top of page

MariaDB for AI: takeaways from MariaDB’s first webinar in Portuguese with Amber

Oct 27
7 min read

Updated: Nov 5

On October 15, MariaDB® hosted its first webinar in Portuguese in partnership with Amber. The session, titled "Driving the Agent-Oriented Enterprise," announced and reinforced the partnership between the two companies and addressed what a business must resolve in its data before putting AI agents to work. I opened the webinar by discussing the challenges companies face in turning technology into results and how Amber operates from diagnosis to ongoing support. Next, Natasha Rinaldi from MariaDB presented how the platform is structured to support AI applications and agents.


This article gathers the takeaways from the session that matter to companies evaluating MariaDB for AI or structuring the database that agents will query. They cover data quality, what businesses now require from a database, the MariaDB platform, governance, and the role of consulting in adoption.


Foca dourada sobre discos, com logos Amber e MariaDB e IA em fundo escuro e brilho laranja.


AI AGENTS DEPEND ON THE QUALITY OF THE DATA THEY RECEIVE


The first takeaway from the webinar is that the primary barrier to using AI agents lies in the data, before the chosen technology. Natasha shared the projection Gartner published in a June 2025 release: by 2028, 33% of enterprise software applications will include agentic AI, up from less than 1% in 2024. In the same release, Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027 due to escalating costs, unclear business value, or insufficient risk controls.


Another data point from the presentation indicates where companies face the most difficulty. In the PEX Report 2025/26, based on a PEX Network survey of more than 200 professionals, 52% point to data quality and availability as the biggest obstacle to adopting AI. Lack of internal expertise follows closely behind at 49%.


An AI agent queries information, cross-references records, and executes tasks based on what it finds in the company’s systems. If the customer record is outdated in the CRM, if the order history sits in another unintegrated system, or if part of the tracking is still done in spreadsheets, the agent responds and acts with wrong or incomplete information. And, unlike a human representative, it repeats the same mistake in every interaction it handles.


This diagnosis confirms what I presented at the opening of the webinar. The first difficulty is digital adaptation: companies of any size need to be digital, and this movement accelerated after the pandemic. The second is executive maturity, because technology advances faster than leadership teams can keep up. The third is what we call tech abundance: there is excess technology and too much information reaching executives and technical teams, which requires greater care in decision-making and clear direction to integrate solutions.


In practice, before choosing which agent to use, a company must answer three questions: where the data it will query is located, which systems must be integrated so this data remains consistent, and who is accountable for information quality. For this reason, in the opening, I placed end-to-end process organization, built on a single database, as the step that follows attracting demand and converting opportunities. This is where database selection enters the discussion on MariaDB for AI, the subject of the next takeaway.




WHAT COMPANIES NOW REQUIRE FROM A DATABASE


The second takeaway is that AI agents change the criteria for choosing a database. In her presentation, Natasha summarized this shift into five market requirements that share the same origin: applications that query data and act on it continuously, without a person reviewing every step. In the table below, each requirement appears alongside a question that a company can bring to its vendor evaluation.


Requirement

What MariaDB presented

Question for the vendor

Scale

Data and users grow beyond database capacity, with workloads distributed across regions and clouds

Does the database move from one server to several without rewriting the application?

Multi-model

Transactions, analytics, and vectors on the same platform

How many systems will be needed to store transactional data, vectors, and analytical data?

Low latency

Sub-millisecond response with high concurrent access

What is the response time measured at our peak access volume?

Context

Reliable and governed base, with memory for agents and auditable trails

Can we audit which data the agent queried before acting?

Governance

Control over data residency and sovereignty

In which country and environment is the data stored?

These questions serve to compare any vendor, including MariaDB itself. In customer service and sales agents, latency and auditing are the most sensitive points: the agent responds to the customer during the conversation, and the company needs to know afterward why it responded that way.



MARIADB FOR AI: A SINGLE DATABASE FOR TRANSACTIONS, VECTORS, AND ANALYTICS


The third takeaway is that MariaDB organized its platform to meet these requirements with a single database, instead of a separate system for each data type. According to the company, the database totals more than 3 billion downloads and over 800 enterprise customers in production. It is also compatible with MySQL, allowing organizations that already use MySQL to evolve their architecture without rebuilding the database.


InnoDB, MariaDB’s storage engine, holds four data models in the same place: rows for transactions, vectors for semantic search, JSON for semi-structured data, and columnar format for analytics. Vectors allow the database to store embeddings, the numerical representations of texts and documents that agents use to find information by meaning rather than exact word match alone. For high-volume analytics, MariaDB presented Exa, developed in partnership with Exasol.


Between the application and the database sits MaxScale®, MariaDB’s proxy that was also discussed in the webinar. It routes writes to the primary server and reads to the replicas, and if the primary server goes down, it automatically promotes the most suitable replica. With the transaction replay feature active, an in-progress transaction is re-executed on the new server without the user noticing the failure. MaxScale also captures data changes (CDC) to keep other systems synchronized.


