Cloud Dialogues

Engineering in the Age of AI

(apologies - audio is not up to our usual standard, but this version is the best I could get it) Guest: Lena Hall (Senior Director of Developer Experience at Akamai, formerly DevRel lead at Amazon Web Services and AI/Data Advocacy Director at Microsoft) joins Georgia and Matt to unpack the latest AI developments and what they mean for how we build software. News Highlights The episode kicks off with several major AI updates. The US government excluded Anthropic from a supplier list over concerns around WMD and surveillance policies - only for OpenAI to sign a government deal shortly after with similar language, raising questions about whether the decision was policy-driven or political. Meanwhile, OpenAI released GPT-5.4, a reasoning-focused "thinking model" with tunable reasoning depth. Early feedback suggests stronger accuracy and less verbosity, though it consumes more tokens and is slightly more expensive to run. The hosts also discuss a pledge from Meta, Microsoft, Google, and Amazon to fund new electricity generation to support AI infrastructure - a move framed as sustainability but widely seen as a practical response to AI's growing energy demand. Finally, a new partnership between CVS Health and Google Cloud highlights a broader shift in hyperscaler strategy: AWS continuing to focus on horizontal infrastructure while Google invests more heavily in vertical AI solutions such as healthcare. The Core Discussion: Engineering in the Age of AI The main conversation explores how AI systems fundamentally challenge traditional software engineering practices. Unlike deterministic systems, AI outputs exist on a spectrum of quality. A system may be operationally healthy yet still produce incorrect or harmful responses, creating a new category of production issues that are harder to detect and diagnose. Lena argues that while organizations don't necessarily need entirely new AI platform teams, platform engineering must evolve. Teams need infrastructure for AI observability, evaluation frameworks, fallback mechanisms, and intervention controls. Without this foundation, individual product teams end up solving the same problems repeatedly. A key takeaway is the need for clearer responsibility across three types of AI failure: capability issues owned by product teams, safety risks defined by leadership, and operational reliability managed by platform engineering. The group also emphasizes the importance of product-level evaluation, focusing not just on model benchmarks but on whether AI actually works for real users. Effective evaluation frameworks measure capability, safety, and operational reliability, with scrutiny increasing for higher-risk applications. For organizations adopting AI, Lena recommends a gradual approach: start with assistants for narrowly defined tasks, move to supervised agents, and only introduce autonomous systems once observability and governance are mature. AI and the Human Factor The discussion ends with the impact of AI on developers themselves. Engineers are spending less time writing code and more time making high-level decisions about architecture, system behavior, and trade-offs. While this can increase productivity, it also raises cognitive load and shifts responsibility toward more experienced engineers reviewing large volumes of AI-generated output. Cool AI Pick Lena's pick is Codex Spark, an ultra-fast model designed for executing well-defined tasks. Her preferred workflow combines reasoning models like GPT-5.4 for planning, then handing execution to Spark - highlighting a broader trend toward specialized models working together in AI development pipelines.

