Platform Engineering Insights

💫2026: How GenAI’s Role Changes in Insurance Simple efficiency gains are done. The basic productivity phase of GenAI is largely complete. What comes next is not smarter AI. It is exposed operations. ⸻ 🔍 What GenAI really surfaces in 2026 As GenAI moves beyond basic automation, it begins to reveal operational gaps insurers have lived with for years: 　⚠️ processes that work only because people fill in the blanks 　⚠️ decisions based on tacit understanding rather than explicit rules 　⚠️ operations that can be explained, but not consistently repeated This is not a technology issue. It is an operational reality check. ⸻ ⚙️ Why this matters now In 2026, GenAI removes the human buffer that used to absorb ambiguity. What was once: • “handled case by case” • “managed by experienced staff” • “good enough in practice” becomes visible, inconsistent, and hard to scale. ⸻ 🎯 Executive takeaway GenAI does not change insurance operations. It reveals which operations were never fully defined in the first place. GenAI is no longer an efficiency tool. It is an operational stress test. #GenAI #Insurance

💡 "AI might grab the spotlight… but it’s the data and platform engineers who plug in the lights." I see everyone's talking about AI like it just happens. Spoiler: it doesn't. Behind every smooth prediction and clever recommendation are two groups of people making sure things actually work. 𝗗𝗮𝘁𝗮 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝘀: Your AI's Reality Check AI models need data. Good data. Not just any data dumped into a folder somewhere. What they're actually doing: → Building pipelines that pull data from ten different sources and make them play nice together → Cleaning up the mess so your model isn't learning from typos and duplicates → Setting up quality checks because one corrupted field can tank your entire output → Creating systems for data storage, transformation, and access that don't fall apart at scale They're not building the AI. They're building the foundation it stands on. Without them, your fancy model is just expensive guesswork running on bad information. 𝗣𝗹𝗮𝘁𝗳𝗼𝗿𝗺 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝘀: The Reason It Doesn't Crash You trained a model. Great. Now how do you actually deploy it without everything catching fire? Enter platform engineers: → Building infrastructure that handles real traffic, not just your test dataset → Managing deployments so updates don't break production at 2am → Monitoring systems to catch issues before your users do → Scaling resources so your app doesn't choke when it actually gets popular → Dealing with security, networking, and compliance (the stuff nobody wants to think about) They make sure your #AI goes from "works on my laptop" to "works for a million people." So, while the Data Scientist is teaching the model how to shine, the Data Engineer is wiring up the lights so they actually turn on, and the Platform Engineer is powering the grid to keep them glowing steadily. Data Engineers holding everything together again 💪🙂 Agree?

𝐄𝐯𝐞𝐫𝐲𝐨𝐧𝐞 𝐰𝐚𝐧𝐭𝐬 𝐀𝐈. 𝐕𝐞𝐫𝐲 𝐟𝐞𝐰 𝐢𝐧𝐬𝐮𝐫𝐞𝐫𝐬 𝐚𝐫𝐞 𝐩𝐫𝐞𝐩𝐚𝐫𝐞𝐝 𝐟𝐨𝐫 𝐰𝐡𝐚𝐭 𝐀𝐈 𝐚𝐜𝐭𝐮𝐚𝐥𝐥𝐲 𝐞𝐱𝐩𝐨𝐬𝐞𝐬. I see this pattern everywhere right now. Insurance leaders rush to deploy AI models for underwriting, claims, or pricing. The models perform well in testing. Everyone celebrates the innovation. Then reality hits. AI doesn't hide data problems. It amplifies them. That clean dataset you thought was ready? AI finds the gaps instantly. Those manual overrides your team made for years? AI reveals the inconsistencies. That tribal knowledge sitting in someone's head? AI exposes how much you've been depending on it. Here's what most insurers miss: AI implementation isn't a technology project. It's an organisational mirror. When AI starts making recommendations, it forces uncomfortable questions: • Why do we have three different definitions for the same risk factor? • Why does our data quality drop after the first renewal? • Why can't we explain this pricing exception from 2019? • Why do different teams use completely different assumptions? These questions existed before AI. We just didn't have to answer them. The insurers winning with AI in 2026 aren't the ones with the fanciest models. They're the ones willing to fix what AI reveals. They treat AI deployment as a forcing function for organisational clarity. Before launching the next AI initiative, ask yourself: • Are we ready to face what our data actually looks like? • Can we handle the transparency AI will create? • Do we have the discipline to fix foundational issues before scaling? AI won't transform your business if your business isn't ready to transform itself first. What's the hardest truth AI has revealed in your organisation? #AIinInsurance #InsuranceLeadership #InsurTech #DigitalTransformation #DataStrategy

