Prioritize real web workflows, programmable execution, and independent verification.
Real browser agents navigating portals and retrieving documents directly fit GuardDog and Birdie automation.
Official session info ↗Programmable execution and isolation offer the closest adjacent match to giving agents reliable control over a computer.
Official session info ↗Trace-based regression testing connects to independent action verification and evidence-led debugging.
Official session info ↗2:55 PM · Attack demo → 3:20 PM · Browser agents → 3:45 PM · Give the model a computer (voice is the backup).
For philosophy: 9:54 AM code quality, 11:55 AM what evals justify, and 1:40 PM human intention and craft.
11:30 AM Exa Snapshot + 12:20 PM Your Next User Is an Agent connect retrieval and agent discovery to GolfTroop’s web presence.
These picks cover retrieval and agent discovery; AI recommendation and ranking signals are useful questions to bring.
Full talks · no overlaps · exact rooms
Filters highlight interests within the route; backups are alternatives, not extra talks to attend.
A shared repository of source documents is a useful model for GuardDog evidence retrieval and Birdie’s access to business context.
Official session info ↗No parallel talk is listed; use this slot to recharge or discuss browser automation, AI-EO, or agent trust with attendees.
Reusable connectors, skills, and controls could help Birdie expand from individual tasks into repeatable GolfTroop workflows.
Official session info ↗No parallel talk is listed; use this slot to recharge or discuss browser automation, AI-EO, or agent trust with attendees.
This examines how to give GuardDog and Birdie more useful autonomy while keeping their actions within trustworthy boundaries.
Official session info ↗No parallel talk is listed; use this slot to recharge or discuss browser automation, AI-EO, or agent trust with attendees.
Its principles for deciding what humans should review directly address how you judge and govern AI-generated software.
Official session info ↗No parallel talk is listed; use this slot to recharge or discuss browser automation, AI-EO, or agent trust with attendees.
Defining what counts as good and done is central to Birdie’s product behavior and your role as an agent builder.
Official session info ↗No parallel talk is listed; use this slot to recharge or discuss browser automation, AI-EO, or agent trust with attendees.
Business semantics, permissions, and durable memory map directly to Birdie’s context and GuardDog’s interpretation of records.
Official session info ↗Its human-calibrated style benchmark could help Birdie produce GolfTroop messages in a consistent brand voice.
Official session info ↗You can apply its dataset-building approach to GuardDog extraction and Birdie workflows that lack an off-the-shelf benchmark.
Official session info ↗Long-running goals and human/agent collaboration map to Birdie’s multi-step business workflows.
Official session info ↗Understanding web retrieval over time can inform GolfTroop visibility experiments and GuardDog source provenance, though this is not an AI-ranking tutorial.
Official session info ↗Source provenance and defensible decisions are directly useful for GuardDog evidence and Birdie audit trails.
Official session info ↗It challenges whether an eval score really justifies autonomy, a critical question for consequential GuardDog findings and Birdie actions.
Official session info ↗Permission-aware retrieval is useful for keeping Birdie customer data and GuardDog restricted records properly scoped.
Official session info ↗Agent discovery, onboarding, and authorization are the closest match to making GolfTroop usable by AI visitors, with ranking signals outside the stated scope.
Official session info ↗System-level failure analysis helps assess harm and authority boundaries beyond a successful agent output.
Official session info ↗62 minutes free between talks; the official lunch window is 12:00–2:00 PM.
Its focus on human intention, craft, and judgment offers a philosophical lens for building agents that serve people well.
Official session info ↗Memory design and prompt reconstruction can help Birdie retain useful context without an ever-growing token bill.
Official session info ↗The live access-control and business-logic investigation is a direct ethical-hacking match for testing GolfTroop and Birdie integrations.
Official session info ↗Delegating authority using lessons from human governance is a strong philosophy alternative for GuardDog and Birdie.
Official session info ↗Tracing delegated identity through MCP tools can help prevent Birdie from performing a valid-looking request for the wrong user.
Official session info ↗Persistent histories linking code, conversations, and collaborators are relevant to Birdie’s development workflows.
Official session info ↗Six concrete attack classes and a safe test lab offer useful threat models for Birdie’s future purchasing or payment actions.
Official session info ↗Unattended-agent crashes, lost progress, and silent failures map directly to GuardDog and Birdie operations.
Official session info ↗Browserbase’s production portal workflows are today’s closest match to hands-on browser automation for GuardDog document retrieval and Birdie external-system tasks.
Official session info ↗Its trace-driven observe, evaluate, and test loop is closely related to verifying browser actions and catching GuardDog or Birdie regressions.
Official session info ↗Giving agents an isolated computer and choosing execution boundaries closely matches programmable browser control and is useful for GuardDog portal automation and Birdie tools.
Official session info ↗Natural turn-taking, latency, and conversation-level testing are useful if Birdie handles customer calls or voice interactions.
Official session info ↗Move to Simons Room (Metropolitan East) for the 4:25 PM welcome and featured keynotes.
Its framework-oriented approach may offer transferable ways to think about AI systems, although the published description is brief.
Official session info ↗No parallel talk is listed; use this slot to recharge or discuss browser automation, AI-EO, or agent trust with attendees.
Models that act in space and time are adjacent to visual computer-use agents, with browser and CAPTCHA coverage unconfirmed.
Official session info ↗No parallel talk is listed; use this slot to recharge or discuss browser automation, AI-EO, or agent trust with attendees.
Open agent foundations could inform your choices about Birdie’s control, extensibility, and self-hosted components.
Official session info ↗No parallel talk is listed; use this slot to recharge or discuss browser automation, AI-EO, or agent trust with attendees.