Christine Lee

New short course: DSPy: Build and Optimize Agentic Apps DSPy is a powerful open-source framework for automatically tuning prompts for GenAI applications. In this course, you'll learn to use DSPy, together with MLflow. This is built in partnership with Databricks and taught by Chen Qian, co-lead of the DSPy framework. Many AI builders spend hours hand-tuning prompts. When given a set of evals, DSPy automates this process. It’s especially useful for optimizing prompts, including few-shot prompts, in complex agentic AI workflows. Further, if you switch an application to a newer LLM, performance can degrade if your prompts were optimized to the previous model. DSPy automatically optimizes the entire system for the new LLM as well, using just a few evaluation examples. This course teaches DSPy works, and best practices for using it. You’ll write programs using DSPy’s signature-based programming model, debug them with MLflow tracing -- to gain visibility into how different parts of a pipeline, as well as how the overall system, are performing -- and automatically improve their accuracy with DSPy Optimizer. Please sign up here: https://lnkd.in/gdjae8AX

4,672

190 Comments

The $1.2 Million Ghost in the Machine: How BigQuery + Looker Turned a Silent Leak into a Profit Engine at Promevo https://lnkd.in/gnx4nxRD For years, the boardroom consensus was clear: Data is a cost center. A necessary burden. A "tax" on doing business. They were wrong. While most organizations are drowning in "data lakes" that act more like digital swamps, a quiet revolution is happening. Look at the Promevo revelation: By centralizing fragmented data into the BigQuery + Looker powerhouse, one organization didn't just "find insights"—they recaptured $1.2 million in annual revenue that had simply vanished into the cracks of manual errors and system silos. That is 10% of their total revenue. Found. Reclaimed. Put back to work. This isn't just "Analytics." This is Financial Alchemy. Here is how the leaders of tomorrow are shifting from a Cost Center to a Profit Center using the Google Cloud AI & Data stack: 🚀 1. From Static Reports to "Data Products" Stop sending PDFs. Start selling intelligence. Promevo didn't just build a dashboard; they built Promevo Insights. By embedding Looker, they turned internal data into a premium, customer-facing revenue stream. They stopped being a vendor and became an indispensable intelligence partner. 🧠 2. The Semantic Layer The Single Source of Truth The "Cost Center" trap happens when Finance, Sales, and Ops all argue over whose spreadsheet is right. Looker’s LookML provides a unified, audited data layer. When everyone speaks the same language, the "speed to decision" drops from months to minutes. ⚡ 3. AI that Actually Acts (Not Just Chats) The future isn't a chatbot that summarizes your problems. It’s an Agentic Workflow. We are seeing the shift from "What happened?" to "What should I do?" Imagine a system that detects a supply chain leak, recommends the fix, and allows you to execute the correction directly within the UI. ⏱️ 4. The Velocity Advantage The old way: 12 months to build a custom BI stack. The Looker way: Months. In the digital economy, being late is the same as being wrong. The Bottom Line: Your data is currently either an expense on your P&L or the most valuable asset on your Balance Sheet. The difference lies in the platform you choose to activate it. If you aren't using BigQuery and Looker to monetize your insights, you aren't just falling behind—you're leaving millions on the table for your competitors to find. Are you ready to stop "managing" data and start profiting from it? #DataMonetization #BigQuery #Looker #GoogleCloud #DigitalTransformation #ThoughtLeadership #AI #DataAnalytics #ProfitCenter

18

Building agentic workflows? There are actually only a few key concepts to be aware of. StateGraph orchestration: A StateGraph is the agent’s blueprint. It consists of nodes that handle tasks with edges (directions) to other tasks, defining the order. Together, they provide the agent with a flow to follow. Conditional routing: At decision points, the agent inspects results and follows “if-else” statements that skip failures, handle edge cases, etc. Memory and threads: Thread memory holds recent conversation turns. Essentially, it stores the short-term memory that you are John Doe and were interested in the weather. Long-term memory stores user facts, letting the agent stay context-aware, reduce repetition, and personalize replies, such as that you have been a customer since 2018. Human-in-the-loop: Strategic hand-offs let people review low-confidence outputs, catch critical paths or errors, and feed corrections back to the model, keeping quality and alignment tight. If this still feels too abstract, I’ve included a more detailed explanation in the article below. https://lnkd.in/eAr-6YAQ

