By The Core Concept: Enterprise leaders who defer distribution decisions until after system design will face costly architectural rework — Danielle Cook‘s three-step action plan makes distribution, latency classification, and Kubernetes standardization foundational, not optional.
The Guest: Danielle Cook, Senior Manager at Akamai and CNCF Ambassador
The Bottom Line:
- Distribution must be an assumption at the start of system design — not a retrofit — with teams asking how close to users they need to be before writing a single line of application code.
- Not all AI workloads are equal: identifying latency-sensitive workloads early and placing them close to users is a discrete, actionable step that prevents performance failures at scale.
- Kubernetes, backed by a maturing CNCF ecosystem of agents and tooling, is the platform standard Cook recommends for reducing complexity and unlocking AI deployment velocity across any environment.
Speaking with TFiR, Danielle Cook of Akamai defined the current state of enterprise AI readiness and outlined the three actions engineering and platform leaders must take immediately to avoid architectural debt in their AI programs.
WHAT SHOULD ENTERPRISE LEADERS DO RIGHT NOW TO PREPARE THEIR AI INFRASTRUCTURE FOR 2026?
Cook’s first directive resequences how organizations approach system design. The default instinct — build first, distribute later — produces architectures that must be fundamentally restructured the moment AI workloads demand real-time responsiveness. The corrective is to make distribution an assumption before design begins, not a consideration added in a later sprint. “You need to assume the distribution at the start. Don’t think about it after the fact. Think about: do I need a distributed cloud? How close to my users do I need to be? What sort of experience do they want?” These questions must be answered at the very beginning of any new application or AI inference workload — not after the first production incident.
Classifying Latency-Sensitive AI Workloads
The second action is a practical filtering exercise that Cook frames as non-negotiable. Not every AI workload requires real-time execution — but the ones that do will fail visibly and immediately if they run too far from users. The discipline is identifying which workloads are latency-sensitive early in the design process and positioning them accordingly. “Identify which AI workloads are latency-sensitive, because not all of them will be. Identify this early, and move any AI workload that requires real-time experiences closer to the user.” This classification step is the bridge between architectural intent and infrastructure placement — and it prevents teams from over-engineering low-stakes workloads while under-serving the ones that define customer experience.
Standardizing on Kubernetes and the CNCF Ecosystem
The third action is platform standardization. Cook, speaking as a CNCF Ambassador, is direct: Kubernetes should be the platform organizations build on. She acknowledges the complexity but frames it as a solvable, ecosystem-supported challenge. The cloud-native community has produced agents and tooling that make Kubernetes progressively more effective, and organizations that standardize now can leverage that maturation rather than rebuilding later on a different foundation. “Look at it as a way to help you remove a lot of complexity. It’s complex in itself, but because of all the different ways the broader cloud-native ecosystem is evolving, there are huge opportunities. Look at it. Use it.”
Broader Context: The Strategic Logic Behind These Three Actions
These three actions are the operational translation of the architectural shifts Cook outlined across her full TFiR interview. Her 2026 predictions establish that AI inference placement is becoming a primary design choice, that distributed cloud is the default architecture, and that Kubernetes is achieving product-market fit as the portable runtime for AI workloads. The advice in this clip makes those macro-level predictions executable at the team level: design for distribution first, classify workloads by latency sensitivity, and standardize the platform layer on Kubernetes before complexity compounds.
Akamai’s managed Kubernetes engine (LKE), GPU-backed infrastructure, and distributed edge represent the production-ready version of this stack — designed so organizations can follow Cook’s advice without building the underlying capability from scratch.
Watch the full TFiR interview with Danielle Cook here.
