Don't Panic! It's Just Data

Companies pouring money into AI are seeing only incremental gains, and One Data Chief Executive Officer Dr Andreas Böhm says poor data is a major reason. In a recent episode of the Don't Panic. It's Just Data podcast, Böhm joined the data whisperer Scott Taylor to discuss why SAP Business Warehouse migrations can give enterprises a valuable opportunity to improve the quality and usability of their data.

Böhm, who started his first data company 20 years ago, told host Taylor that the underlying task has not changed through successive waves of big data, data science and machine learning. Organisations still need to get the right data to the algorithms. The algorithms are what have improved.

Böhm said enterprises are spending heavily on AI but are not yet seeing the return on investment (ROI). This is especially true when it comes to large jumps in efficiency, cost savings or product development. In his view, the cause is that data is not being used properly. He identified three factors that determine whether AI is adopted in practice.

• The first is trust. If finance controllers do not believe the data a system produces, for example because it does not match their own spreadsheets, they will not act on it. Böhm questioned how an AI agent could be expected to act on data that people themselves distrust.
• The second is consistency over time. Getting data right once is relatively easy, he said, but keeping quality high day to day is a continuing challenge. He compared it to asking his daughter to tidy her room: it looks fine for a day.
• The third is context. Data accurate enough for a forecast, where a five per cent swing may be acceptable, may not be good enough for a profit and loss statement or regulatory reporting to the US Securities and Exchange Commission.

Taylor noted that AI amplifies the consequences of poor-quality data. The familiar principle of “garbage in, garbage out” becomes a much larger enterprise risk when flawed data is processed and scaled across AI systems. As Böhm put it, the result is “scaled garbage.”

Böhm said SAP BW systems hold decades of financial and operational data, along with business logic buried in ABAP code, mappings and calculation views. That logic covers how sales forecasts are calculated and how controlling processes work, and the people who built it have often left the company. Simply lifting and shifting that content to a new platform carries the old problems with it, the two agreed. Taylor said it amounts to building a faster way to repeat the same mistakes.

Böhm recommended instead starting with a landscape analysis and end-to-end lineage mapping to see what is actually used. Typically, he said, about 50 per cent of legacy objects can be discarded, and reducing scope raises quality on its own. His team then uses an explainer agent to document what past logic did, so new staff and AI agents can understand it.

He described a client with 32 SAP systems, a mix of BW and ERP, that wanted to free up capital by understanding how cash flowed across the group. BW was trusted at the level of individual organisations but not company-wide. One Data translated the BW logic into a modern lakehouse architecture, then built a group-wide model on top of it. The migration took months rather than years, Böhm said, and the improvement in managed cash was visible in the company's profit and loss account.

Böhm cautioned that AI is not a shortcut around the work. Pasting ABAP code into a large language model does not capture logic spread across classes, reusable objects, hidden calls and customer exits. What has changed, he said, is that AI agents can now handle tasks such as migrating code, quality checks and fixing syntax errors, which would have been far harder four years ago. He dismissed the idea that LLMs remove the need to clean data. The funding angle matters too. Migration spending is already required, Böhm noted, so it can double as the investment in data management that organisations often struggle to secure.

Böhm said governance after migration should follow a data product approach, built on several practices:

• Ownership: Someone must understand and be responsible for the data, or it will not stay clean.
• Readability: Business users need understandable column and table names, not cryptic technical labels.
• Business validation: Checks should go beyond technical rules such as "revenue must be a number" to business rules, such as whether negative revenue is valid because of returns.
• Focus: Not everything needs governing. Priorities should be core entities such as customers, revenues, purchase orders and product lifecycles.

Taylor described governance as a lifestyle rather than a project. Böhm added that it remains a people business, in which data owners take pride in their work and promote it internally.

His closing advice to organisations preparing SAP BW data for AI was to use the migration as a chance to clean up the data landscape and to apply the budget to making at least one or two areas AI-ready. If you would like to find out more, visit onedata.ai or follow Dr Andreas Böhm on LinkedIn.

• The recurring hype cycles in AI and the importance of data quality.
• The critical role of trust, context, and governance in AI adoption.
• Transforming legacy SAP BW systems into modern data architectures.
• Strategies for cleaning up and understanding enterprise data landscapes.
• The significance of data ownership, understanding, and management in data governance.

What happens when an enterprise data agent asks you: “What is our churn this quarter?” The AI agent has access to the warehouse, the tables, and the query engine. It carries out the necessary calculations. However, the answer it provides turns out to be inaccurate. Why?

It’s not because the model is inadequate. It’s because it didn’t know which definition of the term 'churn' the business had in mind, which table was out of date, or which join resulted in customers being counted twice. When you apply this situation to all the questions your enterprise directs at AI, you begin to understand why so many agent projects come to a halt at the demo stage.

