Dan Holle

📊 Just browsed the LangChain State of AI Agents survey — and I’m delighted to be counted as a data point (hopefully a valid one) in this pivotal industry report! As AI agents move from buzzword to reality, the 2025 data is clear: organizations are actively deploying autonomous systems across workflows, and financial services is notably part of this adoption wave. 🔍 Here are the key takeaways shaping the future of financial AI development: • Widespread adoption: Over half of respondents report having AI agents in production — and adoption isn’t limited to tech alone. Financial services remain one of the top represented sectors, showing strong interest in agent-based automation beyond traditional LLM chat. • Quality and trust are critical: The top barrier to production isn’t cost — it’s performance quality and reliability. For finance, where precision and compliance are paramount, robust testing, guardrails, and observability are essential. • Controls and observability: Nearly 90% of organizations now implement observability tools to trace agent reasoning — a must-have for financial applications that must be auditable and trustworthy. • Looking forward: The maturity of AI agent tooling signals that next year’s focus will likely shift toward specialized financial agents, tighter security integrations, and agents embedded deeply into operational workflows. 🚀 Bottom line: AI agents are already reshaping how businesses — including financial institutions — analyze data, automate workflows, and make decisions. I’m excited to be part of this evolution and look forward to what’s next in financial AI. #AI #FinTech #AIAgents #MachineLearning #Innovation #LangChain

5

It starts with lineage. Having proper lineage in place is your best starting point for making your enterprise data fit for AI workloads. A simple gut check for data and AI leadership is to review the lineage you have and ask: do I trust it? Is it up to date? Can I rely on it? Or is it mainly there for audit logs and compliance? I’d start there. 📼 From a recent interview I did on Malcolm Hawker’s CDO Matters podcast. We had such a great conversation and touched on a lot of gems, so I might be sharing a few more clips soon!

32

6 Comments

🥶 9 days ago the head of quantum of a Fortune 50 company called Clay Almy and asked if we could get him commercial data on the cryo market in quantum. Yesterday my team Pallavi Soni Jerzy Szuniewicz Pranshi Saxena delivered the brief; backed by heavy data, reviewing 94 SKUs across 30 vendors and analyzing 1,915 public-procurement records. 1️⃣ Distinct Equipment Tiers: The market is segmented into very clear tiers, covering Dilution Refrigerators (DR), ADRs, 4 K-class optical systems, and Sub-2 K bath cryostats. 2️⃣ The "Premium vs. Low-Cost" Overlap: Within the 4 K-class optical tier, there are distinct low-cost options that successfully match premium specifications at a significant discount compared to premium pricing. 3️⃣ Financing is Highly Nuanced: The lease market is quite thin, and standard vendor rent-to-buy agreements are rare. 4️⃣ "As-a-Service" Models are Growing: Buyers can explore dedicated Hardware-as-a-Service (HaaS) for full-system rentals or Measurement-as-a-Service (MaaS) for discrete measurements without rental commitments. 5️⃣ Shared Facilities are an Option: Shared-access user facilities at major research centers exist as an alternative path, which is cheaper but requires dealing with queue times. 6️⃣ Watch Out for Tariff Traps: Equipment might qualify for duty-free import depending on its classification, but overlay regimes shift constantly. Procuring cryogenic systems requires a deep understanding of not just the technical specs, but the financial, legal, and customs implications. We have detailed price points, finance options and technical KPIs broken down along those 6 POVs. 👇 Delivering such a detailed analysis, backed by massive, proprietary data (parts of it come from confidential vendor relationships) within 1 week is only possible because Global Quantum Intelligence, LLC is the only trusted, fully independent and entirely quantum focused intelligence provider. This powers our GQI Factory model: a proprietary data warehouse and quantum ontology that makes data robust, rich and contextualized. This enables our GQI RAG - any and all AI used by GQI is run on the GQI Factory. No Google news, no press releases, no clues from the AI. Just smart expertise. A unique approach rooted in a decade of effort led by Doug Finke David Shaw that is able to deliver reliable and actionable intelligence - not opinions, surveys or marketing pieces. #QuantumIsComing

43

What problems can AI solve? Had a great chat about this with Gregory LaBlanc at #BerkeleyAISummit last weekend. I've been to more than a dozen of big and small conferences just this year, and a lot of people seem to expect AI and Agents to solve a lot of problems. But does AI solve problems? Or it's humans who do? And AI, Agents, and technology in general help humans solve more problems faster. I think it's the latter. And the implication would be that you need more humans in the end, not less. But these humans would need to understand both business operations and technology, which is not that common - big opportunity for Education!

