Research
AI is becoming systems that plan, decide, and act. Three questions decide whether agents work in the enterprise.
Agent economics
Agent economics is the study of what an agent costs and returns — and why so many pilots never pay back. It treats agents as labor with unit economics: the cost of a task, the value of its outcome, the overhead of supervision, and the point at which automation actually clears its own bar.
Open questions: how to price supervision, how to tell a demo from an operating model, and how to evaluate whether a workflow is ready for automation at all.
Related: Better Models, Different Risks.
Agent reliability
Agent reliability is the study of when agent work can be trusted to run, how it fails, and how it recovers. Durable execution keeps a workflow alive through crashes; reliability asks the larger question of whether the work an agent completes is the work the organization needed done.
Open questions: what belongs in the runtime versus the contract, how failures should compensate rather than merely retry, and what an audit actually needs to reconstruct.
Related: Contract, not runtime · A day in the life · Concord.
Agent alignment and governance
Agent alignment and governance is the practice of keeping agents inside the organization’s intent: every action authorized by someone accountable, overseen at the right altitude, and reversible when it turns out to be wrong.
Open questions: how humans delegate work safely, how agent actions are traced and understood, and what an agent must inherit — role, context, permissions, policy, evaluation — before it acts alone.
Related: Why agents need a contract · The worker and the action · Lattice.
Work with the lab
Avianna advises a small number of startups and enterprise teams on AI strategy, agent workflows, and architecture, grounded in this research. For advisory or collaboration inquiries, contact the lab.