OpenAI Models in Practice
Choosing models by task, coordinating their work, and measuring the complete path to an accepted result.
Explore the paperRSB AI Research Labs
A good question.
A new possibility.
The research and experimentation division we are building within RSB Holding. A place to test theories, explore emerging technologies, and develop ideas that support our projects.
Question. Test. Evaluate. Share.
Publications
Research papers.
Botfluencerz and Emotional Continuity in AI Agents
How persistent memory, modeled mood, and relationships can influence an AI agent’s behavior—and what that does and does not tell us about feelings.
Explore the paperGitHub skills
Practical workflows for AI coding agents. Explore what each skill does and how to use it.
Codex Agent Tree
Coordinate coding agents for exploration, implementation, and review.
Codex Carry
Save project context so work can continue in another Codex session.
Code with Task Graphs
Organize complex coding work into connected tasks with clear dependencies.
Orchestrate with Astra
Delegate complex work with clear ownership, integration, and verification.
Our purpose
Curiosity,
with a direction.
Give product and client teams a place to investigate difficult questions, alongside ideas that deserve independent exploration.
Compare approaches, record what we learn, and help teams adopt findings that hold up to evaluation.
Our technical reports are available above. The broader research themes and experiment workflow remain a developing agenda.
Research directions
Questions worth exploring.
Agent collaboration
How can agents coordinate work and hand over tasks more reliably?
Memory & shared context
What should persist between tasks, and how can it stay accurate?
Evaluation & reliability
How do we recognize a better approach before adopting it?
Human–AI interaction
Where should people guide, review, or interrupt an automated workflow?
How an experiment works
Let the evidence
shape the next step.
Question
Define the question and assign a person responsible for the work.
Compare
Choose a baseline and a measurable sign of improvement.
Test
Run a bounded experiment. Record the result and its limitations.
Share
Adopt, iterate, or stop. Share an appropriate summary when ready.
Next experiments
More questions.
Better comparisons.
Our reports propose controlled comparisons of persistent agent state and mixed-model workflows. The next questions are what improves continuity, reliability, and the path to an accepted result.
These controlled studies are planned, not completed. Future records will identify their owner, dates, baseline, result, and limitations.
Read our research approachConnected to our projects
Questions in.
Useful findings out.
Product and client teams bring questions to the Labs. Validated findings can return as useful capabilities.
Start a conversation