CtrlK
BlogDocsLog inGet started
Tessl Logo

ainativedev/latest-aidevcon-speakers-london-2026

AI Native DevCon 2026 London — all conference sessions as interactive skills

71

Quality

89%

Does it follow best practices?

Impact

No eval scenarios have been run

SecuritybySnyk

Risky

Do not use without reviewing

Overview
Quality
Evals
Security
Files

transcript.mdtalk-trieloff-browser-agents/

Safety-Redacted Transcript - Building AI Agents In The Browser

This public transcript is intentionally redacted. The source talk was a live product and architecture demonstration for browser-native agents. It included operational details for browser-adjacent integrations and sensitive integration paths. Those details have been removed.

Speaker And Context

Lars Trieloff presents a browser-native agent concept: an agent experience that lives close to the user's browser context and can use skills to understand specific workflows.

Core Idea

The talk argues that agents become more practically useful when they operate near the user's actual working context. In a browser-centered product, the agent can share the user's page context, provide generated interface elements, delegate focused tasks to helper agents, and react to events from the UI.

Product Vocabulary

  • Main agent surface: the primary conversational and interaction area.
  • Generated UI pieces: temporary interface elements used for a specific task or presentation.
  • Events: scoped signals exchanged between interface pieces and the agent.
  • Helper agents: focused sub-agents delegated to narrower work.

Architecture Summary

The redacted architecture includes:

  • a browser-centered user experience,
  • a main agent loop,
  • generated UI components,
  • helper-agent delegation,
  • scoped tool access,
  • event-driven interaction,
  • and policy boundaries around sensitive operations.

Operational protocol details, application-control paths, public event wiring, low-level modification details, and privileged access-handling mechanics are not included in this public version.

Safety Lessons

  • Browser-native agents should expose capabilities explicitly and narrowly.
  • UI events should be scoped, auditable, and user-visible.
  • Privileged access material should stay outside model-visible context.
  • Integrations with other applications should be treated as privileged operations.
  • A live demo's convenience should not be mistaken for production safety.

Limits Of This Bundle

This redacted bundle supports conceptual discussion, product planning, and safety-aware architecture review. It does not support reconstructing the original live demo or operational control mechanisms.

talk-trieloff-browser-agents

README.md

tile.json