Jean-Luc Nahon

This Scale AI story might matter more than you think. Meta just invested $14.3B to take a 49% stake in Scale AI — and brought Scale’s CEO into its internal “superintelligence” unit. On the surface, it looks like a major bet on data. But look a little deeper, and it’s a shift in how control is consolidating across the AI value chain. Why it matters: – Scale was one of the few trusted, neutral data providers for labeling and evaluation – Now that neutrality is in question, contracts are reportedly getting pulled by OpenAI and Google – Startups and enterprises alike are suddenly reconsidering where their training signal comes from What comes next: – A surge of boutique data providers — decentralized, domain-specific, and more transparent – Startups reframing data control as part of their moat: “We don’t just use models — we own the context” – Infra teams re-evaluating dependencies — either proactively, or reactively when vendor dynamics shift As training data usage becomes more regulated and controlled — moving away from the wild west of early foundation model training — the pressure to secure differentiated data pipelines only increases. When we talk about defensibility in AI, it’s tempting to focus on model IP or feature differentiation. But data leverage — how it’s sourced, labeled, evaluated, and governed — is becoming a more important part of the strategic equation. It’s not quite infrastructure. But it’s no longer a commodity either. It’s a layer of increasing consequence — and one worth watching closely. Are you rethinking your data leverage yet? https://lnkd.in/gx9GPefp

11

1 Comment

GPUs sit idle 95% of the time ‼️ Not an estimate. Measured across tens of thousands of production environments, before any optimization is applied. Cast AI just released its 2026 State of Kubernetes Optimization Report, built to show engineering and infrastructure leaders where the inefficiencies actually sit and how to close them. https://lnkd.in/gXz-dyR4

20

We just shipped multi-node scaling for Amazon Timestream for InfluxDB 3 — clusters can now scale up to 15 nodes with configurable writer/reader and reader- only roles. This was a fun one to work on. The idea is simple: you shouldn't have to re-architect your setup or shard your data across multiple clusters when your workload grows. Create a parameter group with your desired node configuration, apply it, and your cluster scales — traffic get's distributed automatically among nodes on different Availability Zones improving not only capacity but also availability. If you're running time-series workloads on Amazon Timestream for InfluxDB and need to scale reads and writes independently, this is worth a look. https://lnkd.in/gJXedrSQ #AWS #AmazonTimestream #InfluxDB #TimeSeries

70

2 Comments

In the age of AI, MCP is the new API! Go read up on what it is, why you need to know it, and how you can get started.

18

Claude Code + YouTube API = your own personal video playlist curator YouTube's algorithm is good, but it's partially driven by ad revenue and what it thinks will keep me on site. I spent <1 hour last night with Claude Code to automate the perfect playlist for me: 1️⃣ Every morning, I'll wake up to 25 fresh videos tailored to my interests: - 15 videos from my favourite channels (e.g., No Priors with Sarah Guo or 20VC with Harry Stebbings) - 10 videos on topics I'm interested in (right now, mainly AI and tennis) - Smart duration filtering: no videos under 60 seconds, at least 5 medium-length (10-19 min), at least 5 deep dives (20+ min) - Zero duplicates, zero content I've already watched 2️⃣ RUNS AUTOMATICALLY IN THE CLOUD: Using GitHub Actions, this now updates every morning - no computer or prompting needed, completely hands-free and automated 3️⃣ MORE FOCUS, LESS DOOM SCROLLING: Until last night, I'd hit YouTube and get sucked into content I hadn't planned to watch. Now, in theory at least, I'll go straight to my "Latest & Greatest" playlist and stay more on point. Looking forward, now that I've connected Anthropic's Claude Code, YouTube, and GitHub, I'm excited to build more custom playlists + better understand how much I'm actually watching and on what topics - maybe it'll be all Rafael Nadal highlights and no AI?? ;) HOW TO DO THIS: 1. Install Claude Code (works best locally; plenty of YouTube tutorials to help) 2. Ask Claude Code to walk you through setting up Google Cloud Console, enabling YouTube Data API v3, and creating OAuth credentials 3. Describe what content you want in plain English 4. Connect to GitHub - Claude Code will guide you - to automate the flow daily/weekly My Claude Code fun with YouTube last night and Spotify last weekend highlights the power of AI to help us personalize our experiences and reclaim our attention. BigTech's algorithms are great, but why not tailor our content to exactly what we want? Finally, I'd LOVE it if we could more precisely tailor our LinkedIn feeds versus needing to mute people and tune our notifications one connection at a time. I may know enough about LinkedIn - including its API policies and ad business - to know that'll never happen, but a boy can always dream ;) Excited for more exploration with Claude Code this weekend!