14

1 Comment

The Goodreads datasets are a goldmine for anyone working on recommender systems in the literary world. They go way beyond simple ratings, offering: 📚 Explicit + Implicit Signals: Star ratings and shelf data like “to-read” or “read” 📝 Rich Review Text: Ideal for NLP-powered recommendations 👥 (Sometimes) Social Graphs: Explore influence between readers 📊 Extensive Metadata: Titles, authors, genres, descriptions, and more Whether you're experimenting with collaborative filtering, NLP models, or hybrid systems, Goodreads data offers an incredibly nuanced playground. We just published a deep dive into what these datasets contain, how to use them responsibly, and why they remain essential for pushing the frontier of book recs. 🔗 https://lnkd.in/ev2N3rCr

20

2 Comments

𝐀𝐝𝐨𝐩𝐭𝐢𝐨𝐧 𝐢𝐬 𝐚 𝐓𝐞𝐚𝐦 𝐒𝐩𝐨𝐫𝐭: 𝐓𝐚𝐜𝐭𝐢𝐜𝐬 𝐅𝐫𝐨𝐦 𝐓𝐡𝐞 𝐅𝐫𝐨𝐧𝐭 𝐋𝐢𝐧𝐞𝐬 After my last post on why most data products fail (hint: it’s not the tech), the comments section delivered more practical value than most “best practices” decks I’ve seen. Shout-out to Michael Lee, Patrick Ackermann, Christina Jiang and Juan Carlos C.for surfacing what actually works - these are the tactics that move the needle: 1️⃣ 𝐒𝐩𝐨𝐧𝐬𝐨𝐫𝐬𝐡𝐢𝐩 𝐢𝐬 𝐧𝐨𝐧-𝐧𝐞𝐠𝐨𝐭𝐢𝐚𝐛𝐥𝐞 As Christina put it, visible, vocal backing from senior leaders isn’t optional. If leadership isn’t leaning in, don’t expect org-wide change. 2️⃣ 𝐑𝐞𝐩𝐞𝐭𝐢𝐭𝐢𝐨𝐧 𝐰𝐢𝐭𝐡 𝐑𝐞𝐚𝐥 𝐈𝐦𝐩𝐫𝐨𝐯𝐞𝐦𝐞𝐧𝐭 Change is a grind, not a launch event. Christina’s “repetition with refinement” point is dead-on: keep iterating, keep explaining the “why,” and don’t let up when you hit the “valley of doubt.” 3️⃣ 𝐏𝐨𝐰𝐞𝐫 𝐔𝐬𝐞𝐫𝐬 𝐌𝐚𝐭𝐭𝐞𝐫 𝐌𝐨𝐫𝐞 𝐭𝐡𝐚𝐧 𝐏𝐨𝐰𝐞𝐫𝐏𝐨𝐢𝐧𝐭 Michael and Juan Carlos both highlighted the power of early champions - those who ask the tough questions and drive real adoption. Managers usually know who these are; your job is to identify and empower them. 4️⃣ 𝐏𝐫𝐨𝐜𝐞𝐬𝐬 𝐌𝐚𝐩𝐩𝐢𝐧𝐠 𝐁𝐞𝐚𝐭𝐬 𝐏𝐫𝐞𝐭𝐭𝐲 𝐔𝐈 Patrick nailed it: if your product isn’t mapped to real workflows, it’s shelfware. Integration with daily processes is what separates “nice dashboard” from “must-have tool.” 5️⃣ 𝐖𝐡𝐚𝐭’𝐬 𝐈𝐧 𝐈𝐭 𝐟𝐨𝐫 𝐌𝐞? Michael’s “what’s in it for me?” test is the adoption acid test. If users don’t see a clear benefit, don’t expect usage - no matter how good the tech. My takeaway: 𝘐 𝘴𝘵𝘪𝘭𝘭 𝘴𝘦𝘦 𝘮𝘰𝘳𝘦 𝘢𝘥𝘰𝘱𝘵𝘪𝘰𝘯 𝘧𝘳𝘰𝘮 𝘤𝘩𝘢𝘮𝘱𝘪𝘰𝘯𝘴 𝘢𝘯𝘥 𝘳𝘦𝘢𝘭 𝘸𝘰𝘳𝘬𝘧𝘭𝘰𝘸 𝘪𝘯𝘵𝘦𝘨𝘳𝘢𝘵𝘪𝘰𝘯 𝘵𝘩𝘢𝘯 𝘧𝘳𝘰𝘮 𝘢𝘯𝘺 𝘴𝘦𝘯𝘪𝘰𝘳 𝘴𝘱𝘰𝘯𝘴𝘰𝘳 𝘢𝘭𝘰𝘯𝘦. 𝘉𝘶𝘵 𝘸𝘪𝘵𝘩𝘰𝘶𝘵 𝘴𝘱𝘰𝘯𝘴𝘰𝘳𝘴𝘩𝘪𝘱, 𝘺𝘰𝘶 𝘯𝘦𝘷𝘦𝘳 𝘨𝘦𝘵 𝘰𝘧𝘧 𝘵𝘩𝘦 𝘨𝘳𝘰𝘶𝘯𝘥. What’s your best move for identifying and empowering “champion” users? Any incentive or recognition tactics that actually work? #Analytics #AI #ProductManagement #ChangeManagement #Adoption #Leadership

