Can your AI Platform do this? Agents on Trial: Proving What Your Workflows Actually Did

The cloud native world spent a decade making distributed systems resilient: agents recover, workflows resume, retries just work. But as autonomous agents start making decisions that land in front of regulators, resilience isn't enough. The new question isn't "did it run?" — it's "can you prove what happened, who participated, and that no one altered the record afterward?"

That's a hurdle for anyone shipping agents into legal and regulated systems. Observability tells you what happened, but every log, trace, and audit row requires trust — and trust isn't a chain of custody.

In this session we'll walk through verifiable execution that complement durable execution. Workflow history signing, history propagation, and attestation, all built on Dapr's existing SPIFFE workload identity. SPIFFE answers "who are you?"; verifiable execution answers "how did you get here?" We'll cover why this matters specifically for multi-agent systems that delegate work, invoke tools over MCP, and trigger long-running workflows, and how to enforce that a wire transfer, a claims reimbursement, or a batch release only proceeds when the execution lineage proves every required step occurred.

Live demos show signed history detecting tampering on state load, propagated lineage across child workflows and activities, and securing MCP tool calls, all running on Diagrid Catalyst.

Who this webinar is for

• Developers building production-grade AI agents and multi-agent workflows
• Platform and security engineers responsible for zero-trust and chain-of-custody
• Teams shipping into regulated/legal environments — fintech, healthcare, pharma, insurance
• Compliance-minded engineering leaders evaluating how to make agents auditable
• DevOps/SRE teams supporting agent systems where "prove it" is a real requirement

Demo details

Live, we'll sign a workflow's execution history and then tamper with a record to watch verification fail on the next state load — chain linkage, digest, and SPIFFE identity checks all catching it. We'll show history propagation carrying verified lineage into child workflows and activities, then apply it to a realistic regulated scenario (e.g., a payment or claims workflow that only authorizes when the propagated history proves prior steps ran). We'll finish by securing an MCP tool call as a durable, attested activity. Everything runs on Dapr 1.18 on Diagrid Catalyst.