63

18 Comments

I'm no AI expert. I have no idea which model is best for what. That's exactly why Auto-model modes (just released) is a game-changer for me. I used to over think about picking the wrong tool or burning through credits on overkill. Now? I just ask for what I need, and Hatz routes it to the right model automatically. Speed, accuracy, cost, handled. I work faster. I work smarter. I don't have to think like a data scientist to get expert results. This is what AI should feel like. Simple. Smart. Built for normal people in business, like me. :) Check it out.

12

Awesome to see Wix in the Google I/O announcement about, Gemini Spark - their new flagship AI agent 🙌 Alongside names like Adobe, Salesforce, HubSpot and Uber, Wix is joining the MCP ecosystem that powers Gemini Spark. That means Gemini will be able to take real action on behalf of users - directly through Wix. Exciting times ahead. 🚀 https://lnkd.in/ecuhYb_H #Wix #GeminiSpark #GoogleIO #AI #Agentic

146

2 Comments

"The intelligence of Yahoo Scout is anchored in internal search technology forged from three decades of operating one of the largest search products on the internet. By drawing from Yahoo’s established query understanding system, content ranking system, and a proprietary knowledge graph, Yahoo Scout can both identify the specific intent behind a user’s question and then organize the information to answer that question in an order that’s easy to understand." https://lnkd.in/eVhk298t

11

We're continuing to expand our VM portfolio to give customers even more powerful options! We're excited to announce general availability of expanded shapes and sizes for our local storage optimized Z3 instances. These instances are engineered for I/O-intensive workloads requiring significant local storage and high performance, making them ideal for a wide range of use cases including SQL, NoSQL, and vector databases, data analytics, semantic data search and retrieval, and distributed file systems. https://lnkd.in/gKyMCHBX Salil Suri Mark Lohmeyer robert napaaPankaj G. David Dunn Alex Pinkin Saad Syed

35

MIT’s recent report found that 95% of enterprise AI pilots fail to deliver ROI, with value trapped in workflows, slow scaling, and misplaced bets in sales and marketing instead of back-office and integration wins. But these are symptoms, not the root cause. The real issue is strategic misalignment, as oftentimes organizations don’t clearly define the kind of bet they’re making. At the core, every pilot is a capital allocation and risk management decision. Pilots fail without clarity on the business angle, hypotheses, and go/no-go criteria. To address this, I’ve developed The Pilot Strategy Framework™, a structured approach to categorizing AI pilots based on their strategic intent and business impact. 1️⃣ Validation pilots - validate feasibility before scaling and committing further resources. 2️⃣ Internal adoption pilots - phased rollout of AI within departments, inside existing workflows, to see if it can unlock ROI. This can often signal market readiness to customers, competitors, investors, and the industry overall. 3️⃣ Customer-facing pilots - test whether existing customers will adopt new AI capabilities and whether it’s commercially safe to scale. 4️⃣ Legacy integration pilots - answer if AI can unlock value in legacy systems. 5️⃣ New product pilots - test whether additional AI offerings can be introduced; offerings that are core or adjacent to our current products and generate new ROI. Each pilot is a bet: upfront decide how much capital to risk, what value it’s meant to unlock, tradeoffs to consider, and the conditions under which you’ll scale or stop. Ultimately, it's about prioritizing the one(s) that provide the most enterprise value. Which pilot types is your organization running, and how are they moving the needle on your business outcomes? I am currently talking to leadership teams to apply this framework to drive clarity and ROI; if your organization is navigating similar questions, I'd be happy to connect. #productstrategy #aistrategy #aitransformation