8

5 Comments

𝗥𝗲𝗽𝗹𝗶𝘁 𝗔𝗴𝗲𝗻𝘁 𝟯: 𝗧𝗮𝗸𝗶𝗻𝗴 𝗔𝘂𝘁𝗼𝗻𝗼𝗺𝘆 𝘁𝗼 𝘁𝗵𝗲 𝗡𝗲𝘅𝘁 𝗟𝗲𝘃𝗲𝗹 🔥 If you thought AI agents were cool before, Replit’s latest release, Agent 3, is a serious leap forward. It’s built to do more on its own — test, fix, build, automate — all with minimal supervision. Here’s what you should know👇 🔗 https://lnkd.in/eVXcsH6F 𝗪𝗵𝗮𝘁 𝗶𝘀 𝗔𝗴𝗲𝗻𝘁 𝟯? Agent 3 is the newest version of Replit’s AI agent that helps you move from idea → app more effortlessly. It doesn’t just suggest code; it builds more of the workflow for you, monitors its own work, tests, fixes, and can even create other agents/automations. 𝗞𝗲𝘆 𝗙𝗲𝗮𝘁𝘂𝗿𝗲𝘀 & 𝗨𝗽𝗴𝗿𝗮𝗱𝗲𝘀 • 𝗘𝘅𝘁𝗲𝗻𝗱𝗲𝗱 𝗔𝘂𝘁𝗼𝗻𝗼𝗺𝘆 / 𝗥𝘂𝗻𝘁𝗶𝗺𝗲: Agent 3 can run for up to ~200 minutes autonomously. That means fewer interruptions and more of the “set it and forget it” feel when you have well-defined tasks.  • 𝗔𝗽𝗽 𝗧𝗲𝘀𝘁𝗶𝗻𝗴 + 𝗦𝗲𝗹𝗳-𝗖𝗼𝗿𝗿𝗲𝗰𝘁𝗶𝗼𝗻: It periodically runs real tests in a browser: clicking through UIs, submitting forms, checking APIs, etc. If it finds bugs, it automatically fixes what it can and retests. • 𝗕𝘂𝗶𝗹𝗱 𝗠𝗼𝗱𝗲𝘀: Design-first or Full-stack: You can start by just prototyping the front end/design, or go straight into building full functionality. More flexibility depending on how well you know what you want. • 𝗔𝗴𝗲𝗻𝘁𝘀 & 𝗔𝘂𝘁𝗼𝗺𝗮𝘁𝗶𝗼𝗻𝘀: One of the cooler bits — Agent 3 can build other agents/automations. Think bots (Slack, Telegram), scheduled tasks like daily reports, or workflows integrating tools like Notion, Linear, etc. 𝗪𝗵𝘆 𝗜𝘁 𝗠𝗮𝘁𝘁𝗲𝗿𝘀 𝗦𝗮𝘃𝗲𝘀 𝗧𝗶𝗺𝗲 & 𝗥𝗲𝗱𝘂𝗰𝗲𝘀 𝗖𝗼𝗴𝗻𝗶𝘁𝗶𝘃𝗲 𝗟𝗼𝗮𝗱: Less babysitting. When the agent can test & fix on its own, you don’t need to constantly switch back and forth. 𝗟𝗼𝘄𝗲𝗿 𝗕𝗮𝗿𝗿𝗶𝗲𝗿 𝗳𝗼𝗿 𝗡𝗼𝗻-𝗖𝗼𝗱𝗲𝗿𝘀: If you’re more idea-driven than detail-driven, Agent 3 helps translate ideas into working apps with fewer technical steps. 𝗕𝗲𝘁𝘁𝗲𝗿 𝗔𝗽𝗽 𝗤𝘂𝗮𝗹𝗶𝘁𝘆: Built-in testing & iteration means more bugs caught earlier, more robust behavior out of the box. 𝗔𝘂𝘁𝗼𝗺𝗮𝘁𝗶𝗼𝗻 & 𝗦𝗰𝗮𝗹𝗶𝗻𝗴: With the ability to build agents/automations, you can delegate recurring tasks. Consider automating your workflow, not just building a single app. Thanks for reading ✨𝐀𝐈 𝐏𝐌 𝐁𝐘 𝐃𝐄𝐒𝐈𝐆𝐍. ♻️ Repost to help someone level up with AI. 🚀 Subscribe on Substack to become an Agentic AI PM.