In this episode of the Don't Panic, It's Just Data podcast, host Shubhangi Dua, Podcast Producer and B2B journalist at EM360Tech, sits down with Suresh Srinivas, the CEO and Co-Founder of Collate and also a Co-Founder of the open-source project OpenMetadata. They talk about the context layer and the reasons why AI agents fail when it comes to enterprise data, even though the underlying models are continually getting better. He explains why the solution must be an open context layer rather than a proprietary one.

Srinivas cites a use case to Dua to explain how the solution can be deployed at scale. He talks about OpenAI's own internal data agent, Kepler, which is used by more than 3,500 people and carries out reasoning across about 70,000 datasets and over 600 petabytes of data, as proof that the issue is not particular to Collate's customers.

OpenAI developed a seven-layered context system based on OpenMetadata, and is one of its largest users as well as a major open source contributor. Srinivas maintains that even though there are seven layers, they all come down to three basic elements: the context of the data (what exists), the ontology and semantics (what it means, expressed in business terms such as "revenue" or "churn"), and memory (the corrections and feedback that the agent learns from).

Srinivas states that after OpenAI had invested in the context layer, query performance on Kepler decreased from 22 minutes to 90 seconds, representing a roughly 16-fold improvement, together with improvements in token efficiency.

Anthropic too has published its own research on the importance of context layers for data agents, which shows that there is convergence among the leading research labs, not merely a point made by Collate.

• McKinsey: fewer than 10 per cent of enterprises have scaled AI agents to value
• 80 per cent of enterprises cite data limitations as the scaling barrier
• On the Spider 2.0 benchmark, the top models achieved only 59 per cent accuracy when tested on enterprise data.
• Collate's context layer raised accuracy from 59 per cent to 94 per cent on the same benchmark
• That's an 86% reduction in wrong answers from context alone, not a bigger model
• Token spend fell 75 per cent once a context layer was added
• Database queries per question dropped from 190 to 27, an 86 per cent reduction in compute
• OpenAI's Kepler agent spans 70,000 datasets and 600+ petabytes, built on OpenMetadata
• OpenAI's query performance improved from 22 minutes to 90 seconds with context
• Context layer reduces to three primitives: data context, ontology/semantics, memory
• OpenMetadata has 15,000+ community members and 4,000 production deployments
• Collate curates context per persona rather than serving the full context to every agent
• Collate's AI Governance Studio audits what agents access and the risk they pose

• 00:00 Introduction to AI and Data Context Challenges
• 02:16 The Root Cause of Data Problems: Missing Context
• 04:20 Evolving Definition of Context in AI and Data
• 06:52 Why Data Limitations Still Hinder AI Scaling
• 07:22 The Impact of Context on AI Model Accuracy and Cost
• 09:21 How Providing Context Improves AI Performance
• 11:02 Verifying Data Accuracy in a Constantly Changing Enterprise
• 12:25 AI Agents and Continuous Data Quality Management
• 14:25 OpenAI's Use of Context Layers for Better AI Performance
• 16:03 Avoiding Noise and Cost in Context Management
• 18:47 The Role of Persona-Based Context Curation
• 19:45 Open Source as a Foundation for Context Layers
• 21:37 Accountability for Incorrect or Stale Context
• 22:24 Collaborative Role of Data Teams and AI Agents in Context Management
• 24:01 Guardrails and Deterministic Answers in Enterprise AI
• 25:53 Reducing Token Consumption with Context Layers
• 28:47 Key Takeaway: Building a Robust Context Layer for AI
• 30:37 Upcoming Industry Discussions and Challenges at Big Data London
• 32:28 The Future of AI Agents and the Importance of Context in Data Strategy
• 33:12 Closing Remarks and Next Steps for AI and Data Leaders

@CollateData @enterprisemanagement360

#AIagents #EnterpriseAI #DataGovernance #OpenMetadata #ContextLayer #AIaccuracy #AgenticAI #CIO #DataStrategy #EM360Tech #DontPanicItsJustData #opensourcesoftware #opensourceai

AI agents, enterprise AI, CIO, CTO, IT leadership, data governance, AI governance, context layer, OpenMetadata, data strategy, agentic AI, digital transformation, tech podcast, enterprise technology

Envision a fraud detection model embedded in a bank's AI system, continuously monitoring transactions for unusual activity. The warehouse behind it updates itself every 15 minutes; however, that seems like a short interval. A fraudulent transaction took place 13 minutes ago, and the model still hasn't picked it up.

The thing is, the model is not faulty. The data being fed to it had already overlooked the critical moment. But how?

In this episode of the Don't Panic! It's Just Data podcast, host Shubhangi Dua, Podcast Producer and B2B Tech Journalist at EM360Tech, sat down with guest Sundar Nathan, VP of Product Marketing, Enablement and Academy at Striim. They discuss the gap that exists between the time an event takes place in a production system and the time an AI agent becomes aware of it.

