QCon Software Development Conferences

Uncover emerging trends and practices from the world’s most innovative software professionals to help you validate your software roadmap. Celebrating 20+ years.

[QCon Software Development Conferences @QCon]

The session covers admission control, concurrency limits, production measurements, debugging inference failures at scale, and how organizational structure shaped the architecture.

[QCon Software Development Conferences @QCon]

Netflix serves ML models synchronously for search and asynchronously for batch workloads.

Similar models, different failure modes.

Read the abstract: https://t.co/zK0VUaKpXW

[QCon Software Development Conferences @QCon]

Rajat Shah and Wenbing Bai will share incidents from both modes and what Netflix changed while rebuilding the platform as a shared offering for other teams.

[QCon Software Development Conferences @QCon]

@Cambridge_Uni Those constraints tighten the further you go from basic operations like get and put.

More sessions like this are coming to QCon London 2027, April 13-15.

Find out more about QCon London 2027 here: https://t.co/qxHLSjXsLU

#QConLondon

[QCon Software Development Conferences @QCon]

Multi-cloud comes with real costs.

Martin Kleppmann, Associate Professor @Cambridge_Uni, breaks down the constraints: duplicated storage, egress fees between providers, added operational complexity, and a feature ceiling set by whichever provider supports the least.

[QCon Software Development Conferences @QCon]

@Tansu That one difference helps explain why the two systems look so different operationally.

More sessions like this are coming to QCon London 2027, April 13-15.

Find out more about QCon London 2027 here: https://t.co/GCWhLffspm

#QConLondon

[QCon Software Development Conferences @QCon]

Kafka's core assumption is that resilience and durability come from replicating data between brokers.

Peter Morgan, Founder @tansu.io, explains how Tansu starts from the opposite assumption: storage is inherently durable.

[QCon Software Development Conferences @QCon]

Ronak Nathani, Principal Staff Software Engineer @Netflix, will cover composable skills, rate limits, delete and scale-down protections, access controls, bounded actions, and peer approval for production changes.

[QCon Software Development Conferences @QCon]

LinkedIn’s ops agent supports a Kubernetes platform spanning 500,000+ nodes and 5M+ pods.

See the skills, tools, and guardrails behind it: https://t.co/BsvusPAqib

[QCon Software Development Conferences @QCon]

One agent serves Slack, coding-agent plugins, and deployment-failure workflows. Rather than add a separate MCP layer, LinkedIn made its existing operational CLIs friendly to agents.

[QCon Software Development Conferences @QCon]

@SpiralDB Each step has its own bandwidth ceiling, and they compound.

More sessions like this are lined up for QCon London 2027 (Apr 13-16).

Find out more about QCon London 2027 here: https://t.co/DICkKkQ3hY

[QCon Software Development Conferences @QCon]

Every GPU training run pays a movement tax before a gradient update runs.

[QCon Software Development Conferences @QCon]

Onur Satici, Staff Engineer @SpiralDB and Core Maintainer of Vortex (LF AI & Data), breaks down the path: fetch from S3 to NVMe, read compressed bytes into RAM, decompress on CPU, then transfer uncompressed bytes over PCIe.

[QCon Software Development Conferences @QCon]

@netflix More sessions like this are lined up for QCon London 2027 (Apr 13-16).

Find out more about QCon London 2027: https://t.co/zx1LXn2krM

[QCon Software Development Conferences @QCon]

Distributed tracing is often positioned as the tool for debugging production issues, but it has a specific failure mode.

[QCon Software Development Conferences @QCon]

@netflix That constraint is one reason Netflix's observability team moved toward building a connected data layer that surfaces the right signal without requiring engineers to pre-identify the trace they need.

[QCon Software Development Conferences @QCon]

Deploying to a European cloud region does not protect you from US legal jurisdiction.

Martin Kleppmann is Associate Professor at the University of Cambridge and author of "Designing Data-Intensive Applications".

Find out more about QCon London 2027 here: https://t.co/0mCxXAoTbO

[QCon Software Development Conferences @QCon]

LinkedIn checks whether suggestions reach merged code, not just whether reviewers react to them. More than 60% of sampled suggestions are accepted. The distinction is between generating feedback and producing feedback engineers can use.

[QCon Software Development Conferences @QCon]

More AI-generated code changes the review problem. Min Chen, Principal Staff Engineer@LinkedIn, explains LinkedIn's AI review system at QCon AI New York 2026.

Read the session: https://t.co/WYVnvGNOxz

[QCon Software Development Conferences @QCon]

The design uses multiple review agents, repository-specific context, and validation to filter low-confidence or redundant findings. Review timing matters too: feedback has to arrive before human review begins.