For agents, the platform now includes GridGain, an in-memory caching and processing company recently acquired by MariaDB, as Natasha shared during the webinar. This component stores the context of agent interactions as short-term and long-term memory and offers key-value storage with sub-1-millisecond response times, according to MariaDB. With the launch of MariaDB Enterprise Platform 2027 in November 2026, the platform also gained a semantic cache, which stores answers already generated by the AI model and reuses them when a new question is similar enough, reducing the cost of model calls.


Connection to AI applications happens through RAG, a technique in which the model retrieves relevant passages from company data before responding, supported by the database’s native vector capabilities. There are connectors for the LangChain, LlamaIndex, and Spring AI frameworks, as well as an MCP Server following the Model Context Protocol, a standard used by AI agents to access data and tools. The platform also includes AI copilots for developers and database administrators. In the 2027 release, it added AI Retrieval, with ready-to-use RAG pipelines, and AI plugins for AI-assisted coding tools such as Claude Code and Codex.


In practice, a customer service agent that needs the customer’s purchase history, manual excerpts found by similarity, and the context of the ongoing conversation accesses all three on the same platform instead of depending on three separate integrations. The gain for the company is having fewer systems to integrate, synchronize, and govern.


During evaluation, a company needs to separate what is already available from what is still on MariaDB’s roadmap, such as distributed SQL. To understand the database fundamentals and how it compares to MySQL, read the article What is MariaDB and how it compares to MySQL.




GOVERNANCE: ACCESS RULES, HUMAN OVERSIGHT, AND CONTROL OVER WHERE DATA RESIDES


Autonomy for AI agents requires rules defined prior to deployment. At the opening of the webinar, I showed that every agent project must establish operating boundaries, data access rules, information quality criteria, and human oversight at decisive stages, with gradual validation of results. In a sales agent, for example, this can mean that it checks prices and availability on its own, while a discounted proposal goes through a salesperson before reaching the customer.


This level of care is still a minority practice among companies. The same PEX Report 2025/26 cited in the first takeaway shows that fewer than half of the surveyed companies have an AI governance policy.


Governance also involves control over where data resides. Regulated companies or those with regional requirements must guarantee data residency and sovereignty, and MariaDB presented four deployment options for this: on-premises, with full infrastructure control; multicloud on AWS, Google Cloud, and Azure, without depending on a single provider; serverless, with pay-per-use billing; and BYOA, where the managed platform runs in the customer’s own cloud account. The BYOC (Bring Your Own Cloud) option, which was on the roadmap at the time of the webinar, is now available, as MariaDB announced at the launch of MariaDB Enterprise Platform 2027 in November 2026. It deploys platform components in cloud accounts the customer controls.


In Brazil, these choices must account for the LGPD (Brazilian General Data Protection Law) from the initial solution design. In Amber’s projects, this includes access controls, encryption, log management, and data governance policies. The topic is covered in detail in the article AI and LGPD: how to adopt with legal security, and security criteria for critical systems can be found in Secure database: why critical systems are migrating to MariaDB.




HOW AMBER SUPPORTS THE ADOPTION OF MARIADB FOR AI IN BRAZIL


Technology generates returns when it enters a company through diagnosis, planning, and ongoing monitoring. At the opening of the webinar, I presented the role of consulting in three stages: diagnosis, to map pain points before any investment; planning, with a roadmap to introduce AI safely and select solutions compatible with the existing architecture; and execution, with change management, ongoing support, and return on investment tracking.


In MariaDB projects, these stages translate into concrete activities. Amber starts from the technical diagnosis of the environment and designs the data architecture according to the company’s objectives. Implementation includes compatibility validation, load testing, and contingency strategies, both in initial adoption and in migrating from MySQL or other databases to MariaDB.


Integration with legacy systems, APIs, and enterprise platforms allows MariaDB to run alongside existing processes without interrupting them. Once the database is in production, Amber supports the environment with continuous technical support, performance tuning, and strategic monitoring so that the database supports company growth and new AI demands.


At the close of the webinar, I brought up a point we had not covered in the presentation: beyond the technical argument, MariaDB is a cost alternative. Competitive pressure reaches the final price and leads companies to review processes and architecture, and an open database that connects to the existing structure allows this review without replacing everything at once. Adoption can be modular, resolving technical and financial points in stages rather than through a big bang project.


The proposal presented in the webinar combines MariaDB’s technology with a local team operating in Brazil and Latin America, with projects delivered across eight countries. For companies looking to prepare their data for AI agents, the next step is a joint assessment with Amber, built around the company’s use case and a diagnosis of the current environment, which can be requested via our WhatsApp.

MariaDB® and MaxScale® are registered trademarks of MariaDB plc. MySQL is a registered trademark of Oracle Corporation, and there is no affiliation between MariaDB plc and Oracle Corporation.

 
 
 

Comments

Rated 0 out of 5 stars.
No ratings yet

Add a rating
bottom of page