As the frontier models from companies such as OpenAI and Anthropic become more commonplace, the model ceases to be a source of competitive advantage for any enterprise. What remains then is the data beneath the model and the speed at which that data actually moves.

“I'd go even a step further and state that in 18 months the model you select will be less important to you than the data which is feeding that model,” Nathan said.

Alluding to his personal experience, Nathan tells Dua that his "superpower was memory,” particularly, the ability to remember circuit diagrams and organic chemistry structures as clearly as in a photograph when taking an exam. But with the introduction of Google search, that memory ceased to be an advantage since everyone had the same tool at their disposal.

Why is that experience relevant? It’s similar to the use of AI in the enterprise today. "Your competitor next door has access to the same frontier models at the same price, roughly on the same day," he says.

According to the Striim Marketing VP, what a competitor cannot rent is an enterprise’s operational data, its business insights and "the institutional knowledge accumulated over decades", in particular, how close an agent's view of production data is to the actual data.

"The time gap between the agent and where your production data is measured is the new source of competition," he adds.

Also Read Use Case: UPS Leverages Striim and Google BigQuery for AI-Secured Package Delivery

Read Use Case: The Hyper-Responsive Payments Organisation

• The model matters less than the data feeding it within 18 months, Viswanathan predicts.
• Four Cs of AI-ready data: current, consistent, consumable, contextual.
• Only 23% of enterprises scale AI agents past pilot stage (McKinsey).
• Change data capture reads transaction logs directly, in real time.
• Governance built for non-human users masks PII/PHI before agents see it.

00:00 - Why AI advantage is moving from models to real-time data

02:06 - Sundar’s path from memory-based learning to simplifying complex tech

04:10 - Why “everyone is renting the same brain”

06:52 - Why impressive pilots fail in regulated production workflows

08:30 - The four audit questions that block deployment

10:43 - Why 15-minute refreshes are not real time for fraud

13:25 - Why enterprises need a nervous system, not a warehouse scan

15:52 - How consumable data turns raw events into agent-ready context

18:29 - What Stream’s intelligent pipe does in practice

21:12 - Governance for non-human users and mission-critical guarantees

22:26 - Stop benchmarking models and start benchmarking data staleness

25:45 - Final takeaway: it really is just data, but it must move faster

Two teams pitch the same figure for ‘active customers’ right before a leadership meeting. One dashboard displays 40,000 while the other dashboard displays 52,000. No mistakes were detected by the teams; however, the difference in the stats is due to one team or employee deploying a different data filter and the other using another join path.

Now there's an AI agent involved, who is simply asked to 'retrieve the most recent number of active customers'. Not realising that there are two versions, it selects one, states that figure with complete confidence, and then goes about its business. In that moment, the semantic gap ceases to be merely an internal discussion and becomes a business risk.

On this episode of the Don't Panic! It's Just Data podcast, host Shubhangi Dua, Podcast Producer and B2B Tech Journalist at EM360Tech, is joined by Ananya Devraju, a Business Intelligence Developer at Premier Inn, to discuss the semantic gap and how AI is rendering an old data problem. They look at the situation in which metric definitions are spread across five different dashboards rather than kept in a single central location, and explain why an AI agent can produce a perfectly valid SQL query yet return a completely incorrect answer.

The problem, consistent even with AI in the picture, is governance. AI has just made the issue of governance more apparent. “It was never a data quality problem, but it's a governance of meaning problem, and it's been sitting there long before AI showed up. AI has just made it louder now,” Devraju tells Dua.

Devraju will be presenting a talk at Big Data LDN (BDL) this year on Thursday, September 24, on When Migrations Break Your Metrics: Rebuilding Data for Commercial Analytics. The talk is scheduled for 3:20 pm to 3:50 pm in the Data Architecture Modernisation Theatre.

• Semantic gaps existed before AI; AI has only made them louder and faster.
• The same metric is displayed on different dashboards, each using different date logic, filters, or join paths.
• There can be five different definitions of an "active user" across the five dashboards, all of which are incompatible.
• SQL can be syntactically correct and yet provide the wrong answer to a business question.
• Validation should be placed at the definition level, not at the SQL level.
• AI agents can query the raw staging tables rather than referring to the governed marts by name alone.
• Incorrect joints or the wrong grain result in numbers that look plausible but are actually wrong.
• According to dbt Labs research, 72 per cent of peopleprioritise AI-assisted coding while 24 per cent prioritise validation.
• Incorrect data is more dangerous than having no data; it is the confident errors that pose a risk.
• Since end users are unable to check the answers at the time they occur, trust has to be established earlier on.
• Provide a semantic layer that is governed and has version control, with designated owners for each metric.
• Ad hoc SQL statements on no side-channel should generate the "official" figures that are reported.

