Eventual Consistency | Your Reality Check on What's Actually Happening in Data

Seven cloud data warehouses, one open benchmark, and no single winner. Your warehouse choice probably came down to what someone used at their last job, plus a vendor bake-off with the outcome decided before it started. Then the bill arrived. James Winegar and Ross Katz from CorrDyn, who ran the benchmark, alongside Dave Yaffe, Co-Founder of Estuary break down what happened when they loaded the same data into Databricks, BigQuery, Redshift, Microsoft Fabric, ClickHouse, MotherDuck and Snowflake. You'll come away with a sharper way to shortlist engines, size compute and challenge vendor claims before you commit budget to your data stack. The conversation covers warehouse sizing, concurrency, idle time billing, open table formats like Iceberg, and how AI agents change query load on your data infrastructure. It's built for heads of data, CTOs and data engineering leads weighing a warehouse decision. If you want one tool crowned the winner, this benchmark won't give you that. Key Takeaways Under 10GB, pick the cheapest engine you can run. At 100GB and 1TB, the speed-versus-cost curve flips, and a bigger warehouse can cost you less. "Useful concurrency" combines team size, data volume and a latency target into one sizing number. AI agents can make a ten-person team query like a thousand people. Redshift and Fabric beat their reputations in testing, with caveats that bite if you run 24/7 or juggle many workloads. A second vendor you've tested on your own workload is your best leverage when a three-year contract lands on your desk. Chapter Markers 00:00 Why build a warehouse benchmark at all 02:17 Why Estuary backed the project 03:33 Data size, concurrency and cost dimensions 04:44 How teams really pick warehouses today 06:02 Why vendor-run benchmarks fall short 10:13 Landing 2.2TB into seven engines 14:11 Query validation for apples-to-apples results 15:21 Sizing compute: speed versus cost curves 19:43 Scan aggregates, ClickHouse and MotherDuck 22:10 Defining useful concurrency 27:34 How idle time drives your bill 32:36 Reputations confirmed and contradicted 39:15 Home field advantage caveats 42:14 Open table formats and managed Iceberg 47:29 Agentic analytics and warehouse load 53:10 AI agents inside the data pipeline 58:06 No clear winner across the board 1:01:35 Why autoscaling was switched off 1:03:21 How to read any vendor benchmark 1:05:43 Before you sign a three-year contract 1:08:57 What could change in a rerun Useful Links & Resources Data warehouse buyer's guide and companion app (built on this benchmark) Benchmarks referenced: TPC-DS and ClickBench Apache Iceberg (open table format) CorrDyn: https://corrdyn.com Connect With the Show CorrDyn on LinkedIn: https://www.linkedin.com/company/corrdyn/ Ross Katz on LinkedIn: https://www.linkedin.com/in/b-ross-katz/ James Winegar on LinkedIn: https://www.linkedin.com/in/james-winegar/ David Yaffe on LinkedIn: https://www.linkedin.com/in/davidyaffe/ Website: https://corrdyn.com Which warehouse reputation does your own experience contradict? Tell us where the benchmark matches your workload, and where it doesn't. #EventualConsistency #DataEngineering #DataWarehouse #DataInfrastructure #Iceberg

Your CEO saw the agentic analytics demo. Now someone has to make it work. Leadership wants an AI agent answering every data question, and the CFO is already counting the analyst headcount it will save. You're the one who has to explain what the demo left out. Ross Katz, Principal and Data Science Lead at CorrDyn, argues that an AI agent does an analyst's job without knowing what actually matters to your business. So the real work moves into your semantic layer and context, plus the feedback loop that keeps them accurate. You'll leave with a way to scope a first prototype, the questions to put to any vendor, and a straight answer for the CFO. Jason Bradwell puts the questions to Ross on what agentic analytics is, how the vendors stack up, and where lock-in hides in your data stack. This one is for heads of data and analytics weighing build vs buy as AI disruption reaches BI. If you want a ranked vendor list, look elsewhere. Key Takeaways Your semantic and context layers decay over time. Anthropic hit this building its own agentic analytics system, and it changes the headcount math. Ross has one vendor category he'd steer clear of, and one type of company where he'd say go for it anyway. Keep your documentation in version control and you keep the option to migrate later, even inside a platform built to hold on to you. A $10,000-a-day AI bill is believable. The bigger question is how nobody caught it sooner. Chapter Markers 00:00 Why agentic analytics is everywhere right now 01:29 Defining agentic analytics in plain terms 04:03 From Excel exports to self-serve BI 08:09 Is the agent really an AI analyst? 10:46 Why humans still curate what matters 13:36 Pushing back on an all-in executive 16:23 The semantic layer and the feedback loop 22:36 Data as an argument about priorities 23:25 Vendor tour: hyperscalers and data warehouses 28:16 Keeping your options open on lock-in 29:42 BI tools with AI built in 30:37 The trouble with all-in-one platforms 34:32 The questions to ask any vendor 35:45 When all-in-one does make sense 39:02 What to tell the CFO about headcount 43:56 What We're Watching: the $10,000-a-day AI bill 47:26 Guardrails for token spend 49:51 Wrap-up Useful Links & Resources Ross Katz on LinkedIn: https://www.linkedin.com/in/b-ross-katz/ Ross Katz on X: https://x.com/brosskatz Platforms mentioned: Snowflake Cortex, Databricks Genie, BigQuery, Amazon Bedrock, Microsoft Fabric, MotherDuck, Looker, Omni, Rill, dbt More from CorrDyn: https://corrdyn.com Related show: Data in Biotech, also from CorrDyn Connect With the Show CorrDyn on LinkedIn: https://www.linkedin.com/company/corrdyn/ Website: https://corrdyn.com Listen on Apple Podcasts and Spotify Has your company put an agent in front of its data yet? Tell us what broke first, or which vendor question you wish you'd asked before signing. And if you think Ross is wrong about all-in-one platforms, make the case in the comments. If you want to talk about your data challenges, or you think we got something wrong, find us at corrdyn.com. #EventualConsistency #AgenticAnalytics #DataEngineering #SemanticLayer #CorrDyn

