Kay Malcolm

Kay Malcolm is a database engineer and product management leader whose work at Oracle connects developer education with shared agent memory. She created Oracle LiveLabs, giving developers hands-on access to Oracle technology, and has tackled a related collaboration problem: AI can accelerate an individual’s coding while leaving colleagues without the reasoning needed to continue the work.

Engineering roots and the cost of separate databases

Malcolm studied electrical engineering at the University of Maryland, College Park, earning a B.S. Her career included Southern Company before she joined Oracle. Her technical teaching biography describes starting at Oracle as a sales engineer specializing in Real Application Clusters and Exadata, followed by database product-management leadership. It credits her as the author of multiple certifications covering Big Data, Exadata, Real Application Clusters, and Autonomous Database.

At Southern Company, Malcolm specialized in SQL performance tuning. In her recorded account of that work, she describes a utility application whose false positives and false negatives affected information used by crews restoring power. To examine relationships among nearby infrastructure, she wrote a query with five nested UNION ALL operations. She recalls that it might have taken about 20 minutes to run; the organization subsequently adopted Neo4j. The experience exposed both the difficulty of expressing relationships in relational queries and the operational cost of adding another database: each system brought its own security and patching meetings. She connects that experience to her later advocacy for storing different representations of data together.

Developer education and community

• LiveLabs: LiveLabs turned her teaching work into a platform. Malcolm describes writing it herself and leading a team responsible for platform development, workshop content, quality assurance, and front-end development. Oracle LiveLabs lets learners try Oracle technologies through workshops and temporary access to cloud resources. Her contribution extends to individual teaching materials: Oracle’s repository credits her with creating a cloud-account onboarding lab, while separately naming its contributors and later maintainer.
• Partnerships and mentorship: Her education work also includes institutional partnerships. Malcolm’s team launched an Applied Database Systems course at Howard University, bringing database product managers into the classroom as guest lecturers. She hosts Lashes.Love.Tech, exploring women’s careers in technology, and has supported Oracle’s work with historically Black colleges and universities and Hispanic-serving institutions.

Shared memory for a distributed coding team

Her work on agent memory began with a management problem inside her distributed team. Developers using AI coding tools produced changes faster, but their European and US colleagues did not consistently receive the context behind those changes. A commit arriving overnight carried code without the coding-agent conversation that explained decisions, outstanding questions, or intended next steps. Repositories diverged, and testing and validation continued to consume time. Faster code generation had left a collaboration bottleneck.

The team built a memory broker using the Oracle Agent Memory SDK and an Oracle Autonomous Database. Malcolm describes it sharing context across developers and associating that context with forks, branches, and commits while developers remained in control. The mechanism matters: preserving a decision with the relevant change gives the next person more to work with than the final diff. Her talk presents this as her team’s response to lost context; it supplies no measured productivity improvement.

Memory, retrieval, and converged data

Malcolm places memory within the broader agent harness: the tools, context, retrieval, and safeguards that let a model act. She distinguishes short-term session memory, long-term information retained across sessions, episodic records of earlier interactions, procedural knowledge about steps and tools, and semantic knowledge. Retrieval must select the information relevant to the task. Persisting everything does not make everything useful.

That distinction leads to her argument for converged data management. Different memory jobs can call for relational records, JSON, graph relationships, vector retrieval, or text. When those representations live in separate systems, applications must coordinate them and determine which information to trust. Malcolm advocates bringing them into Oracle’s database platform to reduce that coordination burden. This is her architectural position, grounded in her database-administration experience and team-memory example; storage consolidation alone does not establish agent reliability.

Teaching practical AI use

Her practical explanation of agent memory gives users ways to test continuity: ask an agent to reconstruct an earlier decision, retain a taught process, follow an example format, or wait for several out-of-order updates before reaching a conclusion. Each test examines a specific behavior rather than treating the presence of saved conversation history as proof that the agent remembers effectively.

Through AI For You, Malcolm extends this teaching approach to professionals using AI in everyday work.