• 00:00 Introduction to the Semantic Gap in AI
• 02:56 Understanding Data Interpretation and Governance
• 05:56 The Importance of Centralized Definitions
• 09:05 AI Hallucinations and Data Validity
• 12:02 Navigating Governance in AI
• 14:56 The Role of AI in Data Validation
• 17:56 Final Thoughts on AI and Data Meaning

Ananya Devraju — Business Intelligence Developer, Premier Inn

Shubhangi Dua — Podcast Producer & B2B Tech Journalist, EM360Tech

Listen to more episodes of Don't Panic! It's Just Data on EM360Tech.

#SemanticGap #AIAgents #DataGovernance #DataAnalytics #SemanticLayer #AIData #BusinessIntelligence #DataEngineering #EnterpriseAI #DataQuality

While AI-generated code, also known as vibe coding, is aiding developers in releasing software faster than ever, it seems to be simultaneously causing a greater number of production failures.

According to research carried out by New Relic, 94% of technology leaders think that AI-generated code is of higher quality when it is reviewed, but 78% say they are experiencing more incidents after deployment. Let’s explore why this is happening.

In the recent episode of the Don't Panic It's Just Data podcast, host Christina Stathopoulos, Founder of Dare to Data, was joined by Nic Benders, Chief Technology Strategist at New Relic to talk about "agent debt,” what happens when AI writes great code that breaks in production, and why SREs (Site Reliability Engineers) are left cleaning it up.

Benders states that the issue does not lie with AI itself but rather with a growing phenomenon that he refers to as "agent debt".

After gathering telemetry data, New Relic discovered that the core issues stem from Large Language Models (LLMs). These problems date back to the early days of ChatGPT, when the models were remarkably convincing at generating software code.

“The code looks really good, but it doesn't work in production. It creates incidents down the road,” he tells Stathopoulos.

This is where agent debt comes in.

• AI-generated code can increase production incidents despite passing code reviews.
• 'Agent debt' is emerging as the AI-era equivalent of technical debt.
• Observability must extend beyond applications to AI development workflows.
• Engineering teams should measure AI value—not just token consumption.
• AI may deliver greater value reviewing code than writing it.
• SRE agents will compete on data quality, not model intelligence.
• Human judgement remains essential for high-risk engineering decisions.

• 00:00 Introduction to the episode and guest
• 00:30 Overview of Nick Benders and New Relic's mission
• 01:14 Impact of AI on software development speed and review process
• 02:55 Shift from coding to downstream operations and role changes
• 06:31 Discrepancy between AI-rated quality and incident reports
• 08:08 Understanding agent debt and its signs
• 09:35 The role of observability in AI-generated code in production
• 12:22 Measuring and understanding agent debt through observability
• 14:10 Rethinking the AI software development lifecycle
• 17:30 Automating code review and process checks with AI
• 20:18 Production debugging and the role of AI in incident management
• 21:09 Use case of SRE agents in production environments
• 24:00 Future trends in SRE agents and data focus
• 25:27 Key advice for using AI tools effectively in software and operations

For more information on SRE and agent debt, visit newrelic.com and em360tech.com.

#AgentDebt #AICodingAgents #AIGeneratedCode #AIObservability #SREAgents

agent debt, agent debt explained, agent debt in software development, AI-generated code failures, AI-generated code incidents, AI coding agents, AI coding agents in production, autonomous coding agents, AI code reliability, AI software reliability, AI technical debt, AI-generated software, AI-generated software observability, observability for AI-generated code, AI observability, software observability, production reliability, production incidents, AI incident management, AI-assisted incident response, SRE agents, AI SRE agents, site reliability engineering, AI code review, AI production readiness, AI developer tools, enterprise software development, LLM-generated code, large language model code, New Relic, Nic Benders

For a long time, financial crime prevention has been a massive issue for banks. This is because of several often complex navigation around rules and regulations, such as AML, KYC, CTF, and SAR. However, a bigger issue has emerged. As more and more transactions happen and financial crime networks get smarter, the bigger problem is the way banks handle data.

In the recent conversation on the Don’t Panic It’s Just Data podcast, host Herb Blecher, Research Director, Data and Analytics, Enterprise Management Associates (EMA), sat down with guest Manish Andhy, Financial Services AI & Industry Executive at Teradata.

They talked about how criminal enterprises are changing fast, like the internet, while compliance systems are still using old methods. Andhy tells Blecher, "The difference between how fast criminals can change and how slow compliance systems are is a big weakness."

More than $3 trillion in money moves through the global financial system every year, but only a small part of it is caught. Banks have spent a lot of time and money building systems to stop money laundering. Many of them still have separate systems for onboarding, transaction monitoring, investigations and risk management.