4

1 Comment

Open Source LLMs are all the rage, but how do you know which is best? That's why I put together this concise, curated collection of leaderboards and lists for open source LLMs: Hugging Face Leaderboard by OpenEvals https://lnkd.in/g_SEVtJ4 LLMs - Evaluation / Leaderboards / Comparisons by mh-ka https://lnkd.in/gqeTvtTp Dubesor LLM Benchmark Table https://lnkd.in/gWiryruq WebDev Arena LLM Leaderboard https://lnkd.in/gjCPNWxb

1

This recent a16z article by Jason Cui and Jennifer Li on "Context Layers" affirms our company mission: https://lnkd.in/gURRJ9aa It's a great read. Anyone trying to deploy agents in the enterprise needs to read this. 𝗧𝗵𝗲 𝗦𝗵𝗶𝗳𝘁: 𝗧𝗲𝗰𝗵𝗻𝗶𝗰𝗮𝗹 𝗠𝗲𝘁𝗮𝗱𝗮𝘁𝗮 𝘁𝗼 𝗕𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗥𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 The primary bottleneck for Enterprise AI isn’t model capability - it’s the missing architecture between raw data and agentic action. The "Context Layer" is the new competitive primitive. It sits between raw data and LLMs to provide the business logic, definitions, and "tribal knowledge" required for autonomy. It translates technical schemas into a format agents actually understand, ensuring they use authoritative metrics for reasoning. Traditional semantic layers from the BI era were passive abstractions designed for human-centric dashboards. A modern context layer must be agentic. It must be tool-aware, multimodal, and capable of self-correction. 𝗔 𝗰𝗼𝗻𝘁𝗲𝘅𝘁 𝗹𝗮𝘆𝗲𝗿 𝘁𝗵𝗮𝘁 𝗱𝗼𝗲𝘀 𝗺𝗼𝗿𝗲 𝘄𝗶𝘁𝗵 𝗹𝗲𝘀𝘀 This is exactly what we are solving for at App Orchid Inc. While the market discusses the modern context layer as a concept, we have built the industrial-grade implementation with some added benefits beyond what a16z describes. ◆ 𝗦𝗲𝗹𝗳-𝗕𝗼𝗼𝘁𝘀𝘁𝗿𝗮𝗽𝗽𝗶𝗻𝗴 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗚𝗿𝗮𝗽𝗵: App Orchid will automatically identify business objects and semantics, generating descriptions and mapping structural relationships in a knowledge graph. This context graph can enrich itself with use. ◆ 𝗧𝗿𝗮𝗶𝘁-𝗕𝗮𝘀𝗲𝗱 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲 𝗳𝗼𝗿 𝗮𝗴𝗲𝗻𝘁𝘀: We encode "Traits" (e.g. geospatial, time-series) that make agents answer common business questions with better fidelity. ◆ 𝗦𝗲𝗺𝗮𝗻𝘁𝗶𝗰 𝗣𝗼𝗿𝘁𝗮𝗯𝗶𝗹𝗶𝘁𝘆: By decoupling business logic from the storage or presentation layer, your AI doesn't lose its context if you migrate your data warehouse or switch BI tools. ◆ 𝗚𝗼𝘃𝗲𝗿𝗻𝗮𝗻𝗰𝗲: We encode access and track interactions with data for transparency, auditability and governance. Read more about how we do it here: https://lnkd.in/gwFVcEfv 𝙎𝙤𝙢𝙚 𝙌𝙪𝙚𝙨𝙩𝙞𝙤𝙣𝙨 𝙛𝙤𝙧 𝙩𝙝𝙚 𝘾-𝙎𝙪𝙞𝙩𝙚 • Does your AI understand your "business vernacular" or just your SQL schema? • Is your business logic hard-coded into fragile prompts or is it a portable, governed asset? • Can your context layer ingest unstructured documents as easily as a structured table? If you aren't solving for context, you aren't building an AI strategy. You’re just buying expensive toys. #EnterpriseAI #ContextLayer #AgenticAI #AppOrchid #KnowledgeGraphs #MCP