Nvidia just paid almost $13 billion for a company with $150 million in revenue. That math only makes sense once you stop thinking about revenue. Every big AI acquisition headline feels urgent, but figuring out whether it changes anything about your data stack or your vendor risk is another matter entirely. Ross Katz, principal and data science lead at CorrDyn, works with data teams day to day on exactly the kind of infrastructure decisions this deal touches. He walks through the acquisition using the same framework he'd use with a client evaluating any major shift in the data industry. You'll walk away with a clear read on why Nvidia bought Hugging Face, what "aggregating demand" means in practice, and whether enterprises leaning on Hugging Face need to change anything right now. Ross also flags the one thing worth putting on your 3 to 5 year roadmap regardless of how this deal plays out. This episode covers Nvidia's smile curve strategy, the difference between supply aggregation and demand aggregation, Hugging Face's security breach and what Nvidia's involvement means for trust in the platform, and the regulatory review timeline ahead. It's built for data and analytics leaders who need a clear-eyed read on AI industry M&A, not hot takes. Key Takeaways - The $13 billion price tag is a rounding error for Nvidia, and that's the point: this was never a revenue play, it's about owning one of AI's key demand aggregation points. - Nvidia's real nightmare isn't losing this deal, it's a future where the model layer captures all the value and hardware becomes a commodity. Ross explains why that reshapes every move Jensen Huang makes. - If your team treats Hugging Face as load-bearing infrastructure, Ross has a specific take on what should change now versus what belongs on a 3 to 5 year roadmap instead. - Nvidia has a track record of vertical integration ambitions, and Ross lays out the exact scenario that would push them to acquire a frontier model company outright. Chapter Markers 00:00 Nvidia's $13 billion Hugging Face acquisition 00:58 What is Nvidia actually buying here 04:36 Aggregating supply versus aggregating demand 07:09 Hugging Face as AI's demand aggregation hub 11:44 Nvidia's nightmare scenario: hardware commoditization 14:10 Was Nvidia the only strategic buyer 20:27 Should enterprises rethink Hugging Face dependency 23:46 Regulatory review timeline and what to watch 24:58 Could Nvidia acquire a frontier model company 30:38 Nvidia's lead over AMD, Qualcomm, and Broadcom 33:23 Hugging Face's security breach and trust concerns 35:24 What We're Watching: data warehouse benchmarking findings Connect With the Show Follow Ross Katz on LinkedIn: https://www.linkedin.com/in/b-ross-katz/ CorrDyn on LinkedIn: https://www.linkedin.com/company/corrdyn/ Website: corrdyn.com Do you think Nvidia's Hugging Face acquisition changes anything about how you evaluate vendor risk in your AI stack? Tell us in the comments, we read every one. If you want to talk about your data challenges, or you think we got something wrong, find us at corrdyn.com. #EventualConsistency #DataIndustry #Nvidia #AIInfrastructure #DataStrategy

AT&T runs 45 billion AI tokens a day, and open-weight models already handle a quarter of that load. If your business is burning through a frontier-model bill without knowing which tasks actually need frontier intelligence, you're leaving cash on the table every single day. Ross Katz, Principal and Data Science Lead at CorrDyn, works with data teams on exactly this problem: deciding what to build, what to buy, and what to route. He breaks down a Wall Street Journal piece on AT&T's shift to open-weight models and open-weight AI, and tells you where that logic does and doesn't apply to a normal-sized company. Listen and you'll walk away with a practical way to decide when open-weight models are worth the engineering cost, when routing beats self-hosting, and when Chinese-origin models are a risk versus just a headline. Ross puts a number on it: if the use case isn't worth more than a million dollars a year, don't bother self-hosting. Ross and Jason get into the real difference between open-source and open-weight models, when a routing layer beats hosting your own infrastructure, and why data exfiltration (not hidden backdoors) is the actual risk with Chinese models. They also cover build-cost estimates for self-hosting, and close with a candid take on the AI bubble debate. This one's for data leaders weighing their AI spend, not for anyone just using ChatGPT subscriptions day to day. ## Key Takeaways - AT&T's chief data and AI officer expects open models to handle up to 80% of their AI workload within a few years, but Ross explains why that math doesn't automatically transfer to a smaller company. - Self-hosting an open-weight model isn't just a GPU cost. Ross puts a real number on the engineering talent required, and where the break-even point actually sits. - The Chinese-model security story isn't what most headlines suggest. Ross separates the genuine risk (data exfiltration via APIs) from the one that hasn't been proven (hidden backdoors in the weights). - There's a specific volume threshold, tied to a proprietary dataset, where a small fine-tuned open-weight model can beat a frontier model outright. Ross names the number of examples it took. ## Chapter Markers 00:00 AT&T's token numbers and the episode setup 01:05 Are open models ready for large companies 02:38 Why routing captures value other companies leave behind 07:00 Open-source versus open-weight, explained 09:54 How AT&T's smart router actually works 12:11 Shared inference versus dedicated deployment 14:43 The real cost of self-hosting a model 17:19 Where open weights clearly win today 21:35 AT&T's use of Chinese models 22:33 Is the security risk real or overblown 27:32 What smaller businesses should actually worry about 30:57 Questions every data leader should be asking now 41:51 What We're Watching: is AI a bubble ## Useful Links & Resources - Wall Street Journal: AT&T's open-model strategy (the article that prompted this episode) - CorrDyn: corrdyn.com ## Connect With the Show - Ross Katz on LinkedIn: https://www.linkedin.com/in/b-ross-katz/ - Ross Katz on X: https://x.com/brosskatz - CorrDyn on LinkedIn: https://www.linkedin.com/company/corrdyn/ Are you already routing workloads between open and closed models, or is your AI stack still one provider, one bill? Tell us how you're thinking about it. If you want to talk about your data challenges, or you think we got something wrong, find us at corrdyn.com. #EventualConsistency #DataIndustry #AIInfrastructure #OpenWeightAI #BuildVsBuy

