Research areas.Six disciplines, one architecture.

Codira’s research is organized around six connected areas. Each addresses a distinct problem in building intelligent systems, and progress in one tends to change what is possible in the others.

The areas.And the questions we are working on.

Multi-Agent Orchestration

How specialized agents take distinct responsibilities, exchange information and manage dependencies without the workflow losing coherence.

Open questions

Which decisions belong to individual agents, and which must stay with the orchestration system?
How should work be decomposed so each task can be checked on its own?
How can a failure in one task be contained before it spreads to the next?

In depth The technology

Deterministic AI Governance

How deterministic mechanisms can govern selected parts of AI-assisted execution without removing what makes models useful.

Open questions

How should a model's recommendations be kept separate from the actions a system is authorized to take?
Which policies can be enforced in code, and which need review?
How should probabilistic and deterministic components divide the work?

In depth The technology

Verification and Validation

Methods for evaluating intelligent work throughout a workflow, not only at the end, and for knowing their limits.

Open questions

Which requirements can be stated precisely enough to check by rule?
How should checks be placed so one stage's error is caught before it becomes the next stage's input?
How should a passed check be presented so it is not mistaken for proof?

In depth The technology

Context and State Management

Preserving what a workflow has established, while keeping irrelevant, outdated, contradictory or out-of-scope information away from each agent.

Open questions

What does each responsibility actually need to know?
How should decisions and constraints be carried forward and kept current?
How can information access be scoped while staying traceable?

In depth The technology

Human-AI Collaboration

How AI contributions are reviewed, interpreted and authorized within decisions for which people remain accountable.

Open questions

Where does a workflow need a person, and what should that person see?
How should automation be distinguished from autonomy in practice?
How can AI work be presented so it supports judgment rather than replacing it?

In depth The technology

Model Interoperability

Separating orchestration from model-specific behavior so systems can adopt new models without being redesigned around them.

Open questions

How should tasks be routed by the capability they need?
How can differences in interfaces, outputs and limits be absorbed by the architecture?
How do governance and verification stay unchanged when the model underneath changes?

In depth The technology

Proposed areas

Further areas under consideration. They are proposals, not active programs, until they are approved and staffed.

Agent specialization
Domain-specific intelligence
AI-assisted professional workflows
Reliable autonomous execution

Working on similar questions?We would like to hear from you.

Researchers, institutions and technical partners whose work meets these areas can reach our research team directly.

Research Partnerships Publications