Every company is eager to adopt AI, and with good reason. An organization’s AI strategy now typically carries boardroom-level significance and holds the potential to reimagine and improve productivity. With the advent of Large Language Models (LLMs) and other developments, such as agentic AI, everyone is eager to experiment and discover how generative AI can deliver returns for their businesses.
However, the desire for change is not the same as making it happen. Wanting to adopt AI doesn’t make it easy to adopt AI.
As data engineers and developers encounter push from experimentation towards implementation, this barrier becomes more concrete, and the need to implement AI quickly increases.
What’s the best way to remove these blockages? Let’s take a look.
Blockers, blockers everywhere
The first thing to realize is that blockers to technological adoption don’t come from technology alone. Often, they also come from the organization itself. In fact, when properly unpacked, many AI blockers are actually process issues inherent to the business at an organizational level.
Other blockers are technical in nature. To successfully adopt AI, you must address both the business problem and the technical challenge at the same time.
Making AI easy to implement is now a primary consideration
Rapid AI adoption requires more than access to your data. It requires a flexible, ready-to-go development platform that recognizes SQL as the lingua franca of data.
In this article, we’ll examine the roadblocks that companies encounter when implementing AI and how AI Workflows in Starburst Galaxy address each pain point, faster AI adoption.
AI is no good without the data
Companies face numerous challenges in pivoting to AI. For many, the first challenge involves unlocking the unstructured data currently languishing in data silos across the enterprise.
Historically, this has triggered a lengthy process of centralizing everything into a single repository. But this often doesn’t work as expected. In fact, all this usually accomplishes is delaying the start of AI initiatives by months (if they’re lucky).
Luckily, there is another approach that doesn’t involve default data centralization.
Starburst makes adopting AI achievable within real organizations
Enter Starburst Galaxy, the data lakehouse platform specifically designed to accelerate AI adoption and rapid deployment of new AI agents. Starburst addresses this pain point by providing a unified and flexible approach to data access, collaboration, and governance. Its Icehouse architecture combines the super-fast distributed SQL query engine Trino with the Apache Iceberg open table format.
Unlocking data access and governance, one source at a time
Using Starburst, you can leverage its 50+ data connectors to easily bring your AI lakeside, unlocking unstructured data currently trapped within data siloes. For other types of unstructured content—such as PDFs, call transcripts, or images—Starburst works alongside upstream processing tools that extract, parse, or embed that data before it is stored in Iceberg for querying and governance. Overall, this approach delivers the best of both worlds: a fast and efficient centralization in Iceberg, combined with super-fast retrieval of decentralized data via Trino.
Managing data is the key to managing AI
From speaking with our customers, we know that data management alone is insufficient. To address this, we’ve identified three key patterns that emerge when business and technology challenges necessitate new tools to accelerate AI adoption.
Don’t adopt a one-size-fits-all approach. Different teams may have various levels of AI empowerment within a company. Adopting a diverse approach enables early adopters to drive innovation, with other teams following suit when it makes sense.
The art of innovating locally
Some teams act as localized innovators. When considering AI, start with them. They usually possess a high degree of business and technical acumen, making them highly attuned to the organization’s needs. They may be subject to central oversight, but also have a significant degree of autonomy to help the company navigate the tech industry’s innovative frontiers.
When it comes to AI, these teams require fast access to data across their data estate, both centralized in a lakehouse and distributed on the edges, with minimal data manipulation. They aim to conduct experiments as quickly as possible to rapidly identify which AI solutions will produce the most significant business value.
Driving AI with SQL
Another group of innovators is the SQL innovators. This group is typically made up of software engineers, data analysts, and others, conversant in SQL. They are tasked with building branded AI solutions in collaboration with data science or AI teams.
Given its historic utility as a data-crunching language, it’s no surprise that Python has emerged as a language of choice for many AI agent developers. However, not everyone wants to spend the time learning yet another programming language. These innovators prefer to work with what they know and deliver results quickly.
Working inside pre-existing processes
Finally, we have teams with locked-down approval processes. These teams typically operate in highly regulated industries, handling sensitive data such as financial and healthcare information. Approving a new AI solution typically requires months or quarters.
These teams don’t want a fully packaged solution with embedded AI capabilities. They require a solution for AI agent development that seamlessly integrates with their existing approved AI capabilities and LLMs. An open platform can accelerate the approval process and enable teams to ship more quickly than solutions with tightly coupled AI.
