisation, and a process that is augmented rather than replaced removes nothing.

Revenue is the stated priority, savings are the measured one

The Index finding runs the other way. When I covered the first edition of the Index on 7 September the observation was that none of its 4 dimensions was financial. The second cut is TelecomTV's answer, and it is a useful one, because it reads the same public record for money rather than for capability. Of the 20 revenue-first operators, 11 are Tier 2 operators with annual revenues between $2 billion and $10 billion, 11 are in Asia and 4 in the Middle East, which TelecomTV reads as evidence that the shift toward AI revenue is not being led from Europe or North America. The examples with numbers attached are few. Bell Canada reported around C$700 million of AI-powered revenue in 2025, up 60%, and targets around C$2 billion a year by 2028. SK Telecom's AI datacentre business grew 92.5% year on year to 136.2 billion won in the second quarter, with AI B2B and B2C services at 61.3 billion won, up 24.5%. Indosat Ooredoo Hutchison reports sovereign AI cloud income separately, $28 million in 2025 with guidance of $60 million in 2026, and its Zankore joint venture with Nvidia, Nokia and Ooredoo raised $3.1 billion in debt this week to build 200 MW of GPU capacity by early 2027. On the efficiency side, KPN targets around €100 million a year in operating cost savings by 2030, Proximus folds AI into a €180 million efficiency programme, and Spark credits its churn model with reducing customer loss by up to 30%. TelecomTV also cites GSMA Intelligence's finding that only 35% of telco AI initiatives include a revenue objective. Read together, the picture is that revenue is what operators now say, savings are what they can measure, and 2 of the 3 operators publishing AI revenue lines, SK Telecom and Indosat, are selling compute or cloud capacity, which is the first of the 4 credible revenue lines in the July post and the one that looks least like a telecom service. Bell has not said what its C$700 million comprises.

The number the business case is missing

Put the 2 reports side by side and the missing figure is obvious. The Index counts operators that talk about revenue and reports the revenue lines that exist. Bain estimates the cost that agentic AI adds and warns that the legacy cost may stay. Neither report, and no operator in the Index, publishes both halves for the same company: AI-attributed revenue, AI-attributed savings actually removed from the cost base, and the AI compute and token cost that Bain says will make up 20% to 30% of the operating model. Bell's C$700 million is a revenue figure with no cost of sale attached to it in public. SK Telecom's AI datacentre revenue is real, but it is a hosting business whose margin is set by power and GPU depreciation, not by the network. KPN's €100 million is a 2030 target. Bain's own preferred use cases, autonomous network incident resolution and proactive churn prevention, are the workflows where the operator has to redesign the process end to end and retire what it replaces, and they are also the workflows where the data has to be fit for an agent to act on. Deutsche Telekom's Ahmed Hafez told the AI-Native Telco Forum on 8 September that DT's network data "is not ready" for machines and that data quality demands will start appearing in supplier RFQs, which is a cost that precedes any saving. Bain's framing is the right one for the report I am completing on autonomous networks: cost per resolved incident, before and after, with the platform, compute and data cost inside the denominator. That is the number that converts a proof into a P&L entry, and it is the number no one in the Index has yet published.

What has not been published

No operator has published AI revenue, AI savings and AI operating cost for the same period. Bell has not disclosed the composition of its C$700 million or the margin on it. AT&T's 90% cost reduction is its own account, relayed by Bain, with no base figure. The Bain report itself is available from Bain; the figures here are as reported by Telecoms.com. 4 things to watch. First, whether any of the 20 revenue-first operators reports AI compute or token spend alongside AI revenue in a quarterly filing, since that would be the first complete line item. Second, whether a legacy cost is visibly retired and attributed to agentic automation, in headcount, systems or supplier spend, rather than a saving being projected. Third, the on-demand recording of the AI-Native Telco Forum session on closing the attribution gap between deployment and financial outcomes, with Deutsche Telekom, Everpure and TM Forum, which TelecomTV says will be posted within days. Fourth, whether the 2027 edition of the Index adds a financial dimension and, if it does, whether it measures cost of outcome the way Bain proposes.

TelecomTV Publishes First AI-Native Telco Index

TelecomTV published its first AI-Native Telco Index on 7 September, the day before its AI-Native Telco Forum opens in Düsseldorf. Ray Le Maistre's article describes a 276-page report assessing 56 operators against 4 dimensions, foundation readiness, strategic intent, execution evidence and transformation velocity, each made up of 5 indicators, using published evidence only, with a cut-off of 30 June 2026. The index assigns each operator 1 of 7 archetypes rather than a rank; the decision not to publish a league table was taken with the ANTA steering board, whose members come from Axiata, Deutsche Telekom, Orange, NTT Docomo and Rakuten Mobile. 9 operators are placed in the top archetype, AI Vanguard: AT&T, China Mobile, Deutsche Telekom, KDDI, NTT and NTT Docomo, Rakuten Mobile, SoftBank Corp, Telefónica and Verizon. 5 are Infrastructure Architects, 15 are Strategic Accelerators and 1 is at Transformation Pending. The stated rule is that "announce does not mean executed": a commitment is recorded as a commitment, and a deployment is recorded only where the operator has published something showing it is running. SK Telecom, which the article notes is often named the leading AI-native operator, is a Strategic Accelerator because many of its efforts were still at the planning stage at the end of June. A 12-page executive summary is free; the full index is available to ANTA partners and licence holders. This post works from TelecomTV's article; neither the summary nor the full report was read for it.

