Build an AI Chief of Staff for Your Team

A client asked me this week how many of his 14 AI pilots were worth keeping. Stanford answered before I could. Statistically: one. My top 3 AI news from last week: PwC's AI Performance Study: 75% of AI's economic gains go to the top 20% of companies. The leaders aren't winning on tools. They're winning on governance, frameworks, and the unsexy infrastructure that moves pilots to production. If your AI conversation is still about "saving hours," your competitors are compounding. Stanford's AI Index 2026: 88% of organizations now use AI. Documented incidents jumped from 233 to 362. And 95% of GenAI pilots never reach production. Adoption is climbing vertically. Production isn't moving. Incidents are up 55%. The risk isn't falling behind on adoption. It's the distance between pilot and production. Anthropic is facing backlash over Claude's performance decline (Fortune, this week). Four outages in March. Regressions users can feel. Silence on what changed. You're running critical workflows on a model you can't inspect and a vendor who won't explain what changed. Multi-model architecture is no longer a nice-to-have. It's insurance. Adoption is not the moat. Governance is. The companies winning in 2026 figured that out in 2024.

When AI sends you this message, your money is GONE. Real money I mean. Cost of tokens. You are still playing with AI if you have not seen this message yet. This is not just an ai response nor is it a friendly reminder. Here is the message: "The scope is too large to do serially. I'll run parallel agents to convert each section simultaneously. 4 background agents launched" This is a signal that: - you are not just testing how to build with AI - you are not building simple pages or website - you have the technical engineering depth - you are building an enterprise solution, not demo apps - the complexity of your application requires additional resource allocation Ultimately, it is a signal that you are about to empty your account in a matter of minutes due to the cost of AI tokens required. Heavy resources or coding require multiple AI agents to run simultaneously. While many people are building for the insurance industry, very few are building for reinsurance. After 15 years of building for the insurance, I am currently building for the reinsurance. This time, no more manual coding. All codes written by AI. (very few can get this done) If you are building or driving any enterprise AI solution, let's connect. Happy Monday. #ai #software #engineering #reinsurance #enterprise #innovation

Thousands of atomic APIs. One composition layer. One AI access layer.   That's what it takes to make an insurance platform AI-coding-ready. Not a wrapper on a core system. Not a chatbot. A redesign of what sits between your system of record and your developers.   Most insurance APIs were built for humans to integrate. Big, monolithic, one endpoint does ten things. AI agents can't compose them. Developers spend weeks on plumbing that should take hours.   We rebuilt the API layer from the atom up.   Thousands of atomic APIs, each doing one specific thing. A composition layer (iComposer) that turns them into business workflows. An AI access layer (MCP, Skills, CLI) so AI coding tools can discover and call them natively.   If you're rebuilding your stack for the AI era, drop your line of business in the comments and I'll share what the atomic API surface looks like for it. Motor, home, life, commercial, reinsurance, whichever one you're working on. #Insurance #InsuranceAI #InsuranceAustralia #Insurer #InsurTech #VibeCodingInsurance #Brokers #AI #MGA #Insurancedistribution

Last week, I met someone from a Fortune 1000 bank in SF. I went into the meeting with one assumption: The biggest bottleneck in AI creative is quality. - Can the model generate something on-brand? - Can it follow the brief? - Can it produce something good enough to test? I came out realizing I was only partly right. One thing that surprised me was that they had a separate platform just for reviewing and approving creative assets. That made something very clear: Only solving for generation will not cut it for large enterprises. In paid media, everyone wants more creative diversity. More hooks. More angles. More formats. More tests. AI can now help generate a lot of that. But if you are a large bank, insurance company, healthcare company, or any regulated business, creating the asset is only step one. Then it has to move through brand, product, legal, compliance, and media teams. Now multiply that by: - 25 creatives from each vendor - 10+ product lines - multiple geographies - different rules for every market - several rounds of feedback before anything goes live That is where the leverage breaks. You can generate 100 creatives in a week. But if review and approval still happen in a separate workflow, with separate context, the speed advantage disappears. That meeting changed how I think about AI adoption in large companies. It will not be won only by better models. It will be won by systems that bring analysis, generation, review, approval, and launch into one command center. That is the part we have become obsessed with at GetCrux. Not just generating more creatives. Helping large teams actually get them live. End to end. GetCrux (YC W24)

Your claims AI is working fine. The pipeline underneath it? That's another story. Queues age. Workflows stall. Throughput drops. And none of it triggers an alarm — because the system is technically "up." Uptime ≠ throughput. Silence is expensive. 🎲 We broke down why claims teams keep missing SLAs even after going all-in on AI 👇 🔗 https://lnkd.in/emjHzypM Ryan Shallenberger #ClaimsProcessing #Insurance #AIOperations #IDP #Insurtech

𝐀𝐈 𝐢𝐧 𝐢𝐧𝐬𝐮𝐫𝐚𝐧𝐜𝐞 𝐢𝐬 𝐦𝐨𝐯𝐢𝐧𝐠 𝐟𝐚𝐬𝐭𝐞𝐫 𝐭𝐡𝐚𝐧 𝐦𝐨𝐬𝐭 𝐩𝐞𝐨𝐩𝐥𝐞 𝐫𝐞𝐚𝐥𝐢𝐬𝐞 👇🏽 Everyone's talking about AI replacing underwriters or claims handlers. But the real transformation happening right now is quieter — and frankly more interesting. 𝐈𝐭'𝐬 𝐡𝐚𝐩𝐩𝐞𝐧𝐢𝐧𝐠 𝐢𝐧 𝐭𝐡𝐞 𝐝𝐚𝐭𝐚 𝐥𝐚𝐲𝐞𝐫: → Personalisation models helping insurers shift from reactive to proactive customer engagement → Automated pipelines becoming standard practice as customer behaviour shifts faster than ever → Analytics teams evolving from reporting what happened to predicting what's next The insurers winning aren't the ones with the flashiest AI. They're the ones who've built a solid data foundation to make AI actually work. And in a heavily regulated industry like insurance — where privacy and customer trust are non-negotiable — getting that foundation right matters more than chasing the latest model. Are you working in insurance or financial services? What AI use case are you most excited or cautious about? 👇🏽 #Insurance #AI #DataScience #Insurtech #Analytics

