INSIGHTS · ENGINEERING NOTES

Notes from the engine room

Guides, challenges and thoughts from the engine room.

AI Sovereignty Isn't Data Residency. It's Megawatts, Fiber and Wet-Bulb Temperature. (1/3)

A new arXiv paper by Sergio Cruzes (Ciena) argues AI sovereignty has shifted from a software/legal problem to an infrastructure one. As a DevOps engineer who builds for regulated clients, I agree — and the implications break a lot of European sovereignty discourse that still operates one layer too high.

Read article →

AI Sovereignty Isn't Data Residency. It's Megawatts, Fiber and Wet-Bulb Temperature. (1/3)

All

111 articles

Challenges

AI Sovereignty Isn't Data Residency. It's Megawatts, Fiber and Wet-Bulb Temperature. (1/3)

May 13, 2026·10 min read

Challenges

The Feasible Sovereign Operating Region: Why Your AI Roadmap Hits an Energy–Carbon–Water Wall (2/3)

Cruzes' arXiv paper introduces a Feasible Sovereign Operating Region (FSOR): the joint set where energy, carbon and water constraints can all be satisfied. Most AI roadmaps I've reviewed this year optimise one of the three and quietly assume the other two. They don't compose. Here's what a real FSOR looks like on an infrastructure plan.

May 13, 2026·11 min read

Challenges

Let the LLM Talk, Not Touch: The Closed-Loop Architecture That Actually Survives Production (3/3)

Cruzes' arXiv paper draws a hard line: LLMs in AI infrastructure are advisory and interpretive only. Execution belongs to deterministic agents, validated by a digital twin, with two strictly separated feedback paths. As someone who deploys agentic AI for regulated clients, I think this is the correct architecture — and most current 'agentic' products silently violate it.

May 13, 2026·11 min read

Challenges

McKinsey 2026: AI Trust Maturity Hits 2.3. My Infrastructure Isn't Buying It Yet.

McKinsey reports an AI trust maturity score of 2.3 out of 5 and 23% of organisations scaling agents. As a DevOps engineer working in regulated environments, I read it differently: nearly two-thirds of respondents cite security and risk as the top barrier, and they're right. What you actually need running underneath the slide.

May 12, 2026·10 min read

Challenges

Google Cloud Next 2026: $200B in Capex Doesn't Buy Production Maturity

Thomas Kurian told us the experimental phase is over and three quarters of Google Cloud customers already run AI in production. As a DevOps lead, I read the announcements differently: the slop is a symptom, the real story is the infrastructure tax nobody on stage was willing to itemise.

May 11, 2026·9 min read

Challenges

Coinbase's Solo Operator: Where the One-Man Product Works and Where It Breaks

Coinbase has rolled out one-person pods that fold product, design, and engineering into a single role backed by AI agents. The model is powerful inside a contained domain. The trap is what happens when twenty solo operators hit the same production environment and nobody owns the shared substrate.

May 8, 2026·9 min read

Challenges

53% Recall: Why Microsoft's Own AIOps Confirms the Engineer Is Still Essential

ActionNex, Azure's AI outage manager, reports 71.4% precision and 52.8–54.8% recall on real incidents. What does that mean for a CTO running on-call? A critical reading from DevOps.

May 2, 2026·9 min read

Challenges

The Myth of the 'Self-Healing' Pipeline Without Humans: A Critical Reading from DevOps

A recent IEEE paper forecasts fully autonomous CI/CD pipelines with 'minimal human intervention'. From a DevOps role at enterprise scale, that's not the reality — and it's not the right goal either. What AI does automate, what it doesn't, and why the difference matters for your engineering budget.

May 2, 2026·10 min read

Challenges

AI Failure Modes Are Now a Top-of-Stack Concern: An Engineering Defense Playbook

Hallucinations, overtrust, prompt-level adversarial attacks, deepfakes, bias, opacity. Stanford's 2026 review treats these as features of current AI, not bugs. The engineering defenses are concrete and most teams aren't building them.

April 28, 2026·9 min read

strategy

The Geopolitics of Compute: CTO Strategy in a Fragmented AI Regulatory Landscape

Stargate's $500B target, EU AI Act enforcement, US state-level AI law fragmentation, brain drain, export controls. The Stanford 2026 review reads like a geopolitics brief — and the operational consequences fall on engineering leadership.

April 27, 2026·10 min read

Challenges

AI Agents in 2026: MCP, Memory Limits, and the Interoperability Wall

Stanford's 2026 review names the four limits constraining production agents: memory, reliability, interoperability, efficiency. MCP is solving one of them. The other three are still your problem.

April 26, 2026·9 min read

Challenges

Foundation Model Economics: How to Ship AI Without Owning a Frontier Lab

GPT-4-class training cost ~50M kWh and tens of millions of dollars. Inference is the line item that scales with users. A practical view on what mid-market teams should build, buy, and avoid — informed by the Stanford Emerging Technology Review 2026.

April 25, 2026·9 min read

NEWSLETTER

Technical insight straight to your inbox.

What a founder needs to know to make better technical decisions.

No spam. Cancel anytime.

Prefer to talk directly?

Our articles are the starting point. A call with a CTO is where real change begins.

Talk to a CTO →