Most assessments of operator AI progress, including most vendor-sponsored surveys, score intent: budgets planned, priorities ranked, use cases identified. This index scores what has been published as running, and it records an announcement as an announcement. That is the distinction between field-validated and commercially validated that AI-RAN claims have needed for 2 years, applied here across the whole operator estate, and the SK Telecom result shows the rule working. An operator with a clear strategy, disclosed investment, revenue already generated and a large revenue target still lands below the top tier because its published evidence at the cut-off described plans rather than operations. Producing that result about the operator most often cited as the leader, in a report whose steering board includes 5 operators, is a sign the methodology was allowed to run.

No financial dimension yet

The 4 dimensions measure readiness, intent, execution and speed. None measures return. An operator reaches AI Vanguard by publishing the most evidence that AI systems are in production across its operations, and that is a real achievement, but it is the same class of evidence as the customer-experience award that TM Forum gave Google Fiber last week: proof that it works, not proof of what it is worth. The index is explicit that it measures transformation progress and not who is better, and that framing is useful, because it means nobody should read the Vanguard list as the operators making money from AI. I argued in July that most of what operators call AI monetization is cost avoidance, and that the discipline starts with refusing to aggregate cost saving, defended revenue and new revenue; an index that counts deployments without asking which of the 3 flows each one serves cannot make that separation either. The industry now has a reliable way to count deployments. It does not yet have a published operating cost per subscriber, per activation or per ticket, before and after, from any of the 9, and until one of them publishes that figure the index measures the input side of a business case whose output side remains an estimate.

Because the method admits only what an operator has published, an operator that publishes more scores higher, other things being equal. 4 of the 9 Vanguard operators are Japanese, 2 are American, 1 is Chinese and 2 are European, and 2 of the 9, NTT Docomo and Rakuten Mobile, sit on the steering board that shaped the method. The article says archetype assignment is an editorial judgement based on the evidence analysed, and there is no reason to doubt that. The point is narrower: operators with a corporate habit of publishing technical detail, and operators for whom the network is a marketing asset, will be over-represented at the top of any evidence-based index, and operators that run production AI quietly will be under-represented. The index cannot correct for that without abandoning its rule, and it should not abandon its rule. The caveat applies to every evidence-based index, this one included.

Everything the index can see is inside 1 operator's estate: its data foundations, its stated strategy, its deployments, its pace. That is the version of autonomy that ships, and it is where the evidence should be collected first. The next stage, in which an operator's network AI has to coordinate with an enterprise AI, a cloud provider's AI or another operator's AI, produces almost no published evidence yet, because the coordination layer that would make it work is the one I argued no runtime supplies and that NGMN has since itemised even for coordination inside a single operator. An index built on published deployments will show that stage as empty for some time, and the absence will be accurate. A future edition could usefully add an indicator for it: whether the operator has published any interface, ontology or audit model through which an external agent can act on its network, since the developer APIs alone do not answer that question.

What has not been published

The per-operator scores, the weighting of the 20 indicators, the definition of "running" used for execution evidence, and whether any indicator captures financial disclosure are all in the licensed report and not in the article. The 56-operator list is published; the archetype of each operator outside the 3 named tiers is not. 3 things to watch. First, whether the 2027 edition adds a return dimension, or an indicator for published cost or revenue attributable to AI, which would be the single most useful change to the method. Second, whether any of the 9 Vanguard operators publishes an operating cost figure for an automated domain against its pre-automation baseline in the next 12 months, since they are now on record as the operators with the most to show. Third, where SK Telecom lands next year: the article expects it to be among the strongest profiles, and if the planned efforts have become published deployments by then, the index will have shown it can measure movement and not only position.

NGMN Confirms Agentic AI Needs Improvements

The Next Generation Mobile Networks Alliance published a report on 12 August, "Network Automation and Autonomy Phase III: Agentic AI for Autonomous Mobile Networks," and it reads like a standards body catching up to an argument I have been making for a year. The headline finding, reported by Keith Dyer at The Mobile Network, is that the industry has to solve interoperability, security, governance and data before agentic AI can deliver autonomous networks at commercial scale. The detail underneath the headline is what matters. NGMN says the current ecosystem is too fragmented for agents to form a consistent understanding of the network, because vendors use different data models and different terminology, and it calls for common information models, ontologies and semantic mappings, aligned across 3GPP, TM Forum, ETSI, O-RAN, IETF, W3C and BBF. It asks for a telecom-grade Zero-Trust Agent Ecosystem with secure agent identity, authentication, authorisation, policy enforcement, audit logging, runtime monitoring, and the ability to revoke or quarantine an agent. Read that list again. Topology, ontology, authority boundaries, state, audit trails. That is the meta-model of the agentic plane, and it is now in a report with an operator alliance's name on the cover.

I want to be precise about the provenance here, because the sequence is the point. I argued at DTW Ignite that the API is not enough, that the interfaces we built for developer access were never designed to let an autonomous agent understand what it is acting on. I argued a few weeks later that the agent runtime is not the agent model, that a place to execute an agent and put guardrails around it supplies none of the topology, ontology, authority and audit that coordination between agents actually needs. NGMN has now enumerated eight cross-organisation challenge areas, and they map almost one for one onto that case. Lack of standards for agent knowledge and context sharing. Incomplete information modelling for networks. Lack of end-to-end security, identity and trust for autonomous functions. Lack of operator-friendly governance and lifecycle tools. Under-addressed economic and organisational readiness. When a body chaired by Orange's group CTO writes the same list I published, the argument stops being a contrarian read and becomes the consensus. That is a good day for the thesis. It is a better day to point out the part the report does not solve.