Your data platform used to serve people. Now it has to serve agents too, and that changes what you build and who you hire. Two people ask an AI agent the same question and get two different numbers back, and neither can explain how their agent got there. That's the moment a data leader realises the platform wasn't built for this world. Ross Katz, principal and data science lead at CorrDyn, works with data teams every day as they rebuild their platforms for a world where agents, not dashboards, are the main consumer of data. You'll get a clear framework for what actually changes on a data team when agents become primary consumers, the warning signs that tell you it's time to invest in your data foundation, and a practical way to think through build versus buy versus hire. Ross also makes the case for why judgment, not production, is now the scarce resource on any data team. Ross covers the new org chart split between platform and distribution work for humans and for agents, why judgment has become the bottleneck, and how tacit knowledge gets written into semantic layers and tools like MotherDuck Guides. This is for data and analytics leaders deciding how to structure their teams around agentic AI, not for anyone after a hype pitch on AI replacing analysts. Key Takeaways - The data team's job hasn't changed: curating knowledge and adding context, but who's consuming it has, and that reshapes the platform you need to build. - One specific failure mode is coming for every data team running agents at scale, and it involves two conflicting numbers that nobody can explain. - There are three real reasons to bring in an external team instead of building in-house, and only one of them is about missing skills. - AI should raise your standard for "done", not lower it, but only if you're using it the way Ross argues you should. Chapter Markers 00:00 - Introduction: agents as the new data consumer 00:49 - How the data team's role has evolved with AI 05:11 - Why context now flows to agents, not dashboards 09:43 - Warning signs it's time to invest in your data foundation 16:11 - Build vs buy vs hire: the real decision drivers 21:38 - Designing the org chart for platform and distribution 29:22 - Why judgment is the bottleneck, not production 34:01 - Which data roles disappear, and which get created 36:29 - Turning tacit knowledge into documented context 43:22 - What We're Watching: who AI actually helps 48:09 - Recap and key takeaways Useful Links & Resources - CorrDyn: corrdyn.com - Katie Bauer's platform vs distribution framework, referenced from her work at Hex: https://hex.tech/blog/data-teams-in-ai-era/ - Otis et al. research on AI and Kenyan business owners, discussed in the What We're Watching segment: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4671369 - MotherDuck Guides, for organisation-level data documentation: https://motherduck.com/docs/category/guides/ - Previous Eventual Consistency episode with Dave Yaffe on AI governance: https://www.corrdyn.com/podcasts/eventual-consistency/ep-27-prefect-dagster-orchestration/ Connect With the Show - Ross Katz on LinkedIn: https://www.linkedin.com/in/b-ross-katz/ - Ross Katz on X: https://x.com/brosskatz - CorrDyn on LinkedIn: https://www.linkedin.com/company/corrdyn/ - Website: corrdyn.com Where does your team stand right now: are you still building platforms purely for human consumers, or have agents already started showing up in your query logs? Tell us what you're seeing. If you want to talk about your data challenges, or you think we got something wrong, find us at corrdyn.com.

Prefect just bought Dagster. Two of the biggest names in orchestration are now one company, and the reasons go well beyond a slow grower getting folded into a faster one. If you're trying to work out whether your orchestration tool is about to get sidelined, or whether "agent governance" is really just a new label on an old data stack problem, this one's for you. David Yaffe is co-founder and CEO of Estuary, a real-time data integration platform, and previously ran product at Invite Media (acquired by Google) and co-founded Arbor (acquired by LiveRamp). He's spent his career building the pipelines and infrastructure that data teams actually run on. Ross and David break down what the Prefect-Dagster deal really signals: a shift in buyer from data engineers to IT and the CIO, and a race to own agent governance before anyone else does. You'll walk away with a sharper read on where orchestration is heading, why documentation might beat MCP servers for agent-ready systems, and where the next round of data stack consolidation is likely to land. Expect a candid take on M&A positioning versus reality, the mechanics of agent permissions and observability, and a debate on MCP servers versus CLI documentation for building agent-ready infrastructure. This is for data and analytics leaders trying to separate genuine strategy from press-release spin, not for anyone after a puff piece on the deal. Key Takeaways - The Prefect-Dagster deal has less to do with orchestration and more to do with who owns agent governance and permissions once agentic workflows go mainstream. - The buyer for agent governance tooling isn't the data engineer scheduling jobs. It's the IT department, and getting those two buyers speaking the same language won't be simple. - Well-documented CLIs can outperform MCP servers for agent use, cutting token spend and doing double duty for human users too. - Treat every agent like a new graduate on day one. If your documentation, schemas, and data model aren't clear enough for a novice to follow, an agent will struggle just as much, only more expensively. Chapter Markers 00:00 Prefect's acquisition of Dagster explained 01:39 Welcome David Yaffe, Estuary co-founder 02:02 First reactions to the Prefect-Dagster deal 03:46 Why orchestration alone isn't a sticky business 07:19 Fast MCP and the rise of agent governance 09:47 MCP servers versus CLI documentation debate 12:52 Treating agents like new hires on day one 14:41 Who actually buys agent governance tools 17:40 The "aqua-pivot": why Prefect really bought Dagster 20:07 What happens to existing Dagster users 24:17 Where data industry consolidation happens next 27:27 Governance, liability and the human parallel 29:50 Time-boxed access and the future of agent permissions 31:20 What success looks like a year from now 35:07 What We're Watching: John von Neumann and the death of reading 40:14 Wrap up and final thoughts Useful Links & Resources - David Yaffe on LinkedIn: https://www.linkedin.com/in/davidyaffe - Estuary (David's company): estuary.dev - The Maniac by Benjamin Labatut, referenced in the John von Neumann discussion Connect With the Show - Ross Katz on LinkedIn: https://www.linkedin.com/in/b-ross-katz/ - Ross Katz on X: https://x.com/brosskatz - CorrDyn on LinkedIn: https://www.linkedin.com/company/corrdyn/ - corrdyn.com Got a take on the Prefect-Dagster deal, or a horror story about agent permissions gone wrong? Tell us in the comments. We're curious whether your team sees this as consolidation, or the opening move in the fight for agent governance. If you want to talk about your data challenges, or you think we got something wrong, find us at corrdyn.com. #EventualConsistency #DataInfrastructure #AIagents #DataStack #Orchestration