The strange thing is that big financial institutions have already spent a lot of money to solve this problem. They built data warehouses to get rid of systems and data lakes to put all the information in one place. In fact, they often created the same separate systems they were trying to get rid of.

Also Watch: What is AI Hyper-Personalisation in CX and Why Does it Matter?

• Over $3 trillion of illicit funds moves through the global financial system annually.
• Only about 1% of illicit funds are detected or intercepted.
• AML programs have grown in a fragmented, siloed way across enterprises.
• The pace of change in compliance has historically been slow.
• Traditional rule-based systems generate enormous amounts of false positive alerts.
• Data silos persist due to organizational and technology issues.
• Data products allow for a unified view of data without creating new silos.
• Generative AI can automate parts of the compliance process.
• Financial institutions must treat data as a strategic asset.
• Compliance can inform broader business decisions beyond regulatory obligations.

• 00:00 Introduction to Financial Crime and Data Analytics
• 05:06 Challenges in Financial Crime Compliance
• 10:05 Data Silos and Their Impact
• 15:12 The Concept of Data Products
• 19:57 Modernizing Financial Crime Strategies
• 24:59 Conclusion and Key Takeaways

For more information on financial crime and how financial institutions should manage their data securely and compliantly, follow Terada across its official channels:

• Website: Teradata
• YouTube: @Teradata
• LinkedIn: @Teradata
• X: @Teradata

For more information on enterprise tech analyst-led insights, please visit em360tech.com

• EM360Tech YouTube: @enterprisemanagement360
• EM360Tech LinkedIn: @EM360Tech
• EM360Tech X: @EM360Tech

#Teradata #FinancialCrime #AML #Compliance #DataStrategy #AIinBanking #FinTech #DataSilos #DataManagement #DonTPanicItsJustData #EM360Tech #B2BMarketing #B2BPodcasts

After months of experimenting with large language models (LLMs), enterprises are moving from isolated pilots to broad deployments. However, as adoption speeds up, many tech leaders find that the biggest challenge isn't selecting the right AI model but managing the cost of optimising it.

While AI models might be getting more expensive owing to geopolitical tensions and expanding use cases, the added challenge is that the price per token continues to rise. The problem is that enterprise AI is using a lot more tokens as companies scale new applications across different departments.

In the recent episode of the Don’t Panic It’s Just Data podcast, host Kevin Petrie, VP of Research at BARC, sat down with Eudald Camprubi, Co-Founder of Nuclia (acquired by Progress) and Software Fellow at Progress and Michael Marolda, Senior Product Marketing Manager, Agentic RAG.

They talked about how the next phase of enterprise AI will focus less on model selection and more on context engineering, retrieval strategies, and governance.

As enterprises progress beyond experimentation, success will rely less on picking the next game-changing model and more on using existing models wisely. That means providing precise context instead of excessive context.

Enterprises should select the right model instead of just the latest model. That means they should treat observability, retrieval strategies, and governance as foundational capabilities rather than second priorities.

Camprubí said that enterprise leaders should be wary of industry hype. "Don't trust everything you read on LinkedIn.” In enterprise AI, measurable results will determine which enterprises effectively scale intelligent systems.

Listen to the full episode of Don't Panic! It's Just Data to hear Michael Marolda and Eudald Camprubí discuss Agentic RAG, token optimisation, context engineering, and the future of enterprise AI at scale.

• Token costs are climbing, making cost management critical for enterprises.
• Context is essential for effective AI implementation and user satisfaction.
• Enterprises must focus on providing the right context to LLMs to avoid hallucinations.
• Modularity in AI platforms allows for flexibility and adaptability in solutions.
• Quality metrics are vital for evaluating AI outputs and ensuring reliability.
• The latest AI models are not always the best choice for every use case.
• Understanding user intent is crucial for effective data retrieval.
• FinOps teams are increasingly involved in managing AI costs and strategies.
• Enterprises should consider RAG as a service to reduce maintenance burdens.
• Collaboration among stakeholders is essential for successful AI implementation.

• 00:00 Introduction to AI and Data Context
• 03:26 Understanding Token Costs in AI
• 09:03 The Importance of Context in AI
• 14:37 Exploring the Context Layer and Retrieval Strategies
• 22:31 Model Selection and Cost Management in AI
• 29:11 Key Takeaways for AI Leaders

For more enterprise AI, Agentic RAG, data governance, and enterprise knowledge layer insights, follow Progress Software across its official channels:

• Website: Progress Software
• YouTube: @ProgressSW
• LinkedIn: Progress Software
• X: @ProgressSW

For more information on enterprise tech analyst-led insights, please visit em360tech.com

• EM360Tech YouTube: @enterprisemanagement360
• EM360Tech LinkedIn: @EM360Tech
• EM360Tech X: @EM360Tech

#EnterpriseAI #ContextEngineering #AgenticRAG #AIROI #TokenCosts #AIStrategy #DataManagement #DonTPanicItsJustData #ProgressSoftware #Nuclia

Most enterprises believe they have a data problem. In reality, it is an architecture problem in disguise, and the rise of agentic AI is making that distinction impossible to ignore. That is the central argument Karthik Ranganathan, CEO of Yugabyte, makes in this episode of Don’t Panic! It’s Just Data, hosted by Scott Taylor of MetaMeta Consulting.