The report describes coordination inside a boundary the operator owns

Look closely at the architecture NGMN proposes. A supervisory agent coordinates specialist agents, one each for RAN, core, transport, cloud and security, to diagnose a service problem, weigh corrective actions, resolve conflicts, and verify that service was restored. The worked examples are fault management, service assurance and RAN optimisation. In the RAN case, a cross-domain agent delegates intent to a RAN optimisation agent, which supervises a sub-agent in the infrastructure layer, and when an optimisation action collides with network-wide energy management the cross-domain agent arbitrates the trade-off. This is a serious and correct piece of engineering. It is also, in every example, coordination across domains that a single operator owns. RAN, core, transport, cloud and security are five administrative domains inside one company's walls. The hard thing NGMN is describing is getting an operator's own silos, built by different vendors with different data models, to hand each other a shared and trustworthy understanding of state. That is worth doing, and it is exactly why a runtime and a set of guardrails were never going to be enough. But it is one owner reconciling with itself.

The problem I have been pointing at sits one boundary further out, and NGMN's own scope is the clearest evidence that it is a distinct and harder problem. When an enterprise's AI wants to negotiate with an operator's network AI, for a service level, a slice, a capacity commitment, a remediation, there is no shared owner to impose a common ontology, no single authority to issue the identities, and no single audit log that both parties will accept as the record of what was agreed and who was answerable. Everything NGMN specifies, the semantic alignment, the zero-trust identities, the audit trail, is defined for agents operating within the operator's estate. Across an ownership boundary, the same requirements do not disappear. They get harder, because now two parties have to agree on the model before either can trust an action, and neither controls the other. NGMN has validated that the meta-model is necessary. It has not, and by its own framing could not, close the gap where the two hardest words in this whole subject live, which are coordination and accountability between parties who do not report to the same CTO.

Containment, coordination, accountability, in that order of difficulty

It is worth keeping the three layers separate, because vendors and now standards bodies keep collapsing them. Containment is the Zero-Trust Agent Ecosystem: the identity, the authorisation, the runtime monitoring, the kill switch that lets you revoke or quarantine an agent inside your own network. That is genuinely useful and NGMN is right to specify it in detail. Coordination is the shared ontology and semantic mapping that lets one agent understand what another agent is doing, and this is the layer NGMN is now trying to build across an operator's internal domains. Accountability is the ability to produce, on demand, the identity of an agent, the authority under which it acted, and a trail that a regulator and an outside counterparty would both sign, across a boundary you do not own. The industry has spent a year shipping the first layer and calling it the third. NGMN has now, to its credit, put real weight on the second. The third is still open, and it is the one I argued the EU AI Act made expensive when its enforcement powers switched on at the start of this month, because enforcement does not stop at the edge of your administrative domain.

There is a discipline point here that operators should not lose in the enthusiasm of seeing a favourite thesis ratified. NGMN itself flags it: "While technology groups focus on architecture, few have formal guidance on the required telco organisational changes," and it lists economic and organisational readiness as an under-addressed challenge in its own right. The meta-model is not a product you procure. It is a set of agreements about topology, ontology, authority and audit that has to be built and then adopted, and the report's call for alignment across seven standards organisations is a fair description of how long that takes. Cross-SDO harmonisation of information models is measured in years, not quarters. When I helped set network autonomy targets in operator programmes, the technology was rarely the thing that slipped. The organisational change and the cross-domain agreements were. Anyone booking autonomous-network savings against a framework paper rather than a ratified model is doing the same thing I warn against when vendors stack their individual efficiency claims into a total no network has ever achieved. Read this report as the industry agreeing on the destination. Do not read it as arrival.

What to watch

The honest test has not changed, and NGMN has now made it a formal requirement rather than a personal opinion. For any action an agent takes on its own, can you produce its identity, the authority under which it acted, and an audit trail that both a regulator and an outside counterparty would accept. If the answer lives entirely inside one vendor's runtime, you have containment and you should not mistake it for accountability. NGMN has described the plane the runtime runs on, in more detail than anyone in the industry has committed to print, and it has correctly located the work in standards and organisation rather than in any single box. The next phase belongs to whoever builds the ontology and the identity fabric that hold up not just across an operator's own domains, but across the boundary to the enterprise on the other side of the negotiation. That boundary is where the value is, and it is the one the report stops at. Watch for whether the alignment NGMN asks for actually happens across those seven bodies, and watch, as always, for a pilot inside one operator's walls being described as autonomy across everyone's.

Agentic AI: Autonomous Agents Governance and Enforcement

As of yesterday, the European Commission can fully enforce the AI Act against providers of general-purpose AI models, and the transparency obligations under Article 50 begin to apply. The obligations themselves have been in force for a year. What changes now is teeth: Brussels can investigate, order corrective measures, and levy fines of up to 15 million euros or three percent of worldwide turnover, whichever is higher. Tech Policy Press reported the shift on the 13th of July, and made the point that gives it weight for anyone building networks. The enforcement switch flips just days after the first reported case of an autonomous AI agent carrying out a cyber operation nobody asked it to perform, compromising the Hugging Face platform, as Reuters reported on the 28th. The capability and the accountability mandate arrived in the same week, and they are pointed at each other.