What happens when an AI agent's mistake can cost you the client, not just the demo. You've proven your AI agent works in a sandbox. Getting the business to trust it in production, with real customers watching, is a different problem entirely. Matthew Sekac is Head of Data & Analytics at Welocalize, where he has spent the past 18 months building an AI agent framework to handle the project management complexity behind thousands of translation and localisation projects. He joined Welocalize in 2020 and has over a decade of senior sales strategy experience in IP and life sciences translation before moving into data leadership. Matt walks through exactly how his team proved an AI agent framework could handle over a million manual tasks a year without a mountain of rules-based technical debt. You'll get a concrete method for testing agent reliability against human benchmarks before you ever touch production and a way of thinking about human-in-the-loop verification that scales instead of adding new work. This one covers ethnographic research into how project managers actually work, shadow testing agents against human output, and the evals-based thinking needed to get leadership buy-in for AI at scale. It's built for data and analytics leaders trying to move AI agents from pilot to production, not for anyone looking for a quick AI hype fix. Key Takeaways - Welocalize ran a 30-day shadow test where an AI agent mirrored every human decision on live tasks before a single customer saw it, and that proof, not confidence, is what got leadership to say yes. - The team found their "quality check" agent was inversely correlated with actual accuracy: the more confident it was that a human wouldn't override it, the more likely a human did. The reason why says a lot about where reasoning agents still fail. - Nobody asks if a human process is error-free before automating it, but everyone expects zero errors from AI. Matt argues that a mismatch is the real reason AI trust conversations go sideways. - Before writing a single rule, the team spent weeks just watching project managers work, uncovering a 56-page SOP that turned into a 70-row spreadsheet for one customer alone. Chapter Markers 00:00 Introduction: capability and trust as one problem 01:07 The agentic AI initiative at Welocalize 03:07 Why rules-based automation hit a technical debt wall 06:45 Decomposing project complexity: common cause vs special cause 10:04 The ethnographic study: watching before automating 12:32 Building the shadow test: 30 days, human vs agent 16:04 Real hiccups: parsing failures and UI friction 19:00 Where human-in-the-loop verification goes next 22:52 Benchmarking against human error, not perfection 26:07 Trust, blame, and why systems feel different to people 29:22 Gathering evidence leadership will actually believe 38:42 Building a repeatable agentic engineering practice 40:50 How AI is changing who can build automation 49:00 Evals-based development as a precondition, not an afterthought Useful Links & Resources - Matthew Sekac on LinkedIn: https://www.linkedin.com/in/matthew-sekac-8884894 - Welocalize: https://www.welocalize.com Connect With the Show - Host LinkedIn: https://www.linkedin.com/in/b-ross-katz/ - Host on X: https://x.com/brosskatz - CorrDyn on LinkedIn: https://www.linkedin.com/company/corrdyn/ - Website: https://corrdyn.com If you're wrestling with the same problem, proving an AI system's reliability before anyone will trust it with production work, we want to hear how you're approaching it. Drop a comment with what your organisation's baseline for "good enough" actually is. If you want to talk about your data challenges, or you think we got something wrong, find us at corrdyn.com.

Every major AI lab has just bought itself a consulting arm. Ross Katz explains why that is not a flex, it is an admission that the model alone does not create value. You have been told to "do something with AI" by year end, and now every systems integrator in the market wants the work. How do you tell a partner doing real engineering from one reselling a model licence with a nice deck on top? Ross Katz is principal and data science lead at CorrDyn, where he works daily with enterprise data teams trying to get real value from AI deployments. He reads these lab-consultancy tie-ups as a practitioner who actually does the integration work, not as someone selling the deal. Ross walks through why OpenAI, Anthropic and the big four consultancies are suddenly partnering, and what each side actually gets from the arrangement. You will come away with a clear framework for the layers of work that sit between signing an AI deal and getting anything useful out of it, plus a simple filter for evaluating any implementation partner pitching you. This episode covers the OpenAI and Anthropic consulting and private equity deals, the six layers between a model and real business value, and where money actually accrues in the AI stack. It is built for data and analytics leaders under pressure to show AI results, not for anyone looking for lab hype or a quick fix. Key Takeaways The labs, consultancies and PE firms should not be natural partners, but each is trading something specific: reach, integration knowledge, or portfolio intelligence. Ross breaks down exactly what each side walks away with. Getting a model into production means working through six layers beyond the model itself, and the one nobody wants to talk about is the hardest. Anthropic's own numbers suggest six dollars of services spend for every dollar of software, which tells you where the real work (and the real value) is sitting right now. Ross gives you three questions to ask any AI consultant before you sign, built around exactly how they claim to deliver value. Chapter Markers 00:00 AI labs buying into consultancies and PE 01:25 Why labs need consultancies at all 09:19 Strategic alliance or forced marriage 11:12 The six layers between deal and deployment 17:11 Where the AI money actually lands 20:13 Are consultancies funding their own disruption 22:48 The 1970s mainframe rollout parallel 26:43 Spotting real engineering vs a reseller 29:09 What We're Watching: AI labs and IPO pricing 31:56 Recap and takeaways Useful Links & Resources Ross Katz on LinkedIn: https://www.linkedin.com/in/b-ross-katz/ Ross Katz on X: https://x.com/brosskatz Previous episode referenced on Snowflake and Databricks context layers CorrDyn: corrdyn.com Connect With the Show CorrDyn on LinkedIn: https://www.linkedin.com/company/corrdyn/ Ross Katz on LinkedIn: https://www.linkedin.com/in/b-ross-katz/ Ross Katz on X: https://x.com/brosskatz Which layer is eating your AI budget right now: the data substrate, the governance, or the process and change nobody budgeted for? Tell us where your own AI rollout is stuck, and whether your consultant could actually answer Ross's three questions. If you want to talk about your data challenges, or you think we got something wrong, find us at corrdyn.com. #EventualConsistency #DataIndustry #AIDisruption #BuildVsBuy #DataInfrastructure