The Rise of the AI Platform Engineer

Cloud Dialogues – Episode 39 Guest: Ran Isenberg (Principal Software Architect at Palo Alto Networks, formerly CyberArk) 📰 News Roundup: AI Drama, Agent Governance & Layoff Myths Episode 39 kicks off with a tour through the latest AI headlines – and there was no shortage of spice. 1. Anthropic publicly accused companies including DeepSeek, MiniMax, and Moonshot AI of using fake accounts to scrape and distill their models – a bold move that sparked debate given Anthropic's own history with training data practices. Google reported similar behaviour but stopped short of naming names. 2. We also explored OSO HQ, a new startup building visibility and governance tooling for AI agents operating across enterprise systems – essentially, "what are your bots actually doing?" 3. Meanwhile, rumours of an outage linked to Amazon Web Services' AI coding tool KIRO were clarified as human error rather than rogue AI. A useful reminder that not everything is Skynet. 4. The "Open Claw" / Claude Bot social experiment – later acquired by OpenAI – got a mention too. Interesting concept. Chaotic execution. Classic internet. 5. Finally, the hosts pushed back on the narrative that AI is directly causing tech layoffs. The real story? A correction cycle following years of over-hiring, empire building, and governance gaps – not a sudden robot takeover. 🧠 Main Discussion: AI Platforms – Welcome to the New Wild West The core theme: AI tooling inside organisations is starting to look suspiciously like early cloud adoption. Shadow AI. Tool sprawl. Unmanaged access. Duplicate spend. No clear ownership. Ran argued that platform engineering teams must step into the AI governance vacuum. That means: - Curating approved MCP servers and integrations - Defining and managing organisational "skills" (context files guiding AI agents) - Building observability into agent activity - Providing secure self-service templates for agentic services - Treating governance as an ongoing capability – not a slide deck exercise The key message: publishing a framework isn't governance. Ownership, accountability, and maintenance are. 🔁 AI & The SDLC: Developers as Architects The software development lifecycle is evolving fast. Developers are increasingly acting as architects and product owners – guiding AI agents through structured loops of: Plan → Verify → Validate → Execute Rather than writing every line of code, they're shaping specifications, validating outputs, and managing state through context files. Spec-driven development – where AI maintains project memory – emerged as a particularly promising model. Ran's practical advice: - Test frameworks using real tasks (not demos) - Measure quality, cost, and performance - Gather feedback from actual developers - Roll out via pilot teams before scaling Translation: treat AI adoption like an engineering transformation – not a hype cycle. 👀 Shadow AI: Blocking Isn't Strategy The episode closed with a pragmatic take on "Shadow AI." Blanket bans on tools like ChatGPT don't build capability – they just push usage underground. A smarter approach combines: - Education and cultural norms - Clear guardrails - Detection and observability tooling - Secure internal alternatives Because people will use AI. The question is whether they'll use it safely – or secretly. Visit Ran's blog here: https://www.ranthebuilder.cloud/blog