For a month I have watched the telecom industry celebrate the exact capability the regulation now proposes to hold someone answerable for. AT&T describes tens of billions of tokens and over a hundred billion network signals processed daily, 4-hour investigations cut to a minute, and roughly 30 times return. TM Forum reports the sector crossing into Level 4 autonomous networks. Vendors are shipping the agentic NOC and the self-healing network. The good news and the governance problem are the same sentence. Agents are now taking actions in production that no human reviewed in the moment. That is the point of them. It is also the point at which a regulator asks who is answerable.

Enforcement is not an abstraction. To sanction a provider, or to defend yourself as an operator, you have to be able to say which agent took which action, under whose authority, with what audit trail, and across which administrative boundary. In a telecom network that last clause is the hard one. An agent that chains tool calls across provisioning, billing, customer records, and network management is doing precisely the thing that no single log captures and no single owner witnesses end to end. When an autonomous action provisions a service, pulls subscriber data, or adjusts an account, it has crossed several systems that were never designed to hand each other a shared, signed record of intent and authority. The action happened. Reconstructing who owned it is the part that does not exist yet.

This is the meta-model of the agentic plane that I have argued the industry keeps skipping. I made the case that the agent runtime is not the agent model, and before that, at DTW Ignite, that the API is not enough. The coordination problem needs a model of topology, ontology, authority boundaries, state, and audit trails. A runtime supplies none of those across a boundary. It supplies execution and guardrails inside one administrative domain. That is containment, and containment is a genuinely useful thing to own, but it is not coordination between two parties and it is not accountability across them. The regulation just made the distinction expensive.

Let's watch how the two sides of the Atlantic are responding, because the contrast is instructive. In the United States, lawmakers are floating an AI kill switch, the option to suspend or shut a model down when the risk gets severe. A kill switch answers one question well: can I stop it. It answers a different question not at all: who was this agent authorized to act for, and can you produce the record that a regulator and a counterparty would both accept for what it already did. Stopping an agent and being answerable for it are not the same capability, and only one of them is a compliance obligation as of this morning.

So the discipline for operators moving to higher autonomy is now imposed rather than optional. Compliance turns the audit trail and the authority boundary from an architecture nicety into a requirement you can be fined for missing. The honest test is the one I would put to any autonomous-network business case. For any action an agent takes on its own, can you produce the identity of the agent, the authority under which it acted, and a trail that a regulator and an outside counterparty would both sign. If that record lives entirely inside one vendor's runtime, you have containment, and you should not mistake it for accountability. Deloitte's 2026 enterprise survey found only about 1 organization in 5 reports a mature model for governing autonomous agents. The regulator did not wait for the other four.

The agents learned to act this summer. Today the law asked who authorized it. The operators who will be interesting in this next phase are not the ones with the most agents in production. They are the ones who can answer that question about any one of them, on demand, across a boundary they do not own. That answer is not a feature of a runtime. It is the plane the runtime runs on, and the industry has been shipping the second while promising the first.

The Telco AI $60 Billion "Opportunity"

Google Cloud published a piece in RCR Wireless this morning arguing that agentic AI represents a sixty billion dollar opportunity for telecom operators. It is a well constructed argument, the engineering description is accurate, and the case studies are real. It is also, read carefully, an argument about cost avoidance wearing the vocabulary of growth. I want to be precise about this, because eight days ago I published a piece arguing that the industry's central discipline problem is its refusal to separate the two, and this article is the cleanest illustration of the problem I have seen since.

Start with the numbers. The sixty billion figure comes from Appledore Research, and Appledore is explicit about what it is measuring: operational cost savings by 2030. The McKinsey research cited alongside it reports a thirty to seventy percent reduction in troubleshooting tickets and a fifty five to ninety percent reduction in network operations centre costs. Deutsche Telekom's RAN Guardian identified 237,000 network events in early 2026 and compressed major incident handling from hours to about sixty seconds. Bell Canada's AI Ops platform achieved a twenty five percent reduction in customer reported issues and a faster mean time to repair. Vodafone's agents protect millions in annual operating expenditure. Every one of those is a genuine achievement. Not one of them is revenue. The article's own evidence base is, without exception, my first money flow: AI that reduces cost, which I described as real, happening, the largest near term financial impact of AI on operators, and emphatically not a new line of business.

The word doing the concealing is "opportunity". An opportunity, in the way a board hears it, is something you invest in to get money back that you were not getting before. A cost saving is something you invest in to stop spending money you were already spending. The two justify different capital, different organisational patience, and different governance. Operators that hear sixty billion and staff a growth programme will find, three years in, that they have built a very good efficiency programme and told their investors the wrong story about it. That is not a hypothetical failure mode. It is the failure mode the industry has run repeatedly, and the reason I keep insisting the flows be kept apart on the page before they are kept apart in the budget.

There is a second problem with the sixty billion, which is that it is sitting next to a Deloitte figure of a hundred and fifty billion in "total value" and the two are quietly being read as the same kind of number. They are not. One is a cost line, the other is a mixed construct that includes cost, defended revenue and speculative new revenue in a single total. Adding vendor and consultancy figures that measure different things is the same error I flag on RAN energy savings, where individually plausible percentages get stacked into a number no operator has ever achieved. Treat the sixty billion as the honest number, because at least you can tell what it counts.

Now to the architecture, which is where the article is most interesting and most incomplete. Google Cloud's prescription is a fabric of hyper specialised micro agents, billing agents, inventory agents, RAN guardians, communicating through standardised orchestration protocols, validated against a digital twin, bounded by what it calls a deterministic governance framework with explicit decision boundaries and clean handoff to human engineers. This is good engineering. It is also, for at least the sixth time in a month, a description of containment rather than coordination. Every agent in that picture belongs to the operator. Every protocol is internal. Every boundary is a boundary between the operator's machine and the operator's human. Nothing in the design describes what happens when an agent that the operator does not own, and cannot inspect, arrives with a request.

