Claire (Na) Cheng

A lot of times we get asked about knowledge graphs at Ragie but my hot take is that the overwhelming majority of customer use cases don't require them. It's also no surprise to me that knowledge graph hype is evangelized by graph database vendors wherein vendors often compare KG search against pure semantic search but I think it's a bad comparison. They should be comparing against Agentic Retrieval which I think is more likely to be the future of retrieval in the long term. We recently launched Agentic Retrieval at Ragie and Matt Kauffman did a really nice write up explaining how it works. Check it out: https://lnkd.in/gMeyqD7X

31

1 Comment

Here’s the real story behind Anthropic’s 10-day “Claude Cowork” sprint. The article says Claude wrote 100% of the code — but it’s not fully clear whether Cowork truly evolved itself during those 10 days, or whether another LLM coding setup helped bootstrap it. That distinction matters. What matters more: we’re now at the point where an agent can be built that rapidly improves its own capabilities — and starts writing the next version of itself. This is not “AI helps developers.” This is AI becoming the developer of itself. The Anthropic demo is proof this is no longer theory. This can be done now. https://lnkd.in/dMB-Jw8p

9

As foundation models converge, the moat is shifting. The model layer is becoming increasingly commoditized. Switching between providers is getting easier, cheaper, and faster. The real moat is now the wrapper. The workflow, the context, and where AI lives day to day. Companies like MagicSchool AI, Harvey, Jasper, and Glean are not winning because of a specific model. They are winning because they are embedded inside how people already work. Users do not want another AI tool. They want their existing tools to be smarter. In the long run, distribution and workflow ownership matter more than the model itself.

5

I think Databricks’ Genie Code is clearly less about introducing an entirely new product and more about formalizing the steady improvement of its AI development experience over the last year. Not only has the output quality improved over time, but features such as dashboard authoring, Lakeflow Spark Declarative Pipelines support, etc were under-the-radar additions that positively impacted development speeds, quality, and time-to-delivery. Some competitors have had to rethink or replace rough first-generation coding products, Databricks started with something usable and kept improving it, so hey, might as well celebrate with a new name too! Also, I believe the emphasis on more agentic capabilities points to where the platform is going next, especially as Databricks pushes toward a more unified AI product experience built around governed enterprise data.

199

10 Comments

LLMs can help us a lot. But how do we know when they are right? Problem: LLMs make useful answers and scary mistakes. At scale, human review is slow and costly. Simple scores miss tone, bias, and hidden errors. That makes it hard to trust LLMs in customer support, docs, or any high-stakes use. What this PDF offers: • A clear method to use LLMs as judges to score and explain other LLM outputs. • Ways to spot and reduce common biases like favoritism, verbosity, and order effects. • ChainPoll — a practical mix of step-by-step reasoning and multiple votes to get more reliable judgments. • Luna SLMs — smaller, fast models for cheap, real-time checks in production. • Hands-on guidance: prompts, code snippets, test plans, and metrics such as precision, recall, AUROC, and Cohen’s Kappa. • Practical best practices: component scoring, few-shot examples, adversarial tests, and iterative tuning. Why it matters: Use these ideas to catch bad outputs before they reach users. The guide is written for teams who need repeatable, fast checks that match real needs. Follow Shubham Vora to learn more about AI Agents.

205

41 Comments

Seattle Unveils New Responsible AI Plan and Hackathon Series ▶ Mayor Bruce Harrell announced Seattle’s 2025–2026 AI Plan, expanding AI use across city services. ▶ The policy updates Seattle’s 2023 AI framework to cover all AI tools, not just generative AI. ▶ City staff will receive new training and upskilling programs, with partnerships from universities and tech companies. ▶ More than 40 pilots inform the plan, with active projects in permitting, public safety, infrastructure, and communications. ▶ A new AI leadership role will coordinate projects across departments and evaluate outcomes using clear performance metrics. ▶ The Community Innovation Hackathon Series, with AI House, will develop AI tools for youth mental health, permitting, and city services. ▶ The inaugural hackathon focused on enhancing the Youth Connector app to expand access to enrichment and mental health programs. ▶ Seattle is leveraging public-private partnerships like AI House and the Climate Incubation Hub to build innovation spaces and support AI startups. 🔗 Link to article: [https://lnkd.in/eCps96t7] 📚 The AI Policy Newsletter: https://lnkd.in/eS8bHrvG

12

I just published a new blog on the GenAI stack. Over the last few months, I’ve seen a lot of confusion around how all the pieces actually fit together. Models, infra, agents, evals. People often learn them in isolation, but rarely see the full workflow end to end. So I tried to visualize the entire GenAI stack as one connected system. In the blog, I walk through: ✦ A 30,000-foot mental map of the GenAI ecosystem ✦ How to think about the models layer beyond “bigger is better” ✦ The infrastructure layer that quietly makes or breaks production systems ✦ Orchestration as the real brain of modern AI apps ✦ Evals and observability, aka how you avoid things going off the rails Read here: https://lnkd.in/dHGnhiSY

89

19 Comments