13

4 Comments

This looks useful for folks in research or scanning what's coming out on the frontier: Lumi from Google's PAIR research team, a tool for better understanding papers on arxiv. Reading smarter, not harder -- that tagline brings me back to Mortimer Adler's classic 'How to Read a Book'! From their GH repo: Lumi uses AI to help you quickly read and understand arXiv papers. Features include: ✏️ AI-augmented annotations - read summaries at multiple granularities 🔖 Smart highlights - highlight text + ask questions 🖼️ Figure explanations - ask Lumi about images in the paper Demo: https://lnkd.in/ex7erTBb GH repo: https://lnkd.in/emz_Ghcs Medium article: https://lnkd.in/enky8dKC GitHub discussions: https://lnkd.in/efAmiCcp

19

🤖RLAD Training LLMs to Discover Abstractions for Solving Reasoning Problems. 🔍 Motivation: How do we move large language models (LLMs) beyond pattern matching—toward algorithmic, abstract reasoning on hard problems? Most current RL post-training approaches create longer chains of thought, but they often drift into verbose or degenerate exploration, missing true reasoning depth. 📄 Solution: RLAD introduces 'reasoning abstractions'—concise, natural language descriptions of procedural and factual knowledge that guide models in learning successful reasoning strategies. The approach trains two LLMs: one to generate abstractions and another to solve problems using those abstractions, via a novel RL setup. Crucially, abstractions act as high-level subgoals or skills, helping models explore diverse solution paths and generalize to harder tasks. 🔑 Results: RLAD achieves up to 44% improvement on math reasoning benchmarks (AIME 2025, DeepScaleR Hard, AMC 2023) over state-of-the-art long chain-of-thought RL methods. Moreover, the benefits scale with more compute allocated to abstraction generation versus solution sampling—a key insight for practical deployment. 💡 Implications: Abstraction-driven training broadens the reasoning abilities of LLMs, unlocking new horizons for model interpretability, strategy diversity, and generalization across complex domains. This direction is especially relevant as scaling chain-of-thought length yields diminishing returns. (Attached: Original research PDF)