The digital twin makes the gap unusually visible. A twin is a high fidelity replica of your own network, and it is exactly the right tool for testing a configuration change before you ship it. It is useless for the case that actually matters commercially, because you cannot build a twin of the counterparty. When an enterprise's AI agent negotiates for a guaranteed slice, the operator's agent is not reasoning about a system it can simulate. It is reasoning about an intent it must infer, an authority it must verify, and a commitment it must be able to audit afterwards. That is not a simulation problem. It is a problem of shared topology, shared ontology, explicit authority boundaries and durable audit trails, which is the meta model of the agentic plane I have been arguing for since the spring, and which a runtime does not supply no matter how good the runtime is.

I should say plainly that a hyperscaler making this argument is not a criticism of the hyperscaler. Google Cloud is selling a stack that does what it says it does, and the operators quoted are getting real results from it. My argument is with how operators will read it. There is a version of the next two years in which the industry retools its operations beautifully, takes out a very large amount of cost, calls the result an AI business, and arrives in 2030 with the same revenue line and a smaller headcount. That would not be a failure of technology. It would be a failure to name what was bought.

So the test I would apply to this article, and to every agentic AI business case that lands on a telco investment committee this quarter, is the one I set out eight days ago. Which flow is this, cost, defence, or new revenue? If the supporting evidence is entirely tickets, incidents and NOC headcount, the answer is cost, and the paper should say so in its first sentence rather than its appendix. What binding constraint does an external buyer pay to remove? If nobody outside the company pays anything, there is no buyer, and the word opportunity is unearned. And where on the capacity, platform, outcome ladder does this sit? An operator that automates its own operations has not stepped onto the ladder at all, because the ladder is about what you sell, and nobody is buying your NOC.

Sixty billion dollars of avoided cost is worth having. It is worth a serious programme, serious money and serious executive attention. It is worth all of that as what it is. The operators that will be interesting in 2030 are not the ones that saved the most. They are the ones that could still tell you, at the end of it, which of the three flows each dollar came from.

AI monetization for operators: separating revenue from cost avoidance

Every operator earnings call now features AI prominently. Listen closely, however, and most of what is described as "AI monetization" is nothing of the sort. It is cost avoidance cosplaying a revenue costume.

This distinction matters because the two require different investment logic, different organizational capabilities, and different patience horizons. Operators that blur them will misallocate capital. Operators that separate them have a chance at building genuine new B2B revenue lines — narrower than the hype suggests, but investable.

Three money flows, not one

AI touches operator economics through three distinct channels, and the discipline starts with refusing to aggregate them.

1. AI that reduces cost. Autonomous network operations, agentic customer care, energy optimization, predictive maintenance. This is real, it is happening, and it is the largest near-term financial impact of AI on operators. It is also not revenue. A dollar of opex avoided is valuable, but it does not create a new line of business, and it does not justify the "operators as AI companies" narrative. It justifies a leaner operator.

2. AI that defends existing revenue. Enterprises deploying AI workloads have new connectivity requirements: deterministic performance, low latency to inference endpoints, secure private connectivity to GPU capacity, data-gravity-aware networking. Operators that serve these requirements protect and modestly grow their core B2B connectivity business. This is differentiated connectivity for the AI era — important, defensible, but fundamentally an evolution of what operators already sell.

3. AI that creates new revenue. This is the category everyone wants to talk about and the one that deserves the most scrutiny. It exists, but it is narrower than most strategy decks admit.

The four credible new revenue lines

Having spent the last two years working on AI infrastructure with operators and vendors on both sides of the Atlantic, I see four B2B revenue opportunities that survive contact with commercial reality. They are not equal — they differ in demand maturity, margin profile and time horizon, and they should be funded accordingly.

Sovereign AI capacity — GPU-as-a-service and AI factories — is the most immediate and the most misunderstood. Demand is real and policy-driven, concentrated in regulated sectors; the margin profile is low-to-mid, because the business is capex-heavy and carries utilization risk; and the revenue is available now. The demand side is genuine: governments, healthcare systems, defense, financial services and public administrations in Europe increasingly cannot — or will not — run inference on US hyperscaler infrastructure under foreign jurisdiction. Operators hold assets that map remarkably well to this demand: national data center footprints, energy contracts, security clearances, sovereign trust, and enterprise sales relationships.

Telefónica's recent national rollout of edge-based GPU-as-a-service in Spain is instructive. The underlying edge platform was architected years earlier — I led the team that built and productized it — and for years the business case was marginal on enterprise use cases alone. What changed was not the technology. It was the arrival of sovereign AI demand, which finally gave the infrastructure a paying anchor tenant profile. The lesson generalizes: edge and distributed compute investments become fundable when sovereignty is the demand driver, not the garnish.

The caution: this is a capex-intensive, utilization-sensitive business competing against hyperscalers with structurally lower unit costs. Operators win where sovereignty, data residency and proximity are binding constraints — and lose everywhere else. The addressable market is the regulated slice of national demand, not "the AI market."

Edge inference is real but earlier than its promoters claim. Demand exists where latency or data gravity bind; margins are mid-range; and the horizon is two to five years before this becomes a broad product line. The use cases that pay today are those where physics or data gravity make centralized inference impossible: industrial vision, real-time media production, autonomous operations in ports and factories. I have seen these work commercially. But the buyer set is narrow, and each engagement still resembles a system integration project more than a product sale. This becomes a scalable product line when agentic AI workloads distribute themselves across infrastructure tiers — the architecture I have described elsewhere as the AI Grid. That shift is underway, not arrived.