21

3 Comments

Stop the Data Chaos: Part 1 of Our Data Governance Series is LIVE! Are you drowning in data but starved for insights? Experiencing frustrating inconsistencies, compliance worries, or missed opportunities because your data isn't working for you? You're not alone. Many believe data governance is solely an IT problem – a technical hurdle to be dealt with by specialists. But what if that thinking is costing your business dearly? In Part 1: Demystifying Data Governance: Why It's Not Just for Tech Teams, we're pulling back the curtain to reveal: What data governance truly is and why it's a non-negotiable in today's data-driven landscape. Why it's a business imperative, impacting everything from customer experience to bottom-line profits. Real-world examples of the "Cost of Chaos" – from inaccurate reports to hefty compliance fines. The clear "Benefits Blueprint": better decisions, improved compliance, increased efficiency, and enhanced trust. This isn't just theory; it's about understanding why data governance is for everyone, from the CEO to the data entry specialist. Ready to transform your relationship with data? Click the link below to read Part 1 on my Substack and begin your journey from data chaos to clarity! Don't forget to subscribe to my Substack for the rest of the series, including the upcoming Data Governance Health Check Toolkit! #DataGovernance #DataManagement #BusinessStrategy #DigitalTransformation #DataQuality #Compliance #Substack #Part1

2

These days, a semantic layer isn’t just nice to have, it’s often the piece that makes or breaks how well your data serves the business. What do you think? Picking the right one can be tricky. There are lots of products out there, and they’re not all built the same. The wrong choice can leave you stuck with tools that don’t scale, don’t integrate well, or lock you in long-term. I recently read the “Ultimate Guide to Choosing a Semantic Layer” by David P. Mariani, CTO and Co-Founder of AtScale. It’s one of the clearer, more practical guides I’ve come across on this topic. You can check it out here: Ultimate Guide to Choosing a Semantic Layer - https://bit.ly/4kXMWZo Here’s a quick look at what’s inside: - What core capabilities a semantic layer should have to keep your metrics consistent across BI, AI, and NLQ tools - A checklist of features worth looking for when evaluating different solutions - Pros and cons of different architecture choices, like centralized vs. decentralized setups - Smart questions to ask vendors so you’re not just buying into a good sales pitch - Real examples of how big enterprises are using semantic layers today - Advice on avoiding vendor lock-in and keeping your stack flexible for the future If you’re working on modernizing your data stack, or just want to understand how semantic layers fit into the picture, this is a solid read. You can check it out here: Ultimate Guide to Choosing a Semantic Layer - https://bit.ly/4kXMWZo #semanticlayer #ai #atscale #ravenaondata

135

31 Comments

New data infrastructure lore just dropped: The Pendantic Layer. Joe Reis has a great article introducing this idea [https://lnkd.in/gaXw_TPr], but this quote sums it up perfectly. "The Pedantic Layer slows us down by focusing attention on taxonomy and semantics over outcomes. It makes practitioners feel like they need to pass some purity test of definitions before they can get value. It feeds a culture where thought leaders posture about whether CSVs are structured “enough,” while the practitioners who ignore the debate are quietly building models that drive millions in revenue." Coming from startups where you are obsessed with execution over perfection, the "pedantic layer" has been one of my largest gripes with the wider data industry. In celebration of yet another made-up layer (YAML), leave a comment about your favorite data concept, and I'll tell you why you are wrong for obscure nuances that don't apply to most use cases.