How Starburst Galaxy enables rapid AI innovation
Starburst Galaxy is the data lakehouse solution that enables rapid innovation of new AI solutions. Its AI Workflows is a suite of capabilities designed to accelerate AI by making governed, proprietary data instantly usable.
AI Workflows operate natively and at scale in SQL. That means your data experts don’t have to learn a new language from scratch to get started. Instead, they can start prototyping, evaluating, and deploying new AI solutions today.
Starburst AI SQL Functions: AI for developers
AI SQL Functions in Starburst Galaxy enable anyone conversant in SQL to unlock the power of generative AI. Using this approach, developers can leverage built-in functions to perform sentiment analysis, classify and mask sensitive data, correct grammar, translate between languages, and send prompts to an LLM with supplied data as context.
For example, using the ai.analyze_sentiment function, a developer can use their choice of LLM to return a rating – positive, negative, or neutral – of a user’s disposition towards a product:
SELECT ai.analyze_sentiment('I love Starburst', 'bedrock_claude35'); -- positive
In the call above, the function takes two arguments:
- The text to analyze, which is hard-coded here but can be any field name or SQL expression.
- The LLM that you wish to use. Starburst supports OpenAI-supported models as well as any of the foundational models hosted on Amazon Bedrock.
Why Starburst AI Model Access Management is fully-managed and easy to govern
Starburst Galaxy is a fully-managed cloud solution, incurring minimal operational overhead. Teams can create users, connect data sources to catalogs, and start querying data immediately.
For organizations with high security requirements, Starburst’s AI Model Access Management. This enables organizations to restrict who can access specific models. It can also be used to restrict models from accessing sensitive data, such as personally identifiable information (PII).
Which models does Starburst support?
Currently, Starburst supports GPT-4o on OpenAI, Anthropic Claude on Amazon Bedrock, Llama3 hosted on-premises behind Ollama or vLLM, and any other model that’s compatible with the OpenAI API.
The list is continually growing as new models become available and are added.
Starburst AI Search provides built-in support for RAG
Retrieval-Augmented Generation (RAG) is a method for supplying an LLM with authoritative knowledge outside of its training data sources. The approach is indispensable for generating accurate answers based on external or proprietary knowledge.
Example of Starburst AI Search
Thanks to its nature as a SQL query engine, Starburst Galaxy supports a highly accurate RAG implementation. Developers can use the ai.generate_embedding function to turn data into a vector embedding, a numerical representation of data used in the vector storage format at the heart of context retrieval in RAG:
SELECT ai.generate_embedding('Which chapter should I read to understand how to balance the weight of a Boeing 747?', 'bedrock_titan'); -- [ 0.0061195422895252705, 0.013783863745629787, ...]
You can store these in an Apache Iceberg table. When a user asks a question, you can convert that question to a vector embedding and search the Iceberg table using the cosine_similarity() SQL function:
WITH vector_search AS( SELECT book_title, chapter_number, chapter_title, chapter_intro, cosine_similarity( generate_embedding('Which chapter should I read to understand how to balance the weight of a Boeing 747?', 'bedrock_titan'), chapter_intro_embeddings) AS similarity_score FROM iceberg.example-schema.faa_book_chapters ORDER BY similarity_score DESC LIMIT 5 )
One issue with RAG is that, because it uses approximate search methods instead of exact matches, it may return results that are irrelevant to the query. Starburst Galaxy utilizes a combination of exact nearest neighbor search and filtering to narrow down the result set to the most relevant documents. That translates into faster and more accurate answers from LLMs.
How Starburst helps you achieve results with AI
Rapid AI adoption requires both a solid foundational data architecture and a flexible, AI-ready development platform. As an AI-ready data lakehouse, Starburst Galaxy provides both out of the box with minimal overhead.
Starburst Galaxy doesn’t replace a company’s existing data and AI architecture. It fits squarely into the existing framework, integrating seamlessly with current capabilities. The result is less time spent re-architecting and seeking purchase approvals and more time spent coding the solutions that will power tomorrow’s business.
Technology and organization working together
Recognizing that technology alone isn’t the whole story is also crucial for successful AI adoption. Many organizations struggle not because the tools are missing, but because teams, processes, and priorities aren’t aligned around a shared data strategy. In the end, real progress comes when the organizational and technological foundations evolve together. This happens when data access, governance, and collaboration are treated as company-wide responsibilities, not just engineering tasks.
Unlocking alignment using AI
When that alignment happens, AI adoption stops being a series of stalled experiments and becomes an operational capability. Starburst Galaxy provides the foundation for both sides of that transformation. It gives organizations the technology to move faster and the clarity to move together.