13

1 Comment

There is broad consensus in the data & AI community that agentic architectures hold a lot of potential and that they will play a key role in the future enterprise. But how, exactly, will agentic architectures look? Eric Broda and Davis Broda deliver a compelling answer in Agentic Mesh, a new book hot off the press from O’Reilly. Agents will act in a mesh, and they will play different roles in order to perform tasks, monitor activity and guide the overall direction of these tasks. Agentic Mesh is likely to be the go to manual for setting up agentic architectures at scale for the enterprise. It is my pleasure to welcome Broda and Broda on Data, Explored precisely to explore the topic of Agentic Mesh, a topic that is vital for everyone in the data & AI community to understand. These are the questions I will ask. Let’s open this up with a challenge – and this is perhaps an easy one: Why an agentic mesh? Why not an agentic monolith? What is the difference between AI Workflows, Autonomous Agents and Enterprise-Grade Agents? How do you translate a team and an organization to groups of agents – what’s the fleet and the ecosystem? Enterprise grade agents have seven characteristics: Security, Reliability, Explainability, Scalability, Discoverability, Observability, Operability. No one can remember that without repetition – but, can you describe, what these seven characteristics in total makes an enterprise grade agent be like? What is the agentic mesh registry? A historic perspective: What is the past, present and future of agentic architectures? How was it like writing this book? And what has been the feeling – in terms of timing, are you too early, spot on, to late, what has been the reaction? #agenticai #agents #agenticmesh #mesh #data #ai #datagovernance #aigovernance #agentic #generativeai #largelanguagemodel hope to see you Hala Nelson François Rosselet Anis Aknouche Charlotte Ledoux Peter Baumann Lorenzo De Tomasi Karl Ivo 🎧 Sokolov Darren Wood (MSc, MCIPR, MCMI, ACIM) Tobias Zwingmann Yoann Benoit Kingsley Uyi Idehen Andrew Jones Nilton Kazuyuki U. Edgar N. this webinar is for free and shares insights in the newest thinking on Agentic AI - link in comment

75

11 Comments

This section on how dbt Labs transitioned from a purely PLG motion to layering on enterprise is a fascinating one. Two instructive passages that I have bolded (below from First Round Review's path to PMF series featuring dbt Labs) TLDR: GTM comprises your ICP, channel, and message. When you transition from PLG / bottom up to Enterprise / top down motion, naturally your ICP and channel also changes, but your messaging / proposition needs to change too, e.g., the enterprise may be multiple personas and are buying assurance as much as solution. --- "Handy found product-market fit organically for dbt as an open-source tool mostly used by data practitioners and developers. But a few years into running a commercial business, he realized he had to build a growth curve all over again with C-suite data leaders. “Even though we had an unbelievable amount of market pull, as we initially commercialized, it wasn’t easy for us to transform this open-source command line tool into a product that enterprises would pay a million dollars for,” says Handy. “When you have enough product-market fit, sometimes it allows you to get away with not being super tight on product marketing or sales motions. So around 2022, we went from this gigantic acceleration curve and overnight we realized, we have to sit at the adults’ table and figure things out real fast,” says Handy. After the PLG flame started to fizzle, Handy turned his attention to layering on a sales-led motion for the cloud platform. “We had to focus our efforts on telling cohesive stories to senior data leaders. We had to have a very clear, explainable answer to the question, ‘Why should I use the commercial product and not the open-source product? And it had to be digestible by someone with a C in their title,” he says. Handy’s answer: dbt Cloud can handle complex data for companies of every size. “The longer people used dbt, the more complex their code became,” he says. “It was a problem for the most sophisticated dbt users, who were often at the largest companies. So there was a real opportunity for us to step in and solve that for them with dbt Cloud.” To tell that story to enterprise customers, Handy relied on data, naturally. “At a user conference we presented a chart that showed the number of dbt projects that had a certain number of models in them — over 100, over 1,000, et cetera,” he says. “We watched that number climb and we knew as users ourselves, ‘Oh my God, trying to work in a dbt project with 5,000 models in it is challenging.’ So we started with that quantitative data point and asked folks in our community about their experiences with these very large, complex dbt projects, and validated that this was a pain in the ass without a cloud platform.”