Entering 2026 - The operational state of AI & Cloud

The Operational State of AI & Cloud We’re kicking off 2026 with a reality check. In this episode, Matt, Georgia, and special guest Allen Helton (Ecosystem Engineer at Memento, AWS Hero, and, yes - farmer) dig into what’s actually happening in AI and cloud right now. Less hype, more hard truths. From AI pilots that won’t scale to power grids that can’t keep up, this conversation explores what it really takes to move from experimentation to production. 🎙️ Hosts & Guest * Matt — Host (Texas) * Georgia — Host (London) * Allen Helton — Ecosystem Engineer at Memento, AWS Hero, and farmer 🗞️ Cloud & AI News: What’s Worth Paying Attention To GPT Health: Innovation or Repackaging? The team unpacks OpenAI’s GPT Health launch, questioning whether it’s a genuinely differentiated product or simply a safer wrapper around existing capabilities. Georgia shares how ChatGPT proved unexpectedly useful for post-surgery aftercare - sometimes outperforming traditional medical guidance. AWS Is Back in Growth Mode AWS reported ~20% year-on-year growth in Q3, its strongest in nearly three years. The consensus? AWS has finally caught up on AI - largely thanks to its Anthropic partnership and global access to Claude through Bedrock. Quantum Computing: Is 2026 the Tipping Point? IBM predicts quantum computers will outperform classical systems as early as 2026. The group discusses what that could mean for cryptography, banking, and security - while openly admitting that quantum still needs more expert decoding. Power Is the Real Bottleneck Google flags US transmission infrastructure as the biggest blocker for data-center expansion. That sparks a broader sustainability discussion: hyperscalers can’t depend on aging grids forever, and renewables aren’t optional - they’re inevitable. 🧠 The Operational Reality of AI & Cloud Your Data Foundation Still Isn’t Ready A recurring theme: organizations move “two steps forward, one step back” when AI exposes weak data governance and cloud foundations. As Georgia puts it: AI will not solve your data governance problems. The Education Gap Is the Silent Killer AI initiatives fail when business teams don’t understand the technology they’re adopting. Outsourcing isn’t enough - successful organizations immerse their entire teams so AI outputs are interpreted, validated, and trusted. Are We Really Past Pilots? Some say the pilot phase is over. Alan disagrees. Large parts of the industry are still early on the adoption curve - but the difference now is maturity: guardrails, retrieval systems, and meta-agents are production-ready. 👩‍💻 How AI Is Changing Software Careers AI isn’t just changing how software is built - it’s changing who gets hired. Key shifts discussed: * Programming language choice matters less than ever * Code review, comprehension, and reasoning now outweigh writing from scratch * Systems thinking is becoming table stakes - even for junior roles * “Tech-lead thinking” is creeping into every level Alan’s advice to students and early-career engineers: You still need to understand how it all works - everything you write is part of something bigger. 🧩 Developer Operating Models: What Actually Scales? Ralph at Scale Matt introduces Geoffrey Huntley's Ralph Wiggum development approach: giving an LLM an ordered backlog and letting it execute autonomously across fresh context windows. Powerful - but expensive and hard to sustain. The “Gas town” Model An alternative approach uses 30-40 agents working in parallel across a stack. Fast, impressive… and extremely token-hungry and even more expensive! The Sensible Middle Ground Our hosts argue for balance: AI-accelerated delivery with strong human oversight. Think weeks of work compressed into afternoons - without sacrificing quality, maintainability, or understanding. 🔮 Looking Ahead Regional Model Availability Is a Deal-Breaker Many regulated organizations simply can’t adopt AI due to regional model restrictions. Australia, for example, has access to just one local foundation model - highlighting a global compliance challenge. Sustainability & Reliability Risks If models became unavailable or prohibitively expensive, productivity would fall off a cliff. Competition should help manage costs - but reliability at scale may be the bigger risk. The Adoption Curve Has Never Been Wider AI adoption now spans: * Teams using autonomous coding agents daily * Enterprises still waiting for approval to touch an LLM Most regulated industries haven’t even started formal approval processes. ✅ Key Takeaways 1. Data governance is still the biggest blocker to AI success 2. Developer roles are shifting toward systems thinking and code comprehension 3. Enterprise AI adoption is far lower than headlines suggest 4. Regional model availability is a serious global constraint 5. Power and sustainability will shape the future of cloud growth 6. There’s no single “right” AI operating model 7. Business teams must deeply understand the tech - not just fund it 📬 Closing Notes Alan plugs his newsletter Ready Set Cloud of the Week (readysetcloud.io), where he curates and analyzes the most interesting tech stories each week. As always, we’d love to hear from you. Feedback, guest ideas, and topic suggestions → feedback@cloud-dialogues.com Cloud Dialogues is a podcast for technology leaders navigating cloud, AI, and enterprise transformation—grounded in reality, not hype.

AWS re:Invent 2025 Wrapped

Matt and Georgia recap AWS re:Invent 2025 with special guest Michael Walmsley, AWS Serverless Hero and Global Technology Architect at Accenture. Fresh from the Vegas event with 70,000 attendees, they discuss the major announcements, the shift toward AI agents, and Michael's wild experience coding on a bus for a $100K hackathon prize. Highlights Road to re:Invent Hackathon * 50 developers coded on buses traveling LA to Vegas over 5 hours * Michael's team built "Lucky Loo.me" - an AI bathroom finder using facial recognition * Winning team created "Oric" - an IDE that turns 3 lines into 3,000 lines of AI slop * Prize: $100K split among the winning team The Big Theme: AI Agents Everywhere * "Agents" was the dominant word at every booth * AWS pushing agent capabilities into every service team * Evolution from general AI (2024) to production agent platforms (2025) Announcements we covered: Agent Core Updates * New policy controls for blocking unauthorized actions * Evaluation tools for inspecting agent behavior * Progressive adoption - use pieces without adopting the whole platform AWS Agent Marketplace * Vendors can now sell pre-built agents * Example: Cloud Zero cost management agent Lambda Updates * Lambda managed instances * Durable functions for long-running workflows in code * Alternative for developers who don't want Step Functions S3 Vectors (GA) * Store 20 trillion vectors in one bucket * 90% cost savings vs traditional vector databases * Sub-100ms query times for frequent queries * "S3 is the cheapest database on the planet" CloudWatch Unified Data Store * All logs and metrics exposed in S3 Tables * Cheap, structured SQL querying of observability data AWS Interconnect ⭐ Biggest Surprise * High-speed encrypted links between AWS and Google Cloud * Azure support coming 2026 * Free during preview (pricing TBA) * Major shift from AWS's anti-multi-cloud stance * Acknowledges multi-cloud reality in enterprises Kiro * Rebranding away from confusing "Amazon Q" umbrella * Kiro Powers: AI-activated tool modules * Reduces context bloat in coding agents * Active hackathon scene with significant prize pools Guest Michael Walmsley - AWS Serverless Hero, Global Technology Architect at Accenture, specializing in serverless and SaaS architecture. Fourth year attending re:Invent. Key Takeaway AWS is maturing from general AI capabilities to production-ready agent platforms while finally embracing multi-cloud architectures. The focus has shifted to making agents secure, manageable, and practical for enterprise use.