The conversation traces 30 years of infrastructure evolution in 30 minutes. From Oracle’s dominance as the monolithic backbone of enterprise applications, to the NoSQL revolution of the mid-2000s, and the cloud-native era of the 2010s, it builds toward the rise of agentic AI. In this new phase, systems do not just store and retrieve data; they act on it autonomously.

“The challenges of current architectures under pressure are no longer theoretical. Agentic systems expose every seam, every silo, every bottleneck you've been quietly managing around,” says Ranganathan.

Taylor opens the episode by asking the question many enterprise data leaders are quietly asking: Are today’s architectures actually fit for AI workloads? Raghanathan’s answer is measured. Most are not, and the reasons are structural rather than superficial.

The core issue is fragmentation. Decades of 'good enough' tooling decisions have produced estates where relational databases sit in silos next to document stores, vector indexes live apart from transactional systems, and data pipelines patch the gaps. For traditional applications, this messiness is manageable. For agentic AI, systems that must reason across context, execute multi-step decisions, and maintain coherent state across interactions, it’s a fundamental blocker.

Ranganathan identifies three compounding failure modes: siloed knowledge stores that prevent AI systems from drawing on the full breadth of enterprise information; disconnected memory systems that can't persist context reliably across agent runs; and non-deterministic outputs from large language models (LLMs) that make it difficult to design stable data models around AI-generated results. Together, these problems don't just slow down AI projects; they erode the trust enterprises need to deploy agentic systems at any meaningful scale.

One of the episode's sharpest conceptual moves comes when Ranganathan draws a clean line between knowledge and memory in AI systems, two concepts that get conflated constantly, and at high cost.

Knowledge, in this framing, is the structured, long-term body of facts and context an AI system can draw on: product catalogues, customer histories, domain documentation, and enterprise policies. Memory, by contrast, is the short-term, session-aware state that lets an AI system track what just happened, what's been decided, and what step comes next in a workflow.

Most current architectures treat these as interchangeable or ignore memory entirely, forcing every agent interaction to start cold. The result is factually capable AI but contextually amnesiac; it knows the company's product catalogue but forgets it already recommended three options to this customer ten minutes ago.

This is the gap Meko, YugaByte's knowledge-memory engine, is designed to close. By building a unified layer that handles both the persistent knowledge graph and the operational memory of active agent sessions, Meko allows enterprises to run agentic workflows without patching together vector databases, caches, and relational stores by hand. It's an architectural bet that the knowledge-memory distinction is not a nuance but a first-class design requirement and that enterprises ignoring it will pay the integration tax repeatedly.

Taylor pushes Ranganathan on the perennial tension between AI capability and business value, a conversation that lands differently now that agentic systems are making decisions, not just recommendations. Raghanathan's view is that the critical role of context in AI workflows is still being underestimated by most enterprises.

He cites YugaByte customers who have moved from proof-of-concept AI deployments to production-grade agentic systems by making one architectural shift, treating context as infrastructure, not as application logic. When context management is embedded in the data layer, versioned, auditable, and available across agent boundaries, the reliability bar for AI systems rises dramatically.

For enterprises ready to act, Ranganathan’s guidance is clear: start by auditing existing architectures for knowledge and memory silos before evaluating AI tooling, invest in unified data models that support both relational and non-relational workloads, and treat the context layer as a core engineering concern rather than an afterthought added to LLM integrations. If you would like to find out more, visit yugabyte.com or connect with Karthik Ranganathan on LinkedIn.

• Evolution of data infrastructure for agentic AI.
• Limitations of current architectures and silos.
• The role of knowledge and memory in AI systems.
• Strategies for enterprise data architecture in the AI era.

00:00 Introduction to Data and AI Infrastructure

03:02 The Evolution of Data Infrastructure

05:53 Understanding Agentic Applications

08:55 The Architecture War in Data Management

11:58 Knowledge vs. Memory in AI

15:00 Operational Challenges in Multi-Agent Environments

17:58 The Importance of Context in AI Workflows

20:49 Bridging the Gap Between AI and Business Value

24:10 Customer Success Stories with YugaByte

26:47 Rethinking Data Architecture for Enterprises

AI investment is growing fast, but proving its value remains one of the biggest challenges facing data leaders today. Dashboards are built, models are deployed, and yet when the budget question arrives, most teams still can't clearly demonstrate return on investment.