23

4 Comments

Quantization in Large Language Models (LLMs) As LLMs scale, intelligence is no longer the bottleneck — memory, latency, and cost are. Quantization turns out to be one of the most powerful (and misunderstood) techniques enabling real-world deployment of large models. In this blog, I break down: • Why LLMs are surprisingly quantization-friendly • INT8 vs INT4 vs NF4 — what actually works in practice • How modern architectures absorb low-bit noise • Where quantization fails and how to design around it Quantization is no longer an optimization step — it’s foundational infrastructure for production LLM systems. Link: https://lnkd.in/gqGje75G #LLMs #AIEngineering #Quantization #MachineLearning #GenAI #MLOps

9

2 Comments

Synthetic data is transforming pricing. Within two years, it will become standard practice in pricing design and optimization—unlocking greater diversity in research approaches and enabling pricing models that align more closely with actual value. This post summarizes my current understanding of synthetic data and the role it plays in pricing design. https://lnkd.in/gBVPyyHX #pricing #AI #data #syntheticdata

13

1 Comment

The classic forecasting dilemma: Do you want an interpretable model (like ARIMA) that's hard to scale, or a complex "black box" deep learning model that is? A new paper from Google, ARIMA_PLUS, suggests we can now have the best of both worlds, while leveraging it all inside BigQuery. The paper introduces a system for large scale, in-database time series forecasting and anomaly detection. It's designed to be automatic, accurate, and, most importantly, interpretable. The performance claims are massive: -Accuracy: On the standard Monash benchmarks, ARIMA_PLUS outperformed popular DL models like N-BEATS, N-HiTS, and PatchTST. -Scalability: It forecasted 100 million time series in just 1.5 hours with a single SQL query. -Interpretability: This is the best part. It’s not a black box. ARIMA_PLUS is a structural model that automatically decomposes every forecast into its core components. This means you can finally see why your forecast looks the way it does. The model can explicitly show you the impact of: -The long term trend -Multiple Seasonalities (like daily, weekly, or yearly cycles) -Holidays and custom Events -External covariates (like ad spend or price changes) This is a game changer for any data team running forecasting on BigQuery. It makes powerful, understandable results accessible to anyone who can write SQL. Paper: https://lnkd.in/gju5p38K

134

16 Comments

Skills over headcount Hiring more data scientists isn’t the only lever. Enabling Citizen Data Scientists (CDS) might deliver value faster. Enter CDS - Domain experts + low/no-code = workable models and insights. - They handle “simple to moderately sophisticated” analytics and pass the best ideas to DS for production. Operating model - Educate: core analytics concepts, basic stats, reading model outputs. 📘 - Equip: AutoML, metadata management, data glossaries, self-service marketplaces. ⚙️ - Enable: sandboxes to test safely without touching production. 🧯 Outcome - Higher throughput of experiments. - Better alignment between business needs and DS work. - DS time shifts to advanced science, not “1+1=2” requests. 👉 Question: If you launched a CDS program next quarter, what’s the first capability you’d teach? Bonus - How it works: 1️⃣ Governance 2️⃣ Controls 3️⃣ Education and limits ... the goal is giving the right tools at the right moment for the right tasks. 🔗 Full article published on Medium (link in comments)

