AI Agents, the Trust Problem & Where to Build Next

Oct 12, 2026

Giving an AI agent access is one decision. Understanding everything it might do with that access is a much bigger challenge.

Welcome to the first episode of DeCompute, a podcast from Silence Laboratories exploring security, privacy, and the systems we need to build as AI becomes more autonomous.

Jay Prakash joins Sid Trivedi, Partner at Foundation Capital, for a conversation about what changes when software can reason, choose its own approach, and act on our behalf. Sid brings the perspective of an investor working closely with cybersecurity founders, along with lessons from building the CyberSeed community.

They explore why agent security needs to extend beyond identity and access, how to keep actions aligned with human intent, and what happens when agents from different companies start working together. Along the way, they consider reputation scores for agents, insurance for autonomous decisions, and why personal AI assistants could bring consumer cybersecurity back into focus.

For anyone building agents or the infrastructure around them, this conversation offers a closer look at the problems still waiting to be solved.

What you’ll hear in this episode

Sid’s journey into cybersecurity investing and the story behind CyberSeed.
Why agents challenge security assumptions built around human behaviour and predictable software.
How runtime security can assess an agent’s actions after access has been granted.
The tension between enforcing boundaries and preserving an agent’s ability to find better solutions.
Why “Know Your Agent” involves questions about behaviour, scale, and the cost of monitoring.
How collaboration between agents from different vendors creates new trust boundaries.
Whether agents could develop reputations that help others assess risk.
Why selective data access matters when personal assistants can read, share, and act on sensitive information.
Where Sid sees opportunities for founders, including consumer security and agent orchestration.

You should listen to this episode if you

Build AI agents or products that take actions on someone else’s behalf.
Lead security or engineering teams introducing agents into business workflows.
Work on identity, authorization, runtime security, or privacy infrastructure.
Are designing systems where agents collaborate across companies and applications.
Want an investor’s perspective on emerging opportunities in cybersecurity.

Episode notes

Key Takeaways

Final Takeaway

About the guests

Sid Trivedi

Partner, Foundation Capital

Sid invests in enterprise software and cybersecurity. He also helped build CyberSeed, a community connecting cybersecurity founders with experienced operators, investors, and peers. In this episode, he shares the questions shaping his thinking about agent security and where founders could build next.

Connect with Sid

Giving an AI agent access is one decision. Understanding everything it might do with that access is a much bigger challenge.

Welcome to the first episode of DeCompute, a podcast from Silence Laboratories exploring security, privacy, and the systems we need to build as AI becomes more autonomous.

For anyone building agents or the infrastructure around them, this conversation offers a closer look at the problems still waiting to be solved.

What you’ll hear in this episode

Sid’s journey into cybersecurity investing and the story behind CyberSeed.
Why agents challenge security assumptions built around human behaviour and predictable software.
How runtime security can assess an agent’s actions after access has been granted.
The tension between enforcing boundaries and preserving an agent’s ability to find better solutions.
Why “Know Your Agent” involves questions about behaviour, scale, and the cost of monitoring.
How collaboration between agents from different vendors creates new trust boundaries.
Whether agents could develop reputations that help others assess risk.
Why selective data access matters when personal assistants can read, share, and act on sensitive information.
Where Sid sees opportunities for founders, including consumer security and agent orchestration.

You should listen to this episode if you

Build AI agents or products that take actions on someone else’s behalf.
Lead security or engineering teams introducing agents into business workflows.
Work on identity, authorization, runtime security, or privacy infrastructure.
Are designing systems where agents collaborate across companies and applications.
Want an investor’s perspective on emerging opportunities in cybersecurity.

HOST