Snowflake and Databricks held their flagship conferences within a fortnight of each other and both independently built their entire keynotes around the same claim: the bottleneck for enterprise AI is not the model, it is the context. When two direct competitors land on identical messaging at the same moment, the right response is to check the working. That's exactly what this episode of Eventual Consistency does. Ross Katz joins Jason Bradwell to separate signal from positioning across both conferences, from Databricks' LDAP and the one copy of data promise, to the ontology race, to what Genie One actually tells us about how mature these agentic platforms really are. The through line is bigger than any single announcement: data gravity is no longer the moat it once was, and both platforms know it. The race now is to become the structured intelligence layer of your business and that changes how platform decisions should be made. The real risk for data leaders right now is not backing the wrong preview feature. It is not experimenting at all. Key topics covered >> Why the "context not model" consensus is real and manufactured at the same time >> What LDAP means for your data architecture >> Why the ontology race matters more than the feature announcements >> How to make a Snowflake vs. Databricks platform decision in 2026 >> What a mature agentic AI system actually looks like in practice About the hosts Ross Katz brings a background in analytics and data strategy, working with companies to cut through the noise and focus on what actually drives business value. With experience spanning industries such as e-commerce, education, biotech, and finance, as well as the evolving landscape of AI-enabled work, he focuses on the intersection of data capabilities and business outcomes. He's particularly interested in how shifts in technology change not just what's possible but also how people think about and use data in their daily work. Jason Bradwell is a seasoned B2B marketing leader, founder of B2B Better and hosts Pipe Dream, where he explores how modern B2B companies can build media and marketing strategies that drive real revenue and audience growth. Connect with us: Sponsor: CorrDyn, a data consultancy Connect with Ross Katz on LinkedIn Connect with Jason Bradwell LinkedIn

The Big Four accounting firms are posting more job ads for AI specialists than for auditors. Graduate intake is down 30% at KPMG and 22% at Deloitte. Equity partners are being quietly demoted. The global chairman of PwC is telling the BBC he can't find the engineers he needs. The natural read is that AI is eating audit from the inside. In Episode 10 of Eventual Consistency, Jason Bradwell and Ross Katz spend the episode pulling that narrative apart and find that the more interesting story isn't about audits going away. It's about a value chain being restructured in ways that most of the coverage is missing entirely. Ross introduces a four-phase model of the audit value chain: origination, analysis and production, judgment and synthesis, and relationship and sign-off. AI is compressing the middle (the analysis and production phases) but that compression doesn't reduce the judgment layer. As AI produces more, someone has to verify more. The nature of junior roles isn't being eliminated; it's being transformed from procedural execution toward synthesis and verification. And that transformation requires a different kind of training, a different kind of hire, and a fundamentally different expectation of what new entrants to the profession will do on day one. The conversation gets into why the talent problem the Big Four are describing (not enough AI specialists, stagnant junior salaries, and a competitive market they're not equipped to win) is partly a signalling problem as much as a hiring one. Key topics covered >> What the AI hiring numbers at the big four actually tell us and what they're being used to signal vs. what they mean operationally >> The four-phase audit value chain: origination, analysis and production, judgment and synthesis, relationship and sign-off >> Graduate intake cuts at KPMG, Deloitte, and PwC: one-time recalibration or permanent contraction? >> Whether AI is better for experts (multiplying senior leverage) or better at lifting the floor (enabling juniors to do more) and why the answer determines what the org chart looks like in five years >> The talent market reality: why big four firms are losing the AI hiring competition and what that window means for smaller firms About the hosts Ross Katz brings a background in analytics and data strategy, working with companies to cut through the noise and focus on what actually drives business value. With experience spanning industries such as e-commerce, education, biotech, and finance, as well as the evolving landscape of AI-enabled work, he focuses on the intersection of data capabilities and business outcomes. He's particularly interested in how shifts in technology change not just what's possible but also how people think about and use data in their daily work. Jason Bradwell is a seasoned B2B marketing leader, founder of B2B Better and hosts Pipe Dream, where he explores how modern B2B companies can build media and marketing strategies that drive real revenue and audience growth. Connect with us: Sponsor: CorrDyn, a data consultancy Connect with Ross Katz on LinkedIn Connect with Jason Bradwell LinkedIn

When AI can produce a ten-page analytics report, spin up a data pipeline, or generate a plausible infrastructure assessment in minutes, a question starts nagging at everyone running a professional services business: what exactly are clients still paying for? In Episode 9 of Eventual Consistency, Jason Bradwell and Ross Katz tackle that question from two different angles, Ross from the data services side, Jason from the marketing agency world. They find that the answer has almost nothing to do with AI capability, but has everything to do with three things: whether a client can tell if the work is right, whether you can recover when it goes wrong, and whether the work compounds on a foundation that actually holds. Ross introduces a framework for thinking about where AI genuinely displaces expertise and where it doesn't, with the verifiability test, the recoverability test, and the compounding test. They also dig into the trust problem that's quietly gotten harder for service providers. When a plausible-looking document can be prompted into existence in seconds, the signals clients used to rely on to assess trustworthiness such as a well-produced deliverable or a polished proposal have depreciated fast. The episode closes on a popular discussion on the build vs. buy equation in an AI-accelerated world. Why "I can build it now" is not the same as "I should build it", and whether the heavily subsidised pricing of today's AI tools represents a genuine future risk or an overblown concern. Key topics covered > What the collapsing agency pyramid means for the future of professional services > Why boutique expertise becomes more valuable, not less, as AI handles the general case > The specification cost problem: the more specific your need, the more deeply you have to engage to get AI to meet it, which is exactly where domain expertise lives > The iron triangle illusion: why AI is conditioning clients to expect fast, cheap, and good simultaneously and why the cost just shifts to the future > Value-based vs. time-and-materials pricing in an AI era: what the research says, and why business model stickiness is a real constraint > Whether subsidised AI pricing is a ticking clock or an overblown concern About the hosts Ross Katz brings a background in analytics and data strategy, working with companies to cut through the noise and focus on what actually drives business value. With experience spanning industries such as e-commerce, education, biotech, and finance, as well as the evolving landscape of AI-enabled work, he focuses on the intersection of data capabilities and business outcomes. He's particularly interested in how shifts in technology change not just what's possible, but how people think about and use data in their daily work. Jason Bradwell is a seasoned B2B marketing leader, founder of B2B Better and hosts Pipe Dream, where he explores how modern B2B companies can build media and marketing strategies that drive real revenue and audience growth. Connect with us: Sponsor: CorrDyn, a data consultancy Connect with Ross Katz on LinkedIn Connect with Jason Bradwell LinkedIn

