About Standard
Sensitive agent work needs a boundary the agent cannot control
Agents are moving from answering questions to acting on private data, software, and production systems. Sandboxes contain processes, but sensitive functions also need confidential execution, authority outside the agent, controlled release, and evidence the next system can verify.
Our vision is to make managed confidential execution as practical to integrate as an ordinary cloud function, without reducing the security boundary to an agent’s claim.
Why we’re building Standard
As agents take on more responsibility, people need more than an agent’s account of what happened. They need independently enforced permissions, protection for private data, and evidence they can verify.
That is the ambition behind verifiable autonomy. Standard focuses on a concrete part of it: making sensitive agent functions confidential, controlled, and verifiable.
Founder
Vamsi Komanapalli
Founder, Standard
A word from the founder
Using AI agents for software and vendor work, I hit the same limit: the agent requesting or producing the work could not also define the checks, inspect every private input, and prove what ran. Standard exists to make sensitive functions confidential, controlled, and verifiable without requiring every team to build that infrastructure.
Vamsi brings 20 years of experience spanning cloud security, regulated infrastructure, engineering leadership, and enterprise architecture.
At Oracle Cloud Infrastructure (OCI), he led infrastructure and capacity engineering for Key Management Service (KMS) and technical programs across the Bastion and Secrets teams. His work spans cryptographic infrastructure, roots of trust, secure access, secrets management, and large-scale cloud operations, with a strong focus on regulatory compliance and government cloud security requirements, including FIPS 140-3 initiatives.
At lululemon, he led four engineering teams and more than 20 engineers supporting digital commerce, including the migration of platform services to AWS. Earlier at Oracle, he combined enterprise architecture and presales, working with executives on technology strategy and solution design for major retailers, including Walmart and Kohl’s.
He leads the company’s product direction, architecture, and engineering.
What we believe
People set the permissions
An agent can request work. The owners define what is allowed, which inputs can be used, and which results can be released.
Useful work needs clear data boundaries
Approving a function to use private data should not mean giving the requesting agent unrestricted access to that data.
Evidence should be independently checkable
The next system needs evidence of the approved execution. It should not have to rely on the requesting agent’s account of what ran.
Partner with Standard
Partner with us on agent integrations and protected functions that need managed confidential execution and verifiable results.