After 20 years in insurance operations, I'm seeing a fundamental shift that most carriers are missing. The old playbook for operations was simple: offshore repetitive tasks, optimize cost-per-FTE, measure efficiency in headcount reduction. That playbook is dead. The new reality: Modern insurance operations is about identifying where to apply #AI and #automation to shift from linear to non-linear delivery models. The winners aren't competing on labor costs — they're competing on which use cases actually move the needle. Three things I'm seeing in the market: 1. #Gen AI and #Agentic AI are moving into production — selectively The best outcomes aren't "AI everywhere." They're targeted deployments in underwriting exceptions, claims triage, and policy admin workflows where AI handles volume and humans handle complexity. Companies trying to automate everything are failing. 2. Vendor AI solutions promise near-perfect accuracy. Production reality is 60-80% on average. Every vendor demo shows flawless outcomes. Then you deploy and accuracy drops because real insurance data is inconsistent, incomplete, and full of edge cases the model never saw in training. Carriers struggle to evaluate which solutions actually work vs. which just performed well on sanitized demo data. The gap isn't the technology — it's understanding your specific data quality and process reality. 3. AI companies don't factor in domain and process nuances Tech firms building AI for insurance treat underwriting, claims, and policy admin as generic document processing problems. They're not. Each has decades of business rules, regulatory requirements, and process exceptions that AI models trained on generic data completely miss. The companies winning are those that combine AI capabilities with deep insurance domain expertise. The carriers figuring this out are seeing 40%+ efficiency improvements while improving customer experience. The ones stuck in 2015 thinking are bleeding market share. What am I missing? If you're operating in insurance or building technology for insurance, what's the reality gap between vendor promises and production results? #InsuranceTechnology #AIinInsurance #InsuranceOperations #GenAI

𝗣𝗹𝗮𝘁𝗳𝗼𝗿𝗺 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝗶𝘀 𝗻𝗼𝘁 𝗮 𝘀𝗶𝗻𝗴𝗹𝗲 𝗷𝗼𝗯 𝘁𝗶𝘁𝗹𝗲! 𝗜𝘁'𝘀 𝗮 𝘁𝗲𝗮𝗺 𝗻𝗮𝗺𝗲 𝗮𝗯𝗯𝗿𝗲𝘃𝗶𝗮𝘁𝗲𝗱 𝗳𝗿𝗼𝗺 𝗮 𝗰𝘂𝗹𝘁𝘂𝗿𝗲 𝗮𝗻𝗱 𝗱𝗶𝘀𝗰𝗶𝗽𝗹𝗶𝗻𝗲. Just like DevOps became "DevOps Engineer," we're making the same mistake again. When companies post "Platform Engineer - Must have expertise in Kubernetes, Terraform, CI/CD, observability, security, networking, and developer experience," they're asking for a mythical person who doesn't exist. Just like they did with DevOps. Platform Engineering isn't a single role. It's a team discipline that requires multiple specialized positions working together: 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝘀 who build internal tools, APIs, and developer-facing interfaces. They understand product development, user experience, and software architecture. 𝗦𝘆𝘀𝘁𝗲𝗺𝘀 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝘀 who design infrastructure, manage cloud resources, and ensure platform reliability. They understand distributed systems, security, and operational concerns. 𝗠𝗮𝘆𝗯𝗲 𝗮 𝗣𝗿𝗼𝗱𝘂𝗰𝘁 𝗠𝗮𝗻𝗮𝗴𝗲𝗿 who understands developer needs and can translate business requirements into platform capabilities. --- I was hired as a Platform Engineer and I was overburdened with requests from multiple directions. Frontend teams wanted better deployment pipelines, backend teams needed database automation, security wanted compliance tooling, and management wanted cost optimization dashboards. One person cannot effectively handle all these competing priorities while maintaining quality and sanity. Successful platform engineering requires specialization, not generalization. You need people who can go deep in their respective areas while collaborating effectively as a team. We learned this lesson with #DevOps. Let's not repeat the same mistake with #Platform Engineering.

We just released the State of Platform Engineering Vol. 4, and it’s easily the most practical report we’ve published so far. This isn’t a “platform engineering is important” piece. It’s a snapshot of where teams actually are right now. What they’re measuring. Where adoption breaks. How security, AI, FinOps, and observability are colliding with platform work in very real, very messy ways. Some highlights that stood out to me: A surprising number of teams still don’t measure success at all. Adoption is still more push than pull in many orgs. And while almost everyone talks about platforms as products, only a fraction have the structures and incentives to support that claim. The gap between intention and execution is still huge. What’s different this year is how clearly platform engineering is bleeding into adjacent domains. Data, security, AI, cost, reliability. The report shows that the teams making progress aren’t treating these as separate initiatives anymore. They’re converging them through the platform layer. And AI? It didn’t make platform engineering obsolete. It made the cracks impossible to ignore. Teams trying to scale AI without strong platform foundations are feeling it fast. If you’re leading a platform team, building an IDP, or trying to justify the next phase of your initiative, this report gives you data points, language, and benchmarks you can actually use. Not theory. Not hype. Worth a read if you want to sanity check where you stand and where the industry is really heading. https://lnkd.in/eCh4HV9e

