← All speakers

Bio, Work & Ideas

Remy Guercio

Conference affiliation: Tailscale · 2026

Remy Guercio works on strategic projects at Tailscale, where he develops Aperture, an identity-aware gateway that gives developers and autonomous agents controlled access to AI models without distributing provider credentials. His central security insight is that isolating an agent inside a container means little if that agent retains powerful API keys or authenticated sessions.

Guercio’s earlier work spans several generations of cloud infrastructure. At Cloudflare, he wrote about Cloudflare Workers and serverless applications in 2018 and 2019, emphasizing experimentation and software running at the network edge. At Lambda, his work in 2020 and 2021 addressed GPU cloud infrastructure and machine-learning workloads, including instance provisioning, training-data transfers, performance, and computing costs.

At Tailscale, he applied identity-aware networking to private GitHub Actions connectivity and tsidp, an open-source identity provider that extends existing organizational identities to internal applications and Model Context Protocol servers. His work on dynamic client registration addresses how emerging agent protocols can integrate with established identity infrastructure.

Guercio introduced Aperture’s private alpha in January 2026. Its design centralizes provider credentials, attributes model requests to developers or tagged workloads, and exposes usage, costs, and agent activity without placing sensitive keys inside individual execution environments.

  • The network as a security boundary: Authentication and authorization belong at the network layer, where WireGuard connections, organizational groups, and workload tags determine what an agent can reach. A GitHub Actions review bot can receive a scoped network identity instead of a reusable provider key, as Guercio demonstrates in his AI Engineer talk.
  • Credential-free agent access: Centralized gateways make requests, token consumption, tool calls, and shell commands attributable to specific identities; administrators can revoke network access without recovering secrets from an agent.
  • Existing identity, new agent protocols: tsidp combines Tailscale networking and application grants with OAuth and OpenID Connect to secure internal services without replacing an organization’s primary identity provider.
  • Experimentation without provider lock-in: Aperture CLI and a provider-flexible infrastructure model let teams switch models, vendors, and coding-agent interfaces while retaining centralized access policies and spending controls.

Read the topics behind these talks

1 conference talk

Key ideas

Scroll to read ↓

An agent can run inside a container while holding the credential that grants its access. Moving authorization to an identity-aware network changes where that control lives.

  • A boundary needs permissions
    0:43 ↗
  • Connections carry identity
    3:21 ↗
  • Keep provider credentials at the gateway
    5:38 ↗
  • From a user identity to a request body
    7:22 ↗
  • Follow a PR review bot through the gateway
    9:47 ↗
  • A placeholder key, then a failed word count
    11:41 ↗
  • Shared budgets and gateway hooks
    14:07 ↗
  • Build the same identity boundary into a Go service
    15:31 ↗
  • Configure permissions in the UI or in policy
    17:04 ↗
  • Why configure a base URL explicitly?
    18:46 ↗
  • Model permissions are not yet arbitrary tool permissions
    20:25 ↗
  • Generated code exposes the harder boundary
    21:28 ↗

References