The 2026 dbt Labs State of Analytics Engineering report surveyed 363 data practitioners (not vendors, not analysts, but the people building and maintaining data systems). The headline finding is a tension that most people working in or alongside data teams will recognise immediately: AI is now embedded in daily data work, teams are shipping more and faster, and yet trust as a stated priority jumped from 66% to 83% in a single year. At the same time, 41% of respondents still report ambiguous data ownership, 53% still cite poor data quality, and compute costs are up 50% while only 36% of teams report rising budgets. The report calls it "acceleration without stabilization." In this episode of Eventual Consistency, we use the report as a lens on a bigger question: what is AI actually doing to the relationship between data teams and the rest of the business? Ross Katz from CorrDyn, works with organizations across multiple industries as part of their external data team, offering a vantage point that most internal practitioners don't have. He unpacks why the bottleneck in most data organisations has shifted from infrastructure to accountability, and why the real challenge isn't technical. It's political. Key topics covered >> "Acceleration without stabilization": what it looks like inside organizations and why data teams bear the cost of the gap >> Why the biggest constraint in the AI era isn't time or money, it's attention >> Stakeholder management as a data team survival skill: how to navigate upward in environments that prioritise speed over foundations >> The four layers of data work; generation, integration, maintenance, governance and where AI actually helps vs. creates new demand >> The budget squeeze: compute costs up 50%, team budgets up 36% and what that asymmetry forces data teams to deprioritise >> Trust as infrastructure: when it matters most, where it can be traded off, and what dbt's framing gets right and oversimplifies >> From enablement layer to control layer: how the data team's role is shifting as AI agents enable end-runs around traditional data infrastructure >> The distinction between summarization and synthesis and why it matters for anyone building an organisational knowledge base for AI About the hosts Ross Katz brings a background in analytics and data strategy, working with companies to cut through the noise and focus on what actually drives business value. With experience spanning industries such as e-commerce, education, biotech, and finance, as well as the evolving landscape of AI-enabled work, he focuses on the intersection of data capabilities and business outcomes. He's particularly interested in how shifts in technology change not just what's possible, but how people think about and use data in their daily work. Jason Bradwell is a seasoned B2B marketing leader, founder of B2B Better and hosts Pipe Dream, where he explores how modern B2B companies can build media and marketing strategies that drive real revenue and audience growth. Connect with us: Sponsor: CorrDyn, a data consultancy Connect with Ross Katz on LinkedIn Connect with Jason Bradwell LinkedIn

The threat isn't coming from outside your perimeter. It's already inside, embedded in the open source libraries your engineers pulled in last quarter, the routers running on your network, the security tooling sitting in your CI/CD pipeline. In this episode of Eventual Consistency, host James Winegar is joined by Jamil Bou Khair, founder and CEO of Firezone, and Brian Manifold, senior full stack engineer at Firezone, to unpack what software supply chain security actually looks like in practice and why most enterprises are still building their defenses for a threat model that no longer reflects reality. The conversation is grounded in a real incident: the March 2026 Trivy vulnerability, in which attackers exploited a misconfiguration in Aqua Security's GitHub Actions environment, extracted a privileged access token, and used it to publish a malicious binary that was live in distribution channels for nearly three hours. They also discuss the hardware dimension of supply chain risk, with the US government's ban on foreign-made routers, and the reality that manufacturing on US soil doesn't solve the chip provenance problem. They dig into the tension between moving fast and maintaining operational security, drawing on Jamil's experience scaling Firezone from a startup with three engineers to a team with GitHub Enterprise security policies enforced at the repo level. The episode closes with the practical architecture of defensibility: why blast radius reduction and zero trust segmentation matter more than perimeter security in an AI-accelerated threat environment, how Firezone approaches dependency management including cooldown periods and IP allow listing across SaaS platforms, and why the industry may be approaching a moment where writing your own dependencies rather than pulling in shared libraries, becomes a legitimate security strategy. About the Host James Winegar is a data consultant who has spent years in the trenches helping enterprise organizations actually implement the technologies that vendors promise will revolutionize their business. He specializes in data infrastructure, real-time systems, and the practical realities of what works when the proof of concept becomes production. His approach is skeptical, pragmatic, and focused on the economics of technology decisions, a lens he brings to bear throughout this episode on the real costs of getting MCP wrong and what it actually takes to make AI agents useful in a production enterprise environment. About the Guests Jamil Bou Khair is the Founder and CEO of Firezone, a zero trust network access platform. Jamil brings a founder's perspective on the operational security trade-offs that real engineering teams face when moving fast in a threat environment that is moving faster. Connect with Jamil on LinkedIn: https://www.linkedin.com/in/jamilbk/ Brian Manifold is a Senior Full Stack Engineer at Firezone. Brian works across the stack on security architecture and has direct experience with the dependency management, access control, and incident response challenges discussed in this episode. Connect with Brian on LinkedIn: https://www.linkedin.com/in/brian-manifold-536a0a3a/ Connect with us: Sponsor: https://www.corrdyn.com/ a data consultancy Connect with James Winegar on LinkedIn: https://www.linkedin.com/in/james-winegar/