Speaking on Don't Panic, It's Just Data with host Christina Stathopoulos, Nadiem von Heydebrand, CEO and co-founder of Mindfuel, identified where most organisations go wrong: the interface between data teams and the business. According to von Heydebrand, the reason is straightforward: no use case, no value.

"We get a demand, we believe we've understood it, and we start executing immediately," he explained. Months pass, and nobody can answer why the project exists or what problem it was supposed to solve in the first place. The fix isn't more technology. It's better use case management.

One of von Heydebrand’s core principles is straightforward: before you build anything, you need to really understand the business challenge you're trying to solve. "You have to fall in love with the problem, not with the solution," he said. This matters more than ever in the era of generative AI. With token costs attached to every AI interaction, building the wrong solution isn't just a wasted effort; it's an ongoing financial drain. Use case management has moved from being a nice-to-have to an operational necessity. Good use case management, according to Nadiem, rests on three pillars:

1. Demand exploration: Don't assume you understand the problem. Engage stakeholders, ask deeper questions, and uncover the real business challenge before a single line of code is written.
2. Value management: Every use case needs a value hypothesis. What outcome is expected if this problem is solved? As Nadiem puts it: "The solution itself has a value of zero. Value lives in the problem space."
3. Value tracking: Once live, track performance against the original hypothesis. Define a realistic ROI timeframe and review it consistently.

One of the most common mistakes? Measuring AI success through usage and adoption data alone. "I have enough examples where usage is high, and value is zero or even negative," von Heydebrand warned.

Clicks and logins are a proxy. Business outcomes are the goal. If there's no correlation between the two, the metric is misleading.

The most important distinction in the conversation was the difference between output and outcome. Data teams have historically been measured on output like model accuracy, number of dashboards, and features delivered. But output without impact is just activity. Outcome means the value created for the recipient of your work. Organisations that make this mindset shift from measuring what they produce to measuring what they change are the ones that change their data functions from cost centres into genuine value generators.

For leaders under pressure to prove ROI from AI initiatives, Mindfuel’s CEO advises a pragmatic approach: start now, start small, and be honest. As Stathopoulos summarised: "It all comes back to being intentional about what you build and why." For more information, visit mindfuel.ai, the platform built to help data and AI teams demonstrate, manage, and maximise business value.

Connect with the guest:

Nadiem von Heydebrand: LinkedIn | Mindfuel

• The importance of structured use case management
• Linking AI initiatives to business value
• The impact layer and value tracking in AI projects

00:00 – Introduction to Data and AI Impact Management

03:16 – The Challenge of Connecting AI to Business Outcomes

11:38 – Understanding Use Case Management

17:40 – The Missing Value Layer in Data and AI Initiatives

22:23 – Evolving Mindsets in Data and AI

27:36 – Advice for Leaders on Proving AI ROI

Podcast: Don’t Panic! It’s Just Data

Guest: Michael Marolda, Senior Product Marketing Manager for Agentic RAG at Progress Software

Host: Shubhangi Dua, Podcast Producer and B2B Tech Journalist at EM360Tech

Generative AI has been brewing in the enterprise tech industry for at least three years now. AI pilots are launching every other day, internal copilots are deployed across enterprise divisions, and now teams themselves are experimenting with large language models (LLMs) to automate business workflows. Such additions have sped up research and notably improved productivity.

While the excitement is valid, the truth beneath is often disregarded. Many enterprise AI systems produce answers that sound convincing, even when they are completely wrong.

In the recent episode of the Don’t Panic! It’s Just Data podcast, Michael Marolda, Senior Product Marketing Manager for Agentic RAG at Progress Software, sat down with host Shubhangi Dua, Podcast Producer and B2B Tech Journalist at EM360Tech.

Marolda argued that the problem is not necessarily with the AI models themselves. The real issue is with the enterprise data foundations supporting them.

“Your AI is only as good as the knowledge it has access to,” Marolda explained during the conversation. The question is what the gap is alluded to in the AI enterprise tech space.

Also Read: Build vs. Buy: The Reality of Production-Grade RAG

According to Marolda, around 80 per cent of enterprise data remains unstructured. This includes PDFs, contracts, emails, audio files, presentations, scanned documents, videos, and handwritten notes. This is the kind of information that traditional AI systems struggle to process reliably.

While enterprises are heavily investing in AI infrastructure and model testing, many still do not have systems capable of organising, retrieving, and validating this scattered knowledge. The outcome often turns into a situation where AI tools begin to generate responses without the necessary business context, despite excellent prompt engineering.

“We’ve seen enterprises rush into AI implementations,” Marolda said. “But many pilots fail to scale because the information isn’t grounded in real business data.” It ultimately poses major operational risks for companies, especially in highly regulated industries.

