What do CTOs hate most about GenAI? Tool changes that break stuff
With so many DB vendors to choose from, our vulture claws over the bewildering choices
DataStax recently joined a growing band of database specialists in launching new tooling with the promise of helping customers build GenAI apps on its data platform. Yet the question remains for customers employing multiple databases: Which vendor – if any – should they choose as the main plank to support GenAI application development?
Based around the open source Cassandra database, DataStax has a customer list that includes the likes of CapitalOne, Condé Nast, and Saab. It bought Langflow, the open source visual framework for building AI applications, in April and is now previewing Langflow in its Astra Cloud platform, promising developers a way of quickly designing and piloting GenAI and retrieval-augmented generation (RAG), the popular approach for enhancing LLMs with enterprise data.
DataStax is promising a cloud-hosted, drag-and-drop visual interface for LangChain. Separate product LangSmith meanwhile offers integrations with GenAI tools including OpenAI, Hugging Face, document database MongoDB, and vector database Pinecone. It also offers RAGStack 1.0, a system designed to streamline RAG implementation at enterprise scale.
Speaking to The Register, Jason McClelland, DataStax chief marketing officer, said organizations were finding it hard to build an AI application because the tooling – products like LangChain, LangSmith, and Unstructured.io – are new to the market and rapidly iterating.
"The CTO's frustration is that they have got all these developers spending all their time building the infrastructure as opposed to building the app," he said. "Once they get something working two days later, it breaks because one of the underlying tools has changed again. Our premise then is to work with those partners create a version of their stuff and integrate it with us."
McClelland claims the DataStax AI platform would provide integrated tooling for data preparation, vector embedding, data ingestion, and data chunking – dividing data into smaller more manageable chunks.
However, DataStax is not the only database vendor trying to tempt users to base their development of AI applications on its platform. Fellow NoSQL database vendors Couchbase and MongoDB both offer tooling for developing AI applications, as does Oracle.
The problem is, most large enterprises rely a number of different databases, and some have fundamentally different architectures to others. So why base an AI on any one rather than the others?
- Oracle early leader in pointing vectors at business data, say analysts
- AI query optimization in IBM's Db2 shows you can teach a tech dinosaur new tricks
- From RAGs to riches: A practical guide to making your local AI chatbot smarter
- Oracle Ads have had it: $2B operation shuts down after dwindling to $300M
Carl Olofson, IDC research vice president, said DataStax has produced a facility that makes RAG much faster and easier than is normally the case.
"A reasonable question to ask is, does this enhance, or make stickier, the use of Cassandra, or since LLMs and vectors are normally aimed at unstructured data, does this represent an alternative sales track for DataStax that does not depend on the user actually adopting Cassandra for data management?" he said.
Matt Aslett, director at tech advisory and research company ISG, said database vendors were justified in adding genAI development tooling to their platforms because that's where much of the enterprise data resides.
However, he cautioned: "The core underlying trends are not going to change because of generative AI. Enterprises used multiple databases for multiple types of applications and end use cases. That's going to continue to be the case with AI."
An early or pilot project starting with an existing data platform might make sense, particularly if that's where the skills are.
"But if you accept that a company is going to have multiple data platforms and you accept they're making a heavy investment in AI, then the likelihood is they've got a separate team doing the AI element data science, perhaps AI-specific platforms and tooling, in combination with the databases," he said.
He said organizations looking to tweak existing applications with AI might base that on technology from their database vendors, but those looking to build new applications that exploit GenAI might look for a development platform independent of an individual database vendor. ®