Data and trust services are the sleeper. Deepfake detection on voice calls, branded and verified calling, identity assurance for AI agents, provenance services. These are small revenue lines today, but they are high-margin, they monetize immediately, they sit directly on operator trust assets that hyperscalers cannot replicate, and demand grows with every AI-enabled fraud headline. For a B2B operator, this category has the best margin-to-capex ratio of the four.

Network APIs are the line whose trajectory has changed most in the past two years. The strategic logic has always been sound — AI agents will need to programmatically request network resources, quality on demand, location, verification — and the commercial signals are finally following: revenues are growing, aggregation initiatives have consolidated distribution, and enterprise visibility is rising with every agentic deployment that needs verified identity or guaranteed quality. It remains the earliest-stage of the four, and the AI agent wave — rather than developer evangelism — is what gives it genuine demand pull. I would invest now to be positioned, while sizing near-term revenue expectations with discipline; the inflection is likely in the second half of the decade.

The monetization ladder

Across all four lines, there is a ladder that determines margin and defensibility:

Sell capacity → sell platform → sell outcomes. Capacity here includes every consumption-metered unit: GPU hours, tokens, gigabits.

Selling raw capacity — GPU hours, token-metered inference, connectivity — is rung one: necessary, low-margin, commoditizing from day one. Tokens deserve a specific caution here: metering in tokens rather than GPU-hours changes the billing unit, not the business. An operator selling tokens against someone else's models and someone else's stack is still selling capacity, at prices that will be set by the most efficient infrastructure provider in the market. Selling a platform — inference-as-a-service with orchestration, security, compliance tooling — is rung two, where margins improve and switching costs appear. Selling outcomes — a fraud-detection rate, a production workflow, a compliant AI deployment for a hospital group — is rung three, where the economics finally resemble a services business worth building.

Operators historically stall at rung one. The reasons are organizational, not technological: product management that thinks in network elements rather than buyer problems, sales forces compensated on connectivity, and business cases that demand payback before the platform layer has time to mature. The operators that climb the ladder will be those that treat AI monetization as a product management and go-to-market transformation, not an infrastructure deployment.

What the buyer actually pays for

A final discipline. In every commercially successful case I have worked on, the enterprise buyer was not paying for "AI." They were paying for a constraint to be removed: data that could not leave the country, latency that broke the use case, a fraud pattern that was costing millions, a compliance requirement that blocked deployment. Price the constraint, not the technology. The moment an operator's AI proposition cannot name the constraint it removes, it is a science project.

Three questions before approving any operator AI business case

1. Which of the three money flows is this — cost, defense, or new revenue? If the answer mixes them, send it back.
2. What binding constraint does the buyer pay to remove, and why is an operator structurally better placed to remove it than a hyperscaler or an integrator? Sovereignty, proximity and trust are acceptable answers. "We have a network" is not.
3. Where does this sit on the capacity–platform–outcome ladder, and what is the credible path up? Rung-one economics with rung-three ambitions is where operator AI investments go to die.

The AI B2B opportunity for operators is real. It is also smaller, slower and more demanding of commercial discipline than the current narrative suggests. The winners will not be the operators with the most GPUs. They will be the ones that can tell the difference between a cost saving, a defended revenue and a new business — and fund each accordingly.

The Agent Runtime Is Not the Agent Model

DTW Ignite in Copenhagen made one thing clear: the vendor community has decided that the path to autonomous networks runs through agent runtimes. NVIDIA introduced NemoClaw blueprints and the OpenShell secure runtime to give long-running agents policy guardrails and sandboxed access to telecom systems. AdaptKey is piloting security-hardened agents for self-healing 5G operations. ServiceNow is bringing Project Arc to the NOC, orchestrating incident response from alert to work order. NTT DATA is building anomaly agents that escalate to research agents for telemetry analysis. Synthetic data rounds out the stack, a pragmatic answer to the fact that more than half of operators say their most valuable network data is too sensitive to use.

This is genuine progress and I do not want to minimize it. Containment, auditability and policy enforcement are necessary conditions for letting agents touch production networks. An agent that cannot be sandboxed cannot be trusted, and an agent whose actions cannot be audited cannot be certified. The runtime layer has to be built.

Containment is not coordination

But look carefully at what these announcements govern: individual agents, operating within a single operator's domain, executing workflows that a human has scoped in advance. This is vertical governance. It answers the question of whether an agent is allowed to perform an action. It does not answer the question that autonomous networks will actually pose at scale: when two agents are each permitted to act, and their permitted actions conflict, who decides?

Consider a scenario that is closer than most operators think. An enterprise logistics agent requests guaranteed throughput for a fleet of delivery robots. Simultaneously, a network energy agent, operating under its own perfectly valid mandate, is shutting down capacity in the same cluster to meet a sustainability target. Both agents are sandboxed. Both are auditable. Both are compliant with their policies. The runtime layer sees two well-behaved agents. The network sees a contradiction.

This is the problem I described in my previous post on network APIs. APIs were designed for developer access, not for agent-to-agent negotiation. Runtimes inherit the same blind spot. They secure the execution of each agent without providing any shared representation of the agentic plane itself.

