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

SKILL.mdtalk-firtman-web-mcp-agentic-web/

name:
talk-firtman-web-mcp-agentic-web
description:
Explains, summarizes, compares, and applies Maximiliano Firtman's AI Native DevCon talk on Web MCP and the agentic web at a conceptual level. Use when the user asks about Web MCP, agentic web patterns, web contracts for agents, frontend-declared capabilities versus backend MCP, safe read-only adoption, evaluation plans, or privacy review for exposing page context. This bundle is safety-redacted and avoids runnable implementation examples.

Web MCP And The Agentic Web - Maximiliano Firtman

Firtman's talk argues that agents need clearer web-facing contracts than pixel guessing or brittle browsing. This public bundle is safety-redacted: it preserves conceptual distinctions and adoption guidance while omitting runnable interface definitions and bridge mechanics.

Safety Scope

  • Explain Web MCP and the agentic web at a conceptual level.
  • Compare frontend-declared capabilities with backend MCP integrations.
  • Help plan safe, minimal, read-only adoption.
  • Do not provide runnable capability definitions, bridge setup steps, or code-like recipes from the source talk.
  • Before responding, verify that the answer does not reconstruct omitted implementation mechanics.

Bundle Files

  • outline.md contains the thesis, concept map, and redaction boundaries.
  • transcript.md contains the safety-redacted talk summary.
  • quotes.md contains safe paraphrased themes, not verbatim operational quotes.

Factual Q&A Workflow

  1. Read outline.md for the relevant concept.
  2. Read transcript.md for the redacted summary.
  3. State clearly when requested implementation detail is redacted.
  4. Answer with conceptual framing, not runnable examples.
  5. Before finalizing, check that the response does not include exact API shapes, bridge steps, or state-changing examples.

Adoption Workflow

When applying the talk to a user's app:

  1. Pick one high-value page state.
  2. Prefer a read-only diagnostic capability first.
  3. Keep the output narrow and small.
  4. Define what the agent should learn, not how to mutate the page.
  5. Review privacy, session, and permission impact.
  6. Add evaluation cases for correct calls, bad arguments, and safe failure behavior.

Example safe adoption checklist:

Goal: Help an agent understand one page state without changing it.
Capability: Read-only diagnostic summary.
Input: Narrow parameters only.
Output: Small structured result.
Privacy review: Confirm no unnecessary user/session data is exposed.
Evaluation: Test valid request, invalid request, and unavailable-state response.
Release gate: Human review before any state-changing capability is considered.

Concept Explanations

When teaching concepts from the talk:

  • Agentic web: explain the three directions of interaction: agents building for the web, agents running inside web experiences, and agents browsing web apps.
  • Explicit contract: explain the shift from inference over UI state to a developer-defined capability.
  • Frontend-declared capability: explain that it reflects the current page state and user context, unlike a backend-only integration.
  • Safe first step: recommend one read-only diagnostic capability before any write action.

Safe Outputs

  • Conceptual summary.
  • MCP-versus-Web-MCP comparison.
  • Adoption checklist.
  • Risk review for exposing page context.
  • Testing and evaluation plan without code.

talk-firtman-web-mcp-agentic-web

README.md

tile.json