AI is...everywhere.  It’s in every headline. Every boardroom. Every roadmap.  It feels a bit like a modern-day gold rush, with organizations racing to stake their claim and prove value, fast.  Within the world of insurance, the urgency isn’t misplaced. Opportunity is very real, and for many, already materializing.  According to McKinsey & Company, leading insurers are already seeing ~6% higher shareholder returns from AI. However, most remain in the experimentation phase, with few having scaled AI to deliver meaningful business impact.  As our knowledge of enterprise AI expands, it’s becoming clear that gaps exist not because of access but because of application. Most organizations are still testing use cases in isolation. Where organizations are seeing meaningful impact is when a more integrated approach is taken, embedding AI into core workflows across the business.  For those further along in the AI journey, results show up in very tangible ways: ▪️ 10-20% improvement in sales conversion rates  ▪️10-15% increase in premium growth   ▪️20-40% reduction in cost to onboard new customers   ▪️3-5% improvement in claims accuracy  So, why doesn’t every insurance carrier and agency see results?     It’s not from lack of investment. And it’s not from a lack of tools. More often than not, it’s the foundation.  Put simply, AI doesn’t operate in a vacuum. It depends heavily on systems that talk to each other, data that is consistent and reliable, and teams that trust what’s being surfaced.  When data is fragmented, definitions don’t align, or workflows break across systems, AI doesn’t fix the problem...it exposes it. And even when insurers have the right technology in place, adoption rarely happens without clear governance and effective change management in place.     What it ultimately boils down to is the data. How consistent it is, how connected it is, and whether it can actually support day-to-day decisions.   Because at the end of the day, AI isn’t the advantage on its own...what sits underneath it is.  Check out the full McKinsey & Company article here 👉 https://lnkd.in/eSNgPjHV  

I've been in enough AI kickoff meetings to know what kills momentum before the project even starts. The Same goes for Football, Insurance, Content Creation...You Have to speak in their language and TRANSLATE THE TERMS where they can easily understand. But We're Talking AI Today. Someone drops the acronyms. LLM. API. MCP. RAG. Half the room goes quiet. Not because they're not smart — because nobody handed them a translation. So I built one. Swipe through this. I broke down the full AI stack using the language your team already speaks: The LLM is the engine. It runs. It doesn't know your business. Data is the fuel. Bad fuel blows the engine. Doesn't matter how good the model is. Your AI is only as good as your last dataset (Remember, you're only as good as your last envelope, or your last game, or your last sale?) The API is the USB-C port. It's how things plug in. That's it. MCP is the nervous system. Every tool, every source, every signal — flowing to one brain so it can respond to the whole body. A plumber would call this the plumbing system. Agents are the crew. They take the objective and run. They don't wait to be told every move. You're the GC on the job sight! Governance is the control panel. Every dial, every limit, every warning light. You don't want to drive a car without a steering wheel! The technology isn't the hard part anymore. Getting your team to stop nodding and start understanding — that's the work. People are the toughest part when it comes to deploying AI and Agentic systems within your workforce. The Tech Stack is actually the easy part! What's the analogy that finally clicked for someone on your team? Drop it below. I'm building the full library. ↓ Swipe through the full breakdown

Insurance isn’t being disrupted at the surface — it’s being rebuilt behind the scenes. Here’s how AI is quietly transforming insurance organizations 👇 • 📄 Automated Underwriting – Faster risk assessment using historical + behavioral data • 🔍 Fraud Detection – Real-time anomaly detection across claims patterns • ⚡ Claims Processing – Instant document parsing, damage assessment, payout decisions • 🎯 Risk Pricing & Segmentation – Dynamic, personalized premium modeling • 🧠 Predictive Analytics – Anticipating churn, claim probability, and customer lifetime value • 📊 Operational Efficiency – Automating back-office workflows and reducing manual errors • 🗂️ Document Intelligence – Extracting insights from unstructured policy and claim documents • 🤖 Agent Assist Tools – AI copilots for faster decision-making and compliance checks The real AI revolution in insurance isn’t flashy. It’s silent, scalable, and deeply embedded in decision-making. #AI #Insurance #DataScience #MachineLearning #DigitalTransformation

POV: My executive AI assistant cost me $237 in 8 hours, and I had no idea. Okay, so apparently, when you run Claude Code AND a custom AI agent on the same OAuth token? They share a billing pool. Both can hit your Extra Usage. Both can blow past your "budget" simultaneously while you're sneaking in a round of Marvel Rivals during lunch, thinking you're only being irresponsible by being too busy chasing Jeff to notice the Black Panther in your backline. Spoiler: I was way more irresponsible. Now I've got: → A dedicated API key for my agent ($25/month hard cap, no override) → Extra Usage disabled on my Claude.ai subscription → Cost dashboards I actually look at My weekly cost now: €4-8. Down from €237/weekend. The lesson? OAuth shortcuts are convenient until they're a horror story you tell at 2 am. Use API keys with caps for anything autonomous. What's a billing surprise you've hit with AI tools? Please, please, please tell me I'm not alone. #BuildInPublic #AIAgents #IndieHacker #ClaudeAI #LLMOps