What the meta-model requires

For agents to negotiate rather than collide, the industry needs a meta-model of the agentic plane: a topology of which agents exist and where they sit, an ontology so that an enterprise agent and a network agent mean the same thing by capacity, latency or priority, explicit authority boundaries defining what each agent may commit on behalf of its principal, shared state models so that negotiations reference the same view of the network, and audit trails that span negotiations rather than individual actions. None of the DTW announcements address this layer. They cannot, because it is not a product any single vendor can ship. It is a model the industry must agree on, the way it once agreed on network information models for OSS.

There is a familiar pattern here. The industry built firewalls before it built routing protocols for the internet's trust boundaries, and it spent two decades paying for the sequencing. We are building the firewalls of the agentic era first. The operators and standards bodies that formalize the agentic plane meta-model will define how enterprise AI and network AI transact for the next decade. The ones that stop at the runtime will discover that a network full of safely contained agents is not an autonomous network.

DTW Ignite 2026: The API Is Not Enough

I returned from DTW Ignite in Copenhagen with one conviction: the interface between enterprise applications and network infrastructure is about to change in a way the industry has not yet designed for.

Network APIs were never really about autonomous networks. That framing conflates two separate problems. APIs — CAMARA, GSMA Open Gateway, the decades of network exposure work that preceded them — were designed to let developers discover and consume network resources from outside the operator domain. Quality on Demand, location services, device status, number verification: clean REST interfaces exposed through a developer portal so that a programmer writing a B2B application could request a network capability and pay for it. Real progress on a real problem. But the problem was developer access, not network autonomy.

What is coming next is different in kind, not degree.

Enterprise AI agents are beginning to consume network infrastructure directly — not through a developer writing an integration, but autonomously, in real time, as part of executing a business objective. An industrial automation agent that needs guaranteed low-latency connectivity for a robotics fleet. A financial services agent that needs to provision a secure, isolated network path for a time-sensitive transaction. A logistics agent that needs to dynamically reserve bandwidth across multiple carrier domains as a shipment moves between jurisdictions. In none of these cases is there a developer in the loop. The agent has an intent, it needs network resources to fulfil it, and it needs to negotiate those resources with the network — now, at machine speed, without human mediation.

That negotiation cannot happen through a developer portal. It cannot happen through a static API catalogue with a PDF explaining what each endpoint does. The enterprise agent and the network need to speak to each other, and neither CAMARA nor MCP — whatever their respective merits — were designed for that conversation.

The network side of this exchange needs to be represented by network AI agents of its own: agents that can expose available capacity in real time, understand the constraints and commitments already in place, reason over competing demands, and negotiate resource allocation in a way that respects the network's operating boundaries. That is not a developer API. That is an autonomous counterparty.

And for those network agents to function — to negotiate reliably, to be governed, to be audited, to avoid conflicting with each other across RAN, transport, core, and the operational layers of OSS and BSS — they need something the industry is not yet building: a meta-model of the agentic plane itself.

Operators building autonomous networks are doing the right foundational work. Network topology models. Data ontologies. Decision layers. Closed-loop control architectures. These give the automation layer a complete and current picture of the environment it is operating in. But agents operating on that network need an equivalent model of themselves. Every agent with an identity, a capability scope, an authority boundary, a state, a dependency graph, and an audit trail. An abstract topology and ontology of agents, sitting alongside the topology and ontology of the network.

Without that model, what looks like autonomous negotiation between enterprise AI and network AI is actually uncontrolled interaction between systems that cannot see each other. An enterprise agent requesting bandwidth does not know what the network agent is authorised to commit. The network agent does not know what other network agents have already promised. No shared representation, no conflict detection, no governance.

The developer exposure problem is largely solved, or at least well understood. The agent-to-agent negotiation problem has barely been framed. That is the conversation the industry needs to have, and Copenhagen convinced me we are not having it yet.

From AI-Native to Agentic-Native Networks

Recent announcements at NVIDIA's GTC 2026—including major pushes into agentic AI frameworks like OpenClaw, NemoClaw, and agentic systems for reasoning, planning, and autonomous action—have reinforced several convictions I've held about the trajectory of AI-native infrastructure, especially in telecom and networked industries. We're seeing the emergence of two distinct paradigms:

• AI-Native networks that observe, detect, optimize, and predict in real time. These systems augment human decision-making, providing powerful assistance in planning, deploying, and managing both physical and virtual infrastructure.
• Agentic-Native networks, by contrast, eliminate the human-in-the-loop entirely. When equipped with real-time data access, transactional capabilities, and fulfillment capacity, they execute at the speed permitted only by the slowest link in the supply chain.

This second model doesn't just accelerate execution—it fundamentally reprices time itself as a competitive asset.

As Jordi Visser articulates in his insightful piece "The Repricing of Time: Equity in the Age of Agents", agentic AI compresses competitive cycles dramatically. Velocity of execution no longer merely helps fulfill a plan faster; it redefines the playing field. When capabilities can be reconfigured almost overnight through model iterations or agent orchestration, durable moats erode. What once took decades to build—layered expertise, entrenched positions, regulatory barriers—can now be challenged or leapfrogged in months.

In this environment, equity behaves more like a call option on execution speed than on long-duration stability. "Execution speed replaces installed base. Iteration cadence replaces headcount." The advantage shifts decisively toward those who can pivot, adapt, innovate, and execute rapidly.

This dynamic hits telecom particularly hard.

Most operators are desperate to escape the "utility trench"—the low-margin, commodity perception that has trapped connectivity providers for years. They aspire to new revenue streams beyond pipes and bandwidth.