In March 2026, Digital Applied reported that Anthropic's Model Context Protocol had crossed 97 million installs in just 16 months, a faster adoption curve than most enterprise infrastructure standards ever achieve. OpenAI, Google, and Microsoft are all now shipping MCP-compatible tooling. What started as one company's open standard has quietly become the default interface between AI agents and the systems they work with. But is 97 million installs the moment a standard becomes infrastructure or just a very large number? In this episode of Eventual Consistency, Ross Katz sits down with James Winegar to unpack what MCP's rise actually means for the companies building on it, the enterprises adopting it, and the data leaders who are going to be asked to make decisions about it. James argues that the standardization happened much earlier than most people realize; the install count is a lagging indicator, not the inflection point. The real question now isn't whether to pay attention to MCP, but how to build for a world where it's the default. And the answer, he contends, isn't to throw AI at your raw source systems and hope for the best. The companies that will win are the ones that do the foundational work first: getting data into a warehouse, modeling it properly, and then giving AI agents a well-scaffolded, cost-efficient interface to query against, rather than burning compute calling out to Salesforce, your ERP, and five other SaaS platforms in real time and hoping the answer comes back with reasonable fidelity. The episode also gets into the tension between MCP and CLI tooling, when each is the right interface, why startups are naturally gravitating toward MCP to get to market fast, and why the security and authentication story around MCP is still catching up to what enterprise compliance actually requires. James draws a sharp distinction between MCP servers that authenticate as a privileged service account and those that can actually propagate user-level permissions and explains why that distinction matters enormously once HIPAA, SOC 2, or SEC compliance enters the conversation. Ross and James close with a provocation: by 2028, what does the world look like for a data team that treated MCP as someone else's infrastructure problem? About the Hosts Ross Katz brings a background in analytics and data strategy, working with companies to cut through the noise and focus on what actually drives business value. With experience spanning industries such as e-commerce, education, biotech, and finance, as well as the evolving landscape of AI-enabled work, he focuses on the intersection of data capabilities and business outcomes. He's particularly interested in how shifts in technology change not just what's possible but also how people think about and use data in their daily work. James Winegar is a data consultant who has spent years in the trenches helping enterprise organizations actually implement the technologies that vendors promise will revolutionize their business. He specializes in data infrastructure, real-time systems, and the practical realities of what works when the proof of concept becomes production. His approach is skeptical, pragmatic, and focused on the economics of technology decisions, a lens he brings to bear throughout this episode on the real costs of getting MCP wrong and what it actually takes to make AI agents useful in a production enterprise environment. Connect with us: Sponsor: CorrDyn, a data consultancy Connect with Ross Katz on LinkedIn Connect with James Winegar on LinkedIn

In early March, two major retailers made headlines with strategies that look like opposite bets on the future. Best Buy announced a partnership with OpenAI, opening its product catalog to ChatGPT and supporting Google's Universal Commerce Protocol, positioning itself to be discovered and purchased through AI agents, not just human browsers. Target, meanwhile, announced plans to open more than 30 new physical stores in 2026, backed by a $5 billion capital investment plan that includes expanded fulfillment, food and beverage, and same-day delivery infrastructure built in from the ground up. On the surface, these look like diverging strategies. In this episode of Eventual Consistency, we discuss the case that they're actually two versions of the same underlying bet on where data advantage lives in the next decade of retail. Ross Katz breaks down what it actually takes in the back end for Best Buy to make its product catalog legible to AI agents: clean, structured product data with consistent taxonomy, real-time inventory APIs capable of handling agent-scale request volumes, programmatic pricing that can respond without a human in the loop, and reputation data that feeds trust signals back to the foundation model companies routing purchases. It sounds straightforward. For most retailers sitting on years of technical debt, it isn't. The Target conversation goes deeper than store count. Ross argues that physical stores in the AI era aren't just retail locations, they're data collection infrastructure. The in-store experience becomes the top of a data flywheel that fuels ad targeting, brand loyalty, and fulfillment optimization. The risk for any retailer that goes all-in on agentic commerce without protecting that flywheel? Becoming a fulfillment center for someone else's platform. The episode also tackles the harder strategic question: what happens to the customer relationship when AI agents become the primary interface between buyer and brand? Who owns that relationship? The retailer, or the foundation model company routing the purchase? About the hosts Ross Katz brings a background in analytics and data strategy, working with companies to cut through the noise and focus on what actually drives business value. With experience spanning industries such as e-commerce, education, biotech, and finance, as well as the evolving landscape of AI-enabled work, he focuses on the intersection of data capabilities and business outcomes. He's particularly interested in how shifts in technology change not just what's possible, but how people think about and use data in their daily work. Jason Bradwell is a seasoned B2B marketing leader, founder of B2B Better and host Pipe Dream, where he explores how modern B2B companies can build media and marketing strategies that drive real revenue and audience growth. Connect with us: Sponsor: CorrDyn, a data consultancy Connect with Ross Katz on LinkedIn Connect with Jason Bradwell on LinkedIn

Every week brings another AI announcement, another data center project, another promise about what's possible with enough compute. But there's a constraint most people in tech aren't talking about yet - power. In this episode of Eventual Consistency, Ross Katz sits down with Mickey Peters, former energy executive and Vistage Chair, who spent decades running major power operations across South America for Duke Energy, managing over $1 billion in capital employed. The conversation starts with two converging stories: hyperscalers like Meta, Google, and Microsoft building private power generation to bypass grid constraints - one West Texas project consuming more electricity than all of Chicago - and cities like Denver hitting pause on data center development altogether. The bottleneck isn't capital. It isn't technology. It's whether the electricity exists where you need it, and whether communities will let you build there. Ross and Mickey break down how the major cloud players are each taking radically different approaches to solving the energy problem, from Meta's gas-powered megacampus in Louisiana, to Google's acquisition of an in-house renewable energy developer, to Microsoft reactivating Three Mile Island. They dig into why there's no one-size-fits-all solution, and what the real trade-offs look like between partnering with local utilities versus going behind the meter with your own generation. But the most underappreciated challenge isn't technical, it's human. Mickey draws on his experience managing energy infrastructure in remote Andean communities to explain why community trust is ultimately what makes or breaks a data center project. He unpacks what those conversations between hyperscalers, local regulators, utilities, and communities actually look like, why NIMBYism is more nuanced than a single objection, and what companies consistently get wrong when they show up to make their case. About the hosts Ross Katz brings a background in analytics and data strategy, working with companies to cut through the noise and focus on what actually drives business value. With experience spanning industries such as e-commerce, education, biotech, and finance, as well as the evolving landscape of AI-enabled work, he focuses on the intersection of data capabilities and business outcomes. He's particularly interested in how shifts in technology change not just what's possible, but how people think about and use data in their daily work. Mickey Peters is an entrepreneur and executive coach who helps leaders build stronger teams, make better decisions, and grow profitability through his Houston Vistage peer advisory group. He brings decades of international leadership experience, including 20+ years living and working in Latin America, and senior roles at Duke Energy. Connect with us: Sponsor: CorrDyn, a data consultancy Connect with Ross Katz on LinkedIn Connect with Mickey Peters LinkedIn