Powering AI Data Centers, Energy Demand, and the Renewable Revolution

In this episode, Matt and Georgia sit down with Brad Young (Capgemini Invent) and Alistair Adams (Solution Energy) for a fast-moving conversation about AI’s exploding energy appetite and what it means for the future of data centers, power grids, and sustainability. From geopolitical tension to geothermal innovation, this one covers the full energy spectrum. What We Covered: - AI’s Energy Crunch AI growth is driving unprecedented demand for power. Hyperscalers like Meta, Google, and Microsoft are signing multi-billion-dollar infrastructure contracts at record pace, stretching grids and reshaping global infrastructure priorities. - The Rise of “Power-First” Google’s “power-first strategy” shows the new reality: build data centers where the power is, not where the people are. Nvidia’s Jensen Huang agrees—co-locating at generation sites may be the future. Reliable, renewable baseload power is now the real competitive edge. - Water: The Silent Crisis Energy gets the headlines, but water is just as critical. Google already uses ~70 billion litres annually for cooling—on track to rise tenfold. Innovations like geothermal heat rejection (e.g., the Pawsey supercomputer in WA) offer promising alternatives. - Renewables: What Actually Works Not all green energy is created equal. Wind and solar can’t deliver the 24/7 baseload those massive GPU clusters require. That leaves geothermal and nuclear as the only scalable clean options—though nuclear remains politically fraught in markets like Australia. Regional Realities - Australia: Victoria faces a looming 1.5 GW gap with coal retirement. - UK: Grid constraints limit data center growth. - US: Federal policy is leaning hard into nuclear and geothermal for AI. - Europe: Regulation is reshaping the tech landscape—for better or worse. Cloud’s Hidden ESG Problem Most cloud usage sits in companies’ Scope 3 emissions. As ESG rules tighten, lack of transparency from hyperscalers becomes a real compliance exposure. - Social License Becomes Strategy Community pushback is halting billion-dollar projects. The new game: secure energy, protect water, and bring the community with you. “Permission-based infrastructure” is quickly becoming the norm. - AI, Talent & the Enterprise Gap We discuss the widening skills challenge—junior staff struggle to validate AI outputs, and enterprises claiming “we don’t have use cases” are already falling behind. - Greener Compute Through Smart Pricing Dynamic cloud pricing tied to renewable availability is on the horizon—think “off-peak compute,” automatically routing workloads to greener grids. Standout Insights - We’re in the “Nokia 3210 era” of AI—25+ years of disruption ahead. - Robotics is still more marketing than reality. - Enterprise AI adoption is early; the real environmental impact is still to come. Key Takeaways - Data center location will follow energy, not geography. - Community permission is as critical as capital. - Water use must be part of every sustainability conversation. - Geothermal and nuclear are the only viable clean baseload options. - The next decade will be messy as demand outpaces grid upgrades. - Hyperscalers are accelerating renewable markets—out of necessity. - ESG exposure from opaque cloud emissions is rising fast. Conclusion AI’s growth is forcing a complete rethink of how we power digital infrastructure. The winners will be those who can solve the combined puzzle of clean energy, water management, community trust, and transparent reporting—at a speed the grid has never been asked to move before.