7

2 Comments

The Semantic Leap Beyond LLMs The Data World, has two hot topics : - LLM-based AI → Autoregressive generation, predicting one token at a time - Ontologies/ Knowledge Graphs → Human-driven collaborative agreements about what words mean, taxonomies encoded in semantic graphs Both are great but have their constraints. AI (LLMs) → Great at correlation, no real model of meaning. Just probability distributions over sequences. Statistical mirrors with no deep understanding (2d, 3rd order, causality etc), with hallucinations. Ontologies → Foundational for semantic interoperability, enabled shared meaning across systems for decades. Designed for stability, not adaptability: • Information not execution • shaped by who’s in the room • Static: world often moves faster • Context-sensitive: “Customer” shifts by process/time (Not flaws, design boundaries we can extend) AI is the future but requires three leaps: 1, Semantic Understanding. Before AI can reason, it must comprehend. Not statistically correlate, but structurally & semantically understand. Means moving from "what token probably comes next" to "what does this actually mean, grammatically and semantically?" When you say "Calculate the derivative," the system truly understands not pattern-match to maths content. It recognizes: - The deep actual intent - A specific operation type - A deterministic processing requirement 2, Context State Graphs: Understanding isn't static, every decision has state, reasoning, history, provenance, including, data used to make decision based on specific bus context. Current systems have no memory of why they're in their current state. No trace of the transitions or audit trail of reasoning. We need graphs that capture not just "what is true now" but "how did we get here, and what decisions were made at each transition?" This is the difference between a snapshot & a story. 3, Deterministic but Dynamic Execution: Once you have semantic depth & state history, you can build systems that execute rather than guess. When a clinician asks "Does this drug interaction contraindicate the current prescription?", that's not a token prediction problem. It's a: 1- Query intent classification problem 2- Pharmacological database retrieval problem 3- Interaction rule evaluation problem 4- Risk-scored clinical recommendation problem Each step is executable, empirical, traceable, auditable, repeatable & bounded in a finite decision space. The Convergence Future AI, isn't bigger models it's structured systems that combine: 1- Explicit linguistic intelligence (grammar as routing signal) 2- State driven semantic graph processing (meaning as structure) 3- Domain-specific deterministic pipelines (computation as certainty) Data Object Graphs provide this convergence These types of graph and AI-based solutions will be the systems of record in the future I'm going to post a more detail about this very exciting leap in 2026, so stay tuned. Nebulyx AI

43

48 Comments

So many reactions to OpenAI’s new “agent builder.” While the excitement is palpable, it's important to recognize that orchestration isn't a new concept. Frameworks like LangChain, LangGraph, CrewAI, and visual builders such as Langflow, Make, Zapier, and Flowise have been paving the way for years, making orchestration possible. OpenAI's new tools have certainly drawn fresh attention to this critical layer of infrastructure, but we must also acknowledge the often-overlooked eval layer that is equally important. This distinction matters. Every major platform moment follows a similar trajectory: builders recognize the potential first, and the market catches up later. However, once orchestration becomes accessible to everyone, it transitions from being a differentiator to a necessary infrastructure. The challenge now lies not in connecting steps but in connecting meaning—understanding what to connect it to, the business domain, the data, and how to leverage orchestration effectively. This shift is central to my exploration in The AI Executive’s Handbook: “Prompt engineering was a parlor trick. Context engineering is an architecture.” The orchestration layer connects tools. The context layer connects purpose. The domain layer connects value. This is where the next competitive advantage resides—not in the quantity of agents built, but in their understanding of the world they operate in. Stay tuned for a request for help in selecting a cover tomorrow. #AI #ContextEngineering #AgenticAI #TheAIExecutivesHandbook #Governance #LangChain #Strategy #OpenAI

27

4 Comments

I’ve heard a few people say that Recast is built on top of Robyn. To be clear, it’s not. Recast predates all of the open-source MMM packages. When we started, the open-source ecosystem for media mix modeling was (at best) nascent. No Robyn. No Orbit. No Meridian, etc. If you wanted to do serious marketing measurement, you had to create your own model – or you couldn’t do it at all. The only tools available at the time were Prophet (a generic time series forecaster) and CausalImpact (a Bayesian regression framework for interrupted time series). Prophet didn’t have the flexibility or the statistical foundation we needed. CausalImpact was better – we used it for some early thought experiments – but it was not built for ongoing budget allocation. So we built Recast from scratch. Some of our first attempts used custom Kalman filters to handle adstock, and we actually wrote our own filter variants. But those early models couldn’t handle seasonality or long-term effects in the way we needed. Eventually, we moved to Gaussian processes. That decision was painful because it added a ton of complexity, but it let us model time and saturation effects in a much more principled way. Since then, we've watched the open-source ecosystem grow. We dig into good ideas when we see them, and share our own as transparently as we can. But just to be clear about where our model comes from – it’s tens of thousands of hours of work between my co-founder Michael, myself, and our incredible research team to build the most advanced MMM we’re capable of.

163

6 Comments