Autonomous agents
Can a system hold a multi-step objective across sessions without losing the thread?
DirectionLong-horizon goal retention, constrained tool use, and recovery after interruption.
EvidenceExperiments, evaluations and working notes will be published as results become available.
Orchestration
How do independent capabilities coordinate toward one outcome without collapsing into a black box?
DirectionExplicit plans, dependency handling, permissions and clear boundaries between agents and tools.
EvidenceDesign notes and system evaluations will document how the architecture develops.
Adaptive systems
How does a system improve from failure without pretending the failure did not happen?
DirectionEvaluation loops, failure classification and evidence-led iteration.
EvidenceTechnical notes and evaluation results will be published when supported by reproducible findings.