Tellefsen SPC

On Friday night, Anthropic’s two most powerful AI models went dark for every user on earth, on a US government order — three days after launch . Other models kept running. But if your business ran on Fable 5, here’s the question that actually mattered that afternoon: where did your product live? If it lived inside that one model, Anthropic’s problem became your problem. No model, no product. You wait for Washington. If it lived in your own operating layer — your contacts and the workflows that actually run the business — a model going dark was a config change. Point the layer at another model and keep going. Your customers never notice. That’s the whole thing. The operating layer is everything. The model is a tenant in it, and you should be able to swap one out by the afternoon. For anyone running an enterprise in the Gulf, this isn’t abstract. The order named foreign nationals specifically . Your access to a critical model can be cut overnight — for where you sit, not for anything you did. Frontier models are too good to ignore. But don’t let one become a foundation you can’t replace. We sell the work, not the tool. Friday is why we mean it literally. You can lose a tool. You don’t lose the layer that knows how your business runs. #EnterpriseAI #AI #Oman

Every enterprise we work with can tell you their AI bill. Almost none can tell you what it bought. Link: https://lnkd.in/d4VzN9pi That gap is becoming one of the defining governance problems of enterprise AI. A finance leader sees “one million tokens” on an invoice — not which team, which model, or which workflow produced it. Spend climbs, and there’s no lever to pull. It’s tempting to treat this as a reporting problem. It isn’t. You can’t analyse usage you never captured. Most organisations are blind to their AI spend not because the tools are weak, but because nothing sits between them and the models, recording who is spending what, and why. Visibility isn’t a report you run later. It’s an architecture you put in place from the start. This is the part of AI that gets too little attention. The model is the easy part — increasingly a commodity. The value, and the risk, lives in the layer you build around it: governance, attribution, control. So we built that layer. The AI Spend Control Layer gives an organisation a live, itemised view of its AI spend — by team, by model, by use case — deployed inside its own environment, where the data stays. The economics of enterprise AI are about to matter as much as the capability. The organisations that see clearly will be the ones that built to see. #EnterpriseAI #AIGovernance #FinOps

Adoption used to be the hard part. It isn't anymore. A recent Korn Ferry study put numbers on something we've been watching across the Gulf: more than nine in ten organisations here now use AI in some form. The pilots are everywhere. What almost none of them have done is scale — move AI out of one team's experiment and into the way the business actually runs. It's tempting to read that as a maturity gap that more pilots, more tools, or more training will close. It won't. The wall isn't capability. It's that the AI has nothing underneath it to stand on. In every engagement we run, the pattern is the same. Each system holds its own slice of the truth, and none of them talk to each other. Teams bolt agents onto APIs that don't share context. They make decisions on data that can't see across the business. The model was never the problem — there's no operating layer for it to work on. That layer has four parts, and they're worth naming: A context graph — the shared map of what the business knows. Orchestration — the logic that moves work between systems and people. Governance — the rules for what AI can touch, decide, and log. Connectors — the live links into the systems of record you already run. Build that, and AI stops being a pilot you're impressed by and becomes operations you can trust. Skip it, and you join the nine in ten. We sell the work, not the tool. This is the work. #EnterpriseAI #OperatingLayer #GulfAI

Our founder, Christoffer Tellefsen, recently joined the AmCham Oman Technology & AI Committee for an alignment session with H.E. Dr. Ali Al Shidhani, Undersecretary for Communications and Information Technology, and the leadership team at MTCIT. Two priorities anchored the discussion: building Omani technology talent, and giving Omani startups and SMEs real international exposure. Both point at the same thing. The constraint on AI in the Gulf was never access to the models — those are a download away. The real constraint is the local capability to turn them into operations that actually run: systems that hold up in production, in Arabic, inside Omani data-residency and regulatory realities. That capability doesn't get imported. It gets built — engagement by engagement, by people on the ground who deliver the work and leave the knowledge behind. It's why we build for outcomes, not hours, and why the systems we ship are designed to be owned and operated locally rather than rented indefinitely. Vision 2040 describes a digital economy that contributes to GDP. That contribution shows up when Omani teams and firms can build and run their own AI — not when they stay dependent on someone else's. Grateful to the Committee and to MTCIT for an open, forward-looking conversation. #OmanVision2040 #GulfAI

Your AI pilot didn't stall because the model wasn't good enough. It stalled because there was nothing underneath it to operate on. Every system in an enterprise holds one slice of the truth. The ERP knows the transactions, the CRM knows the customers, the document store knows the contracts — and none of them knows the whole. An AI agent bolted onto one of them inherits exactly that blindness. It reasons brilliantly about its slice and stays blind to everything else. This is the part the market keeps skipping. The 2026 enterprise surveys are blunt about it: integration is now the single biggest reason agents never reach production — ahead of model quality, and ahead even of ROI, which executives now rank near the bottom of their concerns. The capability is here. The substrate isn't. We call that substrate the Operating Layer — the missing middle between your strategy and your stack. Four parts make it work: - A context graph, so AI can see across systems instead of inside one. - Orchestration, so work moves between steps without a person stitching it together. - Governance, so every action is permissioned and auditable. - Connectors, so it runs on the systems of record you already have, rather than replacing them. Get those four right, and the pilot you couldn't scale becomes an operation you can run. Most AI budgets are still being spent one layer too high — on the tools, when the work lives in the middle. That's the layer we build. We sell the work, not the tool. #EnterpriseAI #AIStrategy #OperatingLayer