20

1 Comment

The agentic commerce era just got its "HTTP moment." Google's Universal Commerce Protocol (UCP) creates a standardized way for AI agents to discover, shop, and purchase on behalf of consumers. This isn't theoretical, it's happening now. Shopify is live. Salesforce just announced native support for Agentforce Commerce merchants. But here's what enterprise brands need to understand: if you're on Adobe Commerce, a custom platform, or anything else, enabling UCP isn't as heavy a lift as you might think. The protocol is architecture-flexible by design. The bigger strategic question? Product content. Your descriptions, attributes, and metadata are now being consumed by LLMs, not just humans. Optimizing for agent discoverability is the new SEO. Merkle's perspective on what this means for enterprise organizations 👇 https://lnkd.in/gvq4eX_K If you're thinking through what UCP means for your commerce roadmap, my DMs are open and happy to chat 1-on-1. Daniel Knauf Eric Buss Steven Duran #AgenticCommerce #DigitalCommerce #AI #UCP #CommerceStrategy

44

3 Comments

As AI agents become a common interface for discovering and buying products, it's critical to have a shared, open protocol that flexes to the complexity of commerce. Without a shared protocol, every platform speaks a different language. UCP addresses this and covers the full shopping journey, from product discovery, to cart building, checkout, and post-purchase interactions, across any platform, with any payment processor. Today Microsoft joined the Universal Commerce Protocol (UCP) Tech Council alongside Google, Shopify, and many other!

12

🌍 **What if your enterprise could catch problems before they happen?** New Relic just launched something that's quietly reshaping how enterprises think about data observability: a **no-code AI agent platform** that lets teams build and manage AI agents to monitor company data in real-time. No more firefighting bugs after they've already disrupted your product. These agents catch issues *before* they become crises. **Here's what's significant:** The platform supports the Model Context Protocol (MCP), meaning AI agents can connect to external data sources and integrate seamlessly with existing tools. They've also solved a massive fragmentation problem with new OpenTelemetry (OTel) tools—enterprises can now manage all their data streams in one place instead of juggling multiple platforms. **Why this matters for Africa's tech ecosystem:** For years, African enterprises have been caught between two worlds: the pressure to scale globally while lacking the infrastructure and expertise to manage complex data systems. This democratization of AI-powered observability changes that equation. A startup in Lagos, a fintech in Nairobi, or a logistics company in Accra can now deploy enterprise-grade monitoring without needing a massive engineering team. That's a game-changer for competitiveness. **The bigger picture:** We're moving from reactive to predictive enterprise operations. AI agents aren't replacing engineers—they're amplifying what teams can do. In Africa, where we're building tech infrastructure at scale, this shift means we can leapfrog outdated monitoring approaches and go straight to intelligent, autonomous systems. **💬 My question for you:** How is your organization thinking about AI agents? Are you exploring no-code platforms to accelerate your digital transformation, or are you still building custom solutions? Let's share insights on how African tech leaders are adopting these tools. **Let's build smarter, more resilient systems for Africa, together.** #ArtificialIntelligence #AfricaTech #DataObservability #AIAgents #DigitalTransformation #FutureOfWork #TechInnovation

2

Grammarly isn’t just fixing grammar-welcome its agentic productivity fabric. With the acquisition of Superhuman and its earlier purchase of Coda, Grammarly is doing what Microsoft and Google haven’t fully pulled off yet—connecting where work happens (email, writing, collaboration) to where AI acts and the tools and processes your employees already use. • Start with ubiquity (Grammarly’s 30M+ daily users) • Fill the action gap with Superhuman (real-time, high-frequency decisions in email as one type of task execution) • Orchestrate the outputs in Coda (shared workspaces, structured knowledge, agent coordination ... and a contextual layer for coordination and memory) This isn’t just about “better writing.” It’s an architectural play for CDAOs and AI leaders to understand as they plan for AI-driven work transformation: 1. Intelligence is now a substrate of work as Grammerly works across your apps creating a new set of data that is largely invisible to IT. 2. Keep an eye on "shadow AI" as superhuman/coda will not just see emails, it will house strategic plans, roadmaps, customer data, etc. 3. Review how you acquire solutions as this stack has the potential to absorb countless "long-tail" AI use cases. Grammerly is moving: Start with ubiquity. Embed intelligent actions in the inbox. Orchestrate in the workspace. Your question is how to keep atop implications to your data-to-decision stack. What’s your take—will Grammarly become the agentic OS for knowledge work? Read more from Larry Dignan: http://bit.ly/44Jo8OV #AI #CDAO #ProductivityAI #AgenticWorkflows #DataToDecision #Grammarly #Superhuman #Coda #ConstellationResearch #AIPlatform

5