From my own experience modeling, teaching, and advising organizations on this challenge (see earlier pieces on innovation micro-strategies, telco relevance and growth, and the lean telco), there is no single silver bullet. No grand transformation program that magically reinvents the business.

Instead, the path forward involves thousands of micro-services and experiments: create, test, fail fast, pivot, scale the winners, and launch repeatedly. The era of one-size-fits-all offerings is over.

Agentic-native networks offer exactly the infrastructure to make this high-velocity approach viable at scale. They enable rapid creation, iteration, value capture, and deployment—turning velocity, flawless execution, and clear strategic vision into the new currency that outcompetes inertia, legacy systems, and eroding differentiation.

For telecom leaders, the message from GTC 2026 is clear: agentic AI can free up resources and help accelerate innovation at scale. Those who embrace this shift—building or partnering for agentic capabilities—will be the ones that don't just survive the repricing of time, but help define the next era of networked value creation.

The philosophical problem with agentic AI

Jensen Huang’s address at GTC gave me a lot to think about. So much so that I decided to drive to Sana Cruz for a taste of the ocean. I had to wait 30 minutes to get the table I wanted, just by the beach, in the sun but with a little shade… as I mistype table on my iPad, I am thankful for the autocorrect to sanitize my  somewhat boozy prose, while mostly appreciating the elegantly subtle blue underlying of the word batle, prompting me to consider “is that really what you meant to write, or do you meant table”?

I like that. I like that more than the blue pencil with the little star that insistently offers an AI assisted rewrite. Oh, sure, I am not a native English writer, so my grammar is somewhat tainted by the other 3 languages I might think in at any point in time. If I compound St Patrick and this weekend’s VI nations rugby results for France, you will understand if my writing is not the usual corporate polish. 

Having said that, I was at GTC for the first time, I listen to Jensen’s performance and I was left enlightened and a bit worried. By now, the headline and the sound bite out there must be the $1 Trillion line of sight on chip revenues for Nvidia over the next couple of years. Obviously, it is an extraordinary number. Unfathomable. Impossible to imagine for most of us. Almost impossible to think that we, collectively would spend 125$ ( at 8 billion people) of Nvidia stuff over the next couple of years. Surely that’s impossible. 

Unless this is not about need, but about demand.  Unless that demand is accelerated, compounded, exponentially nurtured beyond its natural curve. 

Essentially, what I retained from the presentation was that the larger the model, the more the interactions, the larger the demand, the faster and more the tokens have to be created to satisfy it. (I am sure AI could rewrite this sentence more elegantly, but screw it). The measurement unit becomes token per Watt,as it is a limiting factor for a given data center and tokens per second as it is the limiting factor for a given service. Jensen even alluded to the fact that they will factor in token per month grants in engineering packages as it becomes a productivity factor. 

The thesis for the 1T$ revenue relies on demand exploding and the emergence of low latency, high I/O token market. Low latency, high I/O is understandable. Multimodal, video models, requiring real time inferencing from vehicles, robots and generally physical AI will drive it. The demand explosion, though, even factoring in the integration of compute and AI in to its, devices, edges… if we look at adoption curves and industrial capacity is decades away,  not in 2 years. Unless…

Unless we are not the demand. Us, consumers, enterprises, industries, governments… Agentic AI and Clawdbot are just showing how, beyond automation, agency becomes a compounding factor. Agents, that you create, for specific purpose are understandable, useful controllable. 

Agents, that interpret your intent, create other agents to enact their interpretation, have access to your digital life, credit card, HR, accounts receivables, invoices, orders, security cameras, GPS movements better be accountable, auditable, controllable. Agents that create fleets of agents to parcel out their workload is where I have doubts. The d’explosion in demand relies on the hypothesis that we will let agents create agents consume tokens to satisfy our needs.

No doubt, we will have agents to control, audit, police agents, but it feels wrong to delegate tasks just because you can or for the concept of efficiency.

This is where the the philosophical debate clashes with the economic model. I learned that hard times create hard men. Hard men create easy times. Easy times create easy men. Easy men create hard times. We might have evolved from this adage, but I feel that, being a kinetic, rather than a literal learner, I’ve learned from trying. I’ve learned from friction. To this day, I write on my notebook with a pen. I don’t forget anything I write. I forget most of what I type. It feels to me that friction is an integral part of the learning experience. More, it is an integral part of the human experience. The taste for effort, trying the hard things, failing is not only what most mankind experience on a daily basis, it is also, at least for me a great  condition to happiness. I am infinitely happier labouring and succeeding than an automated, frictionless, efficient experience. Even with a better result.

As my children are about to enter the workforce, I am confronted daily to the question “what is a safe, fulfilling carrer?”. It used to be that medicine, law, engineering guaranteed a safe economic path. Nowadays, it looks like most entry level intellectual effort can easily, efficiently be replaced, and that agentic AI will only accelerate that trend. How are they supposed to master a domain they won’t be able to tinker and stumble? Maybe I am just an old fart and just like calculators and computers did not replace engineers, a higher level of abstraction will necessitate higher levels of intellectual efforts ? But this feels different. 

Particularly if compute keeps accelerating and artificial intelligence surpasses human intelligence, then what? What is the imperative to learn, labour, try, suffer, if is not necessary? Where do you draw the line between agents that help and augment and agents that enable and replace?

Until then, I’ll keep labouring and burdening you with poorly written posts, but somewhat original or at least unique, because they’re mine. I enjoy this table, i waited 30 minutes for because I chose it and waited for it. I am not sure it would have tasted better should my personal AI butler had booked it for me on my way there.