Nearly $300 billion in market value vanished from software companies in a single week after Anthropic’s AI agent launch reignited fears of a “SaaSpocalypse”. In this episode of Eventual Consistency , guest host Jason Bradwell turns the tables and interviews Ross Katz about what’s really happening beneath the headlines. Is AI truly threatening the seat-based SaaS model, or is this another cycle of hype? Ross breaks down the difference between market panic and operational reality, introduces eight key factors that determine which SaaS companies are vulnerable (or defensible), and explains why data gravity, regulatory moats, and revenue model alignment matter more than ever. The conversation explores how AI agents shift the build vs. buy equation, why consumption-based pricing is on the rise, and what data leaders should prioritize in 2026 and beyond. The verdict? AI won’t kill SaaS, but it will reshape it, creating clear winners and losers in the years ahead. About the hosts Ross Katz brings a background in analytics and data strategy, working with companies to cut through the noise and focus on what actually drives business value. With experience spanning industries such as e-commerce, education, biotech, and finance, as well as the evolving landscape of AI-enabled work, he focuses on the intersection of data capabilities and business outcomes. He's particularly interested in how shifts in technology change not just what's possible, but how people think about and use data in their daily work. Jason Bradwell is a seasoned B2B marketing leader, founder of B2B Better and host Pipe Dream, where he explores how modern B2B companies can build media and marketing strategies that drive real revenue and audience growth. Connect with us: Sponsor: CorrDyn, a data consultancy Connect with Ross Katz on LinkedIn Connect with Jason Bradwell LinkedIn

The era of AI experimentation without accountability is over. Following Davos 2025, where business leaders declared the end of pilot-purgatory after $1.5 trillion in AI investments, we dig into what "AI ROI" actually means for data teams. Matt Sekac from Welocalize shares how they have moved beyond demos to production AI across translation workflows, product development, and internal operations, and why the measurement challenges are more nuanced than boardroom proclamations suggest. We explore the tension between needing disciplined experimentation and executives demanding immediate returns, why live data infrastructure matters more than the models themselves, and how AI value shows up in ways beyond cost reduction. About the host Ross Katz brings a background in analytics and data strategy, working with companies to cut through the noise and focus on what actually drives business value. With experience spanning industries such as e-commerce, education, biotech, and finance, as well as the evolving landscape of AI-enabled work, he focuses on the intersection of data capabilities and business outcomes. He's particularly interested in how shifts in technology change not just what's possible, but how people think about and use data in their daily work. About the guest Matt Sekac leads R&D, the Data Office, and Data Analytics at Welocalize, where he focuses on building data-driven systems and capabilities that improve decision-making, forecasting, and commercial performance across the business. With a career that blends strategy, analytics, and operational execution, Matthew is known for turning complex data into practical tools, models, and workflows that teams can actually use. Connect with us: Sponsor: CorrDyn, a data consultancy Connect with Ross Katz on LinkedIn Connect with Matt Sekac LinkedIn

IBM is acquiring Confluent for $11 billion, and the data world is trying to figure out what it means. We cut through the press releases and LinkedIn hot takes to discuss what's actually happening here. Is this about mainframe modernization, hybrid cloud expansion, or IBM's bet on AI agents driving streaming adoption? We examine why IBM paid a 34% premium, what this means for the future of open-source Kafka, and whether the agentic commerce narrative holds water. Plus, we dig into what data leaders should actually be paying attention to as the industry consolidates around a few major platforms. About the hosts James is a data consultant who has spent years in the trenches helping enterprise organizations actually implement the technologies that vendors promise will revolutionize their business. He specializes in data infrastructure, real-time systems, and the practical realities of what works when the proof of concept becomes production. His approach is skeptical, pragmatic, and focused on the economics of technology decisions, because someone has to pay for all that streaming infrastructure. Ross brings a background in analytics and data strategy, working with companies to cut through the noise and focus on what actually drives business value. With experience spanning industries such as e-commerce, education, biotech, and finance, as well as the evolving landscape of AI-enabled work, he focuses on the intersection of data capabilities and business outcomes. He's particularly interested in how shifts in technology change not just what's possible, but how people think about and use data in their daily work. Connect with us: Sponsor: CorrDyn, a data consultancy Connect with Ross Katz on LinkedIn Connect with James Winegar on LinkedIn

In this episode of Data BS, host James Winegar and Toby Mao, CTO and co-founder of Tobiko Data, discuss the development of SQL Mesh, insights from Toby’s experiences at Netflix and Airbnb exploring the innovative features of SQL Mesh, and the ability to promote data transformations to production seamlessly. Join us for a great conversation about how Toby co-founded his company and what’s next for the future in SQL Mesh. Introduction to Tobiko and SQL Mesh Virtual Data Environments and Instant Deployments Introduction to SQL Mesh and CI/CD Integration SQL Glot: Standardizing SQL Across Platforms Type Annotations and Logical Planning in SQL Glot Future Developments and Multi-Gateway Features in SQL Mesh Connect with Toby Mao on LinkedIn This podcast is sponsored by CorrDyn