45

11 Comments

Hex surveyed data leaders less than six months ago and, while 77% of them said they were excited about AI, only 3% reported making it a priority. However, AI capability has already improved significantly during that time so I was excited to tune into Hex’s recent fireside chat hosted by Charles Schaefer with several data leaders about how they’ve integrated AI into their workflows. I’d definitely recommend listening to the whole conversation but some of my favorite insights were how Kelly Burdine is speeding up development as well as troubleshooting by using AI to better understand the company’s codebase, Stephen Moseley’s framework of “exploration vs. exploitation” to decide when to actually apply AI to a workflow, and Brent Driscoll’s idea that one of the easiest ways to get buy-in from leadership to start using AI would be for the classification and extraction of unstructured data due to its notable efficiency in both time and cost compared to non-AI methods. https://lnkd.in/eJS3ujYd

14

1 Comment

The semantic layer is the data org's white whale. Everyone wants one. Some implement it. Few maintain it. Almost none survives a couple of business cycles. So here's the question worth asking in 2026: do we still need a semantic layer when agents can read code and metadata continuously, in real time? For decades the semantic layer existed because humans couldn't read SQL. It translated raw schemas into business language so analysts and execs could ask questions without engineering training. Agents are now the primary consumers of data. They read code fluently. They write SQL fluently. They compose queries on demand. The vendor response? Rebrand the semantic layer as a "context layer." Same YAML, new acronym. We think there's a more honest answer: most of what a semantic layer provides for agents already lives in well-written pipeline code. Schema contracts, lineage, computation logic, business definitions, all expressible as decorators, type hints, Pydantic classes, and docstrings. In our next webinar we'll walk through the current semantic layer landscape, where headless vs platform-native approaches actually differ, and show real Bauplan pipelines where a single Python file carries what a semantic layer spreads across multiple YAML files and proprietary tools. We'll also be honest about where the equivalence breaks. 🗓️ Tuesday, May 19 🕐 9:00 am PT 🌉 💻 Virtual 🔗 Register: https://lnkd.in/gCU-gKwm

32

4 Comments

Collate, the creator of the leading open-source data intelligence project OpenMetadata, has raised $10M in a Series A funding round. Collate is solving the last-mile data challenges for modern data teams by providing an agentic data intelligence platform. The platform automates tasks and drives collaboration, completing the virtuous cycle with AI to enable data teams to deliver the data needed for AI-readiness. Its orchestrated suite of agentic workflows helps customers like Mango and Decisiv easily find, understand, and govern their data, increasing productivity and accelerating innovation. The funding was led by Venrock, with additional investment from Unusual Ventures and Karman Ventures. The funding will be used to accelerate Collate's AI-first mission for enterprises and support the continued growth of the OpenMetadata open-source community. Congratulations to co-founders Suresh Srinivas, Sriharsha Chintalapani, and the Collate team. Yahoo Finance: https://lnkd.in/ddtq3_jU. #DataIntelligence #OpenSource #AI #DataManagement #SeriesA #Funding #DataGovernance

22

1 Comment

The Strategic Value of Methodological Precision in Data Services Core Principles of Shaolin Data Services Over the past week, we've outlined the statistical rigor—from understanding Parsimony in Multivariate Regression to the necessity of ANOVA over simple means—that underpins every engagement at Shaolin Data Services. This isn't academic indulgence; it is a strategic imperative. The Fractional CDS strategy is a surgical, no-nonsense approach laser-focused on identifying and striking at key data weaknesses. That level of precision is only possible when every variable is correctly classified, every comparison is robustly tested, and every model is built with contextual intelligence (Bayesian). Uncompromising Standards and Client Defense The quality of our deliverables is a direct function of our methodological uncompromising standards. We refuse to take companies seriously if their internal culture allows for inconsistent, exploitative, and fundamentally flawed data practices. Our rigor is your defense. Call to Action and Relevant Hashtags → Get the uncompromising methodology your organization needs. Explore our Fractional CDS services. #StatisticalRigor #DataStrategy #Methodology #FractionalCDS

