SAP's Data + AI Play: Why Your Architecture Just Changed

SAP's acquisition of Dremio and Prior Labs signals a major shift in enterprise AI: the "AI on top of data" era is over. In this episode, we break down why your data architecture is now your AI strategy and what this means for builders. We also cover Aircall's acquisition of Vogent, which brings production-grade voice agents into the mainstream, and the growing business risk from AI-enhanced identity fraud. For a practical takeaway, we walk through how to experiment with a voice agent in your own workflow in under an hour. Follow No-BS AI Briefing for concise analysis that helps you build smarter.

Transcript

SPEAKER_00 0:00

Today on No BS AI Briefing, SAP's massive double acquisition signals the end of the AI on top of data era. We'll also cover why production grade voice agents are finally here, how AI identity fraud is becoming a major business risk, and what Google's latest moves with Gemini mean for builders. Let's dig in. No BS AI Briefing brought to you by Proactive AI. Welcome back. I'm your host, Vikash, and this is where builders get straightforward AI news without the fluff. We've got a lot of ground to cover today, mostly centered around a big theme, consolidation and integration. The era of just playing with a generic chatbot API is closing, and the era of deeply integrated vertical AI is here. First up, in a move that validates the voice AI space, AirCall has acquired Vogent. In plain English, this means AirCall, which is a business phone platform, is bringing a sophisticated AI voice agent natively into its product. This isn't about some clunky third-party integration you have to stitch together. We're talking about production grade capabilities like human like turntaking, low latency responses, and even custom voice modeling, all built right in. For builders, this is a huge signal. Voice has been a really tough nut to crack, but this acquisition shows that the demand is real and the tech is finally getting good enough for serious business use cases like sales qualification, support triage, and appointment booking. It also shows that for many AI startups, the winning move isn't to be a standalone tool, but to get acquired and become a core feature of a larger platform. The value accrues when you solve a vertical problem natively. Next, in the world of infrastructure, Lattice Semiconductor acquired AMI for a hefty $1.65 billion. Now, this might sound like a niche hardware story, but it points to a really important shift in the AI stack. Lattice makes FPGAs, that's a type of programmable chip, and AMI makes infrastructure management software. Putting them together creates a vendor agnostic way to control and observe everything happening inside an AI data center. Think real-time telemetry on power consumption, thermal management and utilization, not just for GPUs but for CPUs and any other custom chips you're running. So why does this matter for you? It tells us the bottleneck in scaling AI is moving. It's not just about getting more raw compute anymore, it's about observability and control. As you build and deploy more complex models, managing the AI factory efficiently becomes the real challenge. Neutral full stack platforms that can wrangle hardware from different vendors are becoming absolutely essential for serious enterprise AI operations. And speaking of enterprise, our third headline is a big one and it's our deep dive for today. SAP has acquired not one but two companies, Dremio and Prior Labs. Dremeo is a data lakehouse company that helps you access data across different systems, SAP and non-SAP alike, without a ton of painful ETL. Prior Labs is an AI startup that builds foundation models specifically for tabular or structured data. The goal here is pretty clear. SAP wants to let its AI models understand the messy structured business data sitting in its customer systems natively in real time. For builders, this is probably the most important strategic signal of the week. This isn't about AI on top of data anymore. This is about co-designing your data architecture with your AI. The idea that you can just point a generic LLM at a giant enterprise database and get magic is officially dead. The real value is in unifying the data layer and the model layer. We also have a flurry of updates from Google which is expanding Gemini across work, health, and even defense. The company announced Gemma 4, its latest open reasoning model, and a new tool called Deep Research Max, which is positioned as an autonomous research agent for enterprise workflows. That's a big step up from just a chat interface. They're also expanding memory and personalization features in Gemini. But it's not just about the tech, it's about deployment. Google announced a partnership with the US Department of Defense, a $10 million commitment to train healthcare workers in rural areas and deeper Gemini integrations into consumer products like Fitbit and Translate. For builders, this shows Google is fighting a war on multiple fronts. They're playing the open source game with Gemma, building high-end agentic tools for enterprise, and simultaneously pushing into heavily regulated real-world sectors. It's a reminder that a winning AI strategy isn't just about having the best model, it's about having the best distribution and the most diverse set of applications. And finally, a story that should be a wake-up call for anyone building a product that handles user identity or agreements. AI-powered identity fraud is emerging as a major business risk. A new report from Lumen found that a staggering 55% of businesses in New Zealand have been targeted by AI-enhanced identity fraud, and 90% of them are worried about vulnerabilities in their digital agreement workflows. I mean, think about that. The very tools we are building can be turned against us to create hyper-realistic fake IDs, deep fake videos for verification, or sophisticated phishing attacks. As a result, businesses are cranking up their spending on AI-powered identity verification tools by 67%. The takeaway for builders is simple and direct. Fraud detection and identity verification are no longer optional, nice to have features. They are becoming core non-negotiable product requirements. If your product touches documents, contracts, money, or personal identity, you need to be thinking about your defense against AI-driven attacks right now. Alright, let's do our deep dive on the SAP story because it's the one that I think carries the most weight for anyone building a business in the AI space. The headline is that SAP