During the podcast, Marolda mentioned a high-profile case involving an airline chatbot that provided customers with incorrect policy information, leading to legal consequences for the company. The issue was not due to malicious intent or a technical failure at the model level — it was due to unreliable data grounding.

For enterprises using AI in customer service, HR, legal operations, finance, or internal knowledge systems, such errors are not rare. In fact, they’ve become a governance issue.

Enterprises tend to rely on data lakes as centralised storage for vast amounts of information. However, Marolda makes a point about how storage is no longer enough in the age of AI. “A data lake is just cheap storage,” he explained. “A knowledge layer is what actually activates that information for AI.”

This difference is increasingly important as enterprises move from testing to operational AI deployment. Traditional storage systems can hold documents, but they cannot interpret relationships between data points, retrieve context semantically, or validate AI-generated outputs against source material.

An enterprise knowledge layer, on the other hand, is designed to fill that gap. Marolda tells Dua that modern retrieval-augmented generation (RAG) systems can process unstructured data, apply optical character recognition (OCR), convert speech to text from video and audio, and build semantic connections across enterprise content.

This enables AI systems to retrieve not just documents, but highly specific pieces of contextual information, including paragraph-level citations and timestamped video references.

For enterprise leaders, the implications are significant. Rather than viewing AI as a separate assistant, enterprises are increasingly seeing AI as a retrieval and reasoning layer built on top of their knowledge ecosystems.

The economics of AI was a critical discussion Marolda had with Dua. He noted that while many AI providers continue to push for higher token consumption and larger workloads, enterprises such as Progress Software are now beginning to value efficiency instead.

Unlike NVIDIA’s enterprise philosophy, as proposed by its CEO Jensen Huang, is a new compensation model where engineers receive annual AI token budgets worth half their base salary on top of regular pay. During a live interview on the All-In Podcast, recorded in San Jose, California, in March 2026 at Nvidia's GPU Technology Conference (GTC), Huang stated:

"If a $500,000 Engineer Did Not Consume At Least $250,000 Worth of Tokens, I'm Going To Be Deeply Alarmed."

“We’re actually trying to reduce token consumption,” he explained. Such an approach contrasts with broader industry trends focused on maximising AI use at scale. As enterprise AI budgets become more established, CIOs and CFOs are scrutinising infrastructure costs, energy consumption, and long-term operational sustainability.

It’s particularly relevant as enterprises pit multiple LLMs against each other for quality, relevance, and cost efficiency. According to Progress’s Sr. Product Marketing Manager, the next phase of enterprise AI adoption won’t be driven by model capability alone. It will be guided by practical governance, meaning identifying which systems produce the best results at reasonable costs.

Overall, successful AI adoption is not just about selecting the right model but, in fact, pivoting towards building the right knowledge architecture.

For instance, enterprises continue to invest in generative AI; the enterprises that thrive may be the ones that can effectively structure, govern, retrieve, and validate their institutional knowledge.

• Enterprise AI hallucinations increase without grounded enterprise data.
• Agentic RAG helps enterprises reduce AI hallucinations and improve accuracy.
• Unstructured data is the biggest challenge in enterprise AI adoption.
• Enterprise knowledge layers improve AI governance and traceability.
• AI token reduction lowers enterprise AI infrastructure costs.
• RAG architecture helps enterprises scale trustworthy AI systems.

• 00:00 Introduction to Enterprise AI and Knowledge Layer
• 02:13 Challenges with Unstructured Data in AI
• 08:11 The Importance of a Knowledge Layer
• 12:04 Trust and Governance in AI Solutions
• 16:48 Progress's Unique Approach to AI Solutions
• 19:15 Agentic RAG: A New Paradigm in AI Retrieval
• 24:52 Real-World Applications of Agentic RAG
• 26:39 Maintaining Quality and Performance in AI Systems
• 28:01 Key Takeaways for IT Decision Makers

For more enterprise AI, Agentic RAG, data governance, and enterprise knowledge layer insights, follow Progress Software across its official channels:

• Website: Progress Software
• YouTube: @ProgressSW
• LinkedIn: Progress Software
• X: @ProgressSW

For more information on enterprise tech analyst-led insights, please visit em360tech.com

• EM360Tech YouTube: @enterprisemanagement360
• EM360Tech LinkedIn: @EM360Tech
• EM360Tech X: @EM360Tech

#AgenticRAG #AIGovernance #EnterpriseAI #GenAI #RAGSystems #UnstructuredData #KnowledgeLayer #DataRetrieval #AIHallucinations #ProgressSoftware #TechPodcast #B2BTech #CIOInsights #LLM #AICostReduction #EnterpriseSearch #AIStrategy #DontPanicItsJustData #MichaelMarolda #ShubhangiDua #TechNews