Resources: Lisbon AI Week PagerDuty + Cloudflare Meetup

👋 Hi community,

What an amazing evening yesterday at the Lisbon AI Week: PagerDuty + Cloudflare Meetup!

In case you couldn’t be there or want to revisit the insights shared, we’ve got you covered:

🎤 PagerDuty MCP Server:
José Corte-Real (Staff Software Engineer, PagerDuty) & Hélder Duarte (Senior Software Engineer, PagerDuty)

⭐ Talk Highlights:

• Start from user journeys and competency questions; design the minimal, ergonomic tool set rather than auto-converting OpenAPI to MCP.
• Keep tool count low and avoid semantically overlapping tools; too many/similar tools degrade LLM selection and increase confusion.
• Design tools for agents: use categorical params (e.g., scope = mine|teams|all) and include helpful metadata (e.g., “records_count”) to reduce LLM errors.
• Evals focus on tool trajectories, not final answers: verify the right tools, order, and parameters across models; mock tool outputs for reproducibility.
• Avoid tiny/empty tool responses—they invite hallucinations. Ensure responses are informative and context-relevant.

🎤 Agentic RAG / AI Search:
Gabriel Massadas (Senior Systems Engineer, Cloudflare)

⭐ Talk Highlights:

• Managed RAG: ingestion from buckets/web, multimodal text extraction (vision/OCR), chunking, embeddings, and vector storage handled for you.
• “Agentic search” lets the LLM iterate searches, refine queries, and decide when it has enough evidence before summarizing.
• Agentic loop: extract knowledge from retrieved results, judge sufficiency, iterate up to a cap, then answer—improves quality vs. single-pass.
• Conversation-aware query rewriting merges follow-ups into a single contextualized query to preserve chat context.
• Reference implementation with Cloudflare Agents SDK deployable on Workers; integrates with AI Search, Vectorize, and AI Gateway.

🎤 PagerDuty SRE Agents:
Pedro Vicente (Senior Engineering Manager, PagerDuty)

⭐ Talk Highlights:

• Reduce toil and MTTR by surfacing key context fast: past incidents, runbooks, recent changes, related services, live metrics/logs.
• Categorize incidents by novelty/repeatability to set autonomy: novel (assist), partially understood (guided), recurrent (higher autonomy).
• SRE agents also coordinate: capture meeting/chat context, summarize, and keep stakeholders aligned during crises.
• Prioritize diagnosis first; gate remediation autonomy with confidence, repeatability, and customer-defined permissions/guardrails.
• Vision: shift time from firefighting to innovation—“build more, babysit less.”

Big shout out to our friends at Cloudflare for partnering up with us and Lisbon AI Week for the invite to be part of this debut initiative!

Subscribe to PagerDuty Community Events and don’t miss the next ones!! 

Have you attended this meetup? Leave a comment telling us what you thought and/or ask the speakers questions 💚