bought two companies, Dremeo and Prior Labs. So what really happened here? SAP, one of the biggest enterprise software companies in the world, didn't just buy an AI company. They bought a data architecture company, DreMio, and a specialized AI model company, Prior Labs, at the exact same time. Dremyo's whole thing is making it easy to access and query data that's scattered all over the place in different formats without having to go through a long, expensive process of moving and transforming it, known as ETL. Prior Labs, on the other hand, focuses on a very specific, very hard problem, making AI models that are actually good at understanding structured tabular data. We're talking about the rows and columns you find in a spreadsheet, a CRM, or an ERP system. And why does this matter so much right now? Because it signals a fundamental shift in how the enterprise world is thinking about AI. For the last couple of years, the hype has been all about general purpose large language models. The dream was that you could just point a model like GPT-4 at your company's data and ask it anything. But anyone who has actually tried this in a real business setting knows the reality. These models often struggle with the rigid logic of tabular data. They hallucinate numbers, they misinterpret columns, and they can't reliably perform the kind of multi-step analysis that businesses actually need. SAP's move is a massive admission of this reality. They're saying that the AI on top of data approach is a dead end. The future is co-designing the data layer and the model layer to work together seamlessly. So who should really be paying attention to this? Well, if you're a founder of a vertical SaaS company, this is a massive validation of your core thesis, your defensibility, your moat isn't just your workflow software anymore. It's your deep understanding of your customers' specific data structures and your ability to build specialized models that can reason over them. This is how you beat the generic players. If you're a product manager, this means you need to stop thinking about AI as a feature you can just bolt on. The question is no longer how can we add a chatbot? The question is how does our underlying data architecture need to change to enable truly intelligent, reliable features? And for you infrastructure folks, this move validates the modern data stack. Things like the lakehouse architecture that Dremio champions. But it adds a twist. That stack is no longer just for analytics. It's now the foundation of your company's AI. The one model fits all idea is fading fast. So, how would I think about this as a builder? I'd use an analogy. Imagine your company's data is a massive library. The old approach was like hiring a very eloquent, very charismatic person who has read a lot of books in general but has never been in your library before. You ask them a complex question and they wander around, pull a few books off the shelves, and give you a confident sounding but probably wrong answer. SAP's new approach is different. With Dremio, they're hiring a master librarian who knows exactly where every single book, journal, and financial record is located. They can get you the right information instantly. And with the prior labs, they're hiring a domain expert, a forensic accountant who can actually read those dense financial ledgers and give you a precise, reliable, and auditable answer. You need both. Your data architecture is the library's organization system and your model is the specialist reader. One without the other is useless. My Nobia's take on this is pretty straightforward. This is a smart, strategic, and very long-term move by SAP. Don't expect to see some magical new product from them next quarter. Integrating these companies will be complex and tabular foundation models are still a relatively new field. But the strategy itself is spot on. It confirms that the future of enterprise AI isn't the chat interface, it's the unsexy, deep in the stack work of unifying data access and specialized intelligence. For startups, the message is clear. Own the data layer. That's where the real defensible value is. If you're finding this useful, hit follow in your podcast app right now. It takes two seconds and it's the best way to make sure you don't miss the next briefing. Alright, if you want one practical takeaway from today's episode, here it is. Experiment with deploying a production grade voice agent into a simple customer workflow. The AirCall and Vogent acquisition shows that this technology is crossing the chasm from cool demo to real business tool, and the only way to build intuition for what it can actually do for your product is to get your hands dirty. Your goal here isn't to build a fully fledged perfect system in an hour, it's to experience the technology and spark a strategic conversation with your team. Here's how you can try this in under 60 minutes. First, sign up for a service that offers a voice agent API. You could check out AirCall's new Vogent integration or look at other platforms in this space. The key is to find one that lets you set up a simple agent without a ton of custom code. Second, pick one and only one hyperspecific repetitive task. Don't try to boil the ocean. A perfect example would be qualifying a new inbound sales lead with two or three basic questions or triaging a customer support call by asking, is this about billing, a technical issue, or something else? Third, script out that very simple conversation flow and set it up in the tool. Then call the number yourself. Record the call, have a teammate call it, just listen, how is the latency? Does the turntaking feel natural or does it feel like a clunky old IVR system? How does it handle an unexpected question? The reason this specific experiment is so valuable is that voice is a fundamentally different interface from text or clicks. It's emotional, it's high bandwidth, and it's deeply personal. Hearing a fluid responsive AI agent handle a task that used to be handled by a clunky phone tree or a human can trigger a whole new set of ideas. It forces you to ask, what if this was cheap, reliable, and infinitely scalable? How would that change our customer onboarding, our support model, our sales process? That's a conversation worth having. And it all starts with one 60 minute experiment. That's it for today's No BS AI briefing. If this helped, follow the show in your podcast app and share it with one builder you know. And if you've got questions or topics you want covered, connect with me on LinkedIn and send them over. See you in the next briefing.