1

1 Comment

[NEW SERIES]: Putting Agents in your Data Architecture - Are we Ready? 🦞 I have been spending time reading research papers, blogs & understanding various system internals to address a question. "Are our existing data platforms ready to support Agentic workflows?" Traditional data platforms were built for kinda deterministic pipelines. - Engineers design data pipelines - Systems execute scheduled jobs (Apache Spark, Flink) - Queries tend to be bounded by human speed and reasoning The focus has been on repeatability and importantly stability. However, agentic workflows introduce a very different pattern in terms of how they will work inside our data architecture. Agents don’t execute a fixed pipeline. They explore systems dynamically. - They probe schemas - Issue iterative queries - Test multiple hypotheses - Refine strategies based on intermediate results This creates an entirely new class of workload for data platforms. So, we need to think of architectural primitives that can support these capabilities. The first primitive I want to talk about in this series is: ISOLATION! If agents are allowed to experiment with data (whether rewriting queries, testing transformations, evolving schemas), they need isolated environments. Agents may evaluate many “what-if” scenarios in parallel, each representing a speculative path toward solving a task. For that, they need safe environments to experiment and fail without consequences. So, How do we provide isolation for agents exploring large data systems like #lakehouse? ✅ Branch-able data states: allow agents to fork isolated copies of datasets to test hypotheses without touching production ✅ Snapshot-based versioning: enable agents to explore historical or point-in-time views of data safely ✅ Fast rollback mechanisms: since most speculative agent paths fail, systems must discard branches instantly ✅ Copy-on-write storage: create thousands of lightweight isolated environments without duplicating entire datasets. In many ways, the lakehouse with formats like Apache Iceberg might already be closer to agent-ready data infrastructure than we realize. And we already see several data systems already moving in this direction. For e.g. Neon has branching databases, allowing developers (and increasingly AI agents) to spin up isolated copies of production data instantly. Replit's snapshot engine allows AI agents to create instant filesystem forks, experiment in isolated environments, and roll back changes safely. Bauplan uses copy-on-write branching and commit semantics, allowing agents to run pipelines on temporary branches and merge changes only after verification. I am working on a new blog series that will go deep into these primitives. Helpful references in comments. #dataengineering #softwareengineering #ai

145

17 Comments

LLMs are powerful tools, but are they really the gateway to AGI? Richard Sutton, widely regarded as a pioneer of Reinforcement Learning believes they are far from it. In his 2019 article "The Bitter Lesson", Sutton argued that over time, methods that scale with computational power tend to beat approaches that rely on human-crafted domain knowledge. At a glance, this seems to suggest that LLMs fall into the first category as they are a product of massive amounts of compute; however, they lean towards the latter as they incorporate curated data, supervision and RLHF across the training pipeline. Recent developments also suggest that simple scaling rules (e.g. Kaplan's scaling law) have real limits and nuances. In a market saturated with LLMs, Sutton offers a refreshing, first-principles take on intelligent systems. Well worth a watch, kudos to Dwarkesh Podcast for the interview. https://lnkd.in/gqADcTpF

6

Hey, my Iowa State University - Ivy College of Business grad class students. You can get a jump on Thursday night's class, where we'll explore how to build AI models that leverage #CausalAI to avoid or address human decision-making traps. This is the focus of Step 7, "Map Scores to Decision Recommendations," in the Thinking Like a Data Scientist methodology for those studying from home. #ThinkingLikeaDataScientist #TLADS

8

1 Comment

DaaX.ai achieves industry-leading score of 77.7 on the FACTS Benchmark: https://lnkd.in/e2JRr87H See this Technical Note on DaaX's methodology and results: https://lnkd.in/eJey6dGS Learn about the FACTS benchmark here: https://lnkd.in/e9i6gWJY ..."The FACTS Benchmark Suite, introduced by Google DeepMind and Google Research, is a comprehensive framework to measure the factual accuracy of LLMs." DaaX used the benchmark to measure the factuality provided by their neuro-symbolic AI technology. Source: https://daax.ai/

6

3 Comments

Agree with this article that macroeconomic data collection can/should be modernized to incorporate newly-available digital data in real time vs. over-reliance on imperfect & delayed survey data. This is a good argument for increased investment in collection agencies like BLS & Fed. Instead we’re getting budget cuts and vendetta political firings of statistical professionals when they report truthful data that annoys the President.

3

I think we are witnessing a new role arising in the data space, and I have seen people in the know talking about it. Now that LLMs have brought data science to all technical functions, the lines are staring to really blur between specializations. There's a new creature forming- mostly in the SAAS startup world for now but bound to spread- and that is a person that both writes software, uses the latest data science toolkits, and is moving beyond insights to action in the form of building products. Are you seeing this too, and what should it be called? Let's give it a Kripkean baptism! I propose a "foreword deployed data science engineer" FDDSE! Clumsy, I know. Have a better name?

1

13 Comments