Layers in your architecture get owned. Connective tissue gets starved. That's the entire reason most platform initiatives fail, and it has nothing to do with the technology. When you draw your platform as a layer, the team closest to it claims it, and the teams above and below treat it as someone else's problem. When you draw it as the connective middle, nobody claims it, and it gets starved of headcount, funding, and political cover. Both diagrams kill the platform. Just on different timelines. The fix isn't a better picture. It's an operating model where the platform team owns the contract with every layer it touches, and the surrounding teams are measured on how well they consume and feed it. If your platform team is on uptime SLAs and your app teams are on feature velocity, the platform loses every prioritization fight until it dies quietly. I've watched this pattern play out across four companies now. The platforms that survived weren't the ones with the cleanest reference architecture. They were the ones where leadership made a deliberate choice: reshape the architecture to fit the org, reshape the org to fit the target architecture, or do both at once. The default is to do neither, and let the org's gravity quietly bend the architecture into whatever shape the existing reporting lines allow. That's how platforms die. Not from bad design, from unmade decisions. What's the deliberate choice you've seen work? #PlatformEngineering #EnterpriseArchitecture #TechnologyLeadership

Does AI Kill Platform Engineering? AI is disrupting almost every layer of software. Code, testing, security, support, product management. It is reshaping how systems are built and operated. So it is fair to ask what it means for platform engineering. Two questions keep coming up in conversations with enterprise leaders, platform teams and investors: 1. If AI can operate infrastructure, why do we need platform engineering at all? 2. As AI infrastructure becomes dominant, do cloud-era platforms still matter? Let’s start with the first. The original case for platform engineering was productivity. Self-service. Golden paths. Reducing cognitive load. But if AI becomes the interface, that argument weakens. So what’s left? Control. Enterprises do not optimize purely for capability. They optimize for accountability. Someone still owns the cloud bill, the compliance audit, data residency, security posture, and the blast radius of failure. An AI agent can provision infrastructure. It cannot assume responsibility. As AI increases velocity, governance becomes more important, not less.And this is where declarative (intent based) APIs matter. Agents need structured, stable, idempotent interfaces. They need to declare intent, not execute fragile imperative steps. They need policy enforcement and reconciliation built in. Platform engineering becomes less about productivity tooling for humans and more about defining the declarative control plane that agents operate against. Now the second question. AI workloads introduce GPUs, accelerators, model registries, inference endpoints. But underneath, it is still compute, networking, storage, identity, policy, and cost. The workload changes. The hardware shifts. The need for a governed substrate does not. If anything, AI increases heterogeneity, cost volatility, and regulatory scrutiny. What I’m seeing in Fortune 500 companies: Platform teams are not shrinking. They are being asked to support traditional workloads plus AI infrastructure, across more clouds, at higher velocity, under stricter compliance. The scope is expanding. The real debate isn’t whether AI kills platform engineering. It’s whether enterprises still want sovereignty and policy control over infrastructure in an AI-driven world. From what I’m seeing, they clearly do. Curious what others are experiencing. Is AI shrinking your platform scope, or redefining it? #PlatformEngineering #AIInfrastructure #CloudNative #Crossplane #EnterpriseIT

Something I see often in #bioinformatics - across #RandD, services and product teams - is the assumption that pipelines and platforms are basically the same thing. They are not, and the gap between them is where most complexity and technical debt begin. #Pipelines are execution-focused: a defined sequence of steps that turn an input into an output. They are narrow in scope, assume context and are often tightly coupled to a dataset or team. #Platforms are system-focused: everything that surrounds those pipelines to make them reliable, reproducible, maintainable and scalable. They provide the environment, guarantees and lifecycle that allow many pipelines to coexist without breaking each other. Most people think they understand this distinction, yet in practice it is one of the most misunderstood areas in our field. **Key differences** - Pipelines optimise for output. Platforms optimise for consistency and sustainability. - Pipelines bind to assumptions. Platforms formalise assumptions with clear interfaces and contracts. - Pipelines are hard to migrate. Platforms make migration possible through standards and abstraction. - Pipelines deliver a result. Platforms deliver a foundation. **A few examples of platform-level architecture** - Designing versioning that guarantees backwards compatibility across teams. - Defining interface layers so pipelines behave predictably, no matter who built them. - Introducing orchestration with automation, dynamic retries, audit logs, monitoring and resource control. - Building environment models (containerisation, dependency governance, IaC) for true portability. - Creating standards so new pipelines plug in without reinventing the structure. These are not "pipeline tasks" - they are architectural decisions. If everything is treated like a pipeline, you end up with dozens of incompatible, brittle workflows. Platform thinking makes the whole ecosystem clearer, safer and easier to scale. For anyone working in bioinformatics #engineering, recognising whether you are building a pipeline or a platform is one of the most important mindset shifts you can make.