QCon Software Development Conferences

In an Anthropic study, participants learning a new Python library with AI assistance scored 50% on a mastery quiz, compared with 67% for participants coding without AI.

A staging environment can reproduce the software and still miss the behavior of the data. Lada Indra, Head of Data Platform at Pleo, on using a shadow environment to find data-dependent failures before they reach production.

JPMorgan didn’t mandate adoption of its internal AI tool, LLM Suite. Matthew Skelton, CEO & Principal at Conflux and Co-Author of Team Topologies, on opt-in adoption and “friendly FOMO” instead of a top-down rollout.

Deployment freezes, saturation, and incident severity shape resilience decisions. Vanessa Huerta Granda hosts this track at QCon San Francisco. View the sessions: buff.ly/YHft7nw

At QCon AI New York, Nancy Wang shares 1Password's approach to agent authorization. The program also leaves time to question speakers about production decisions. Explore the sessions: buff.ly/Pt1hGF2

Is another prompt instruction fixing your agent or adding debt? @dbreunig demonstrates four context failures and practical fixes at QCon San Francisco. Read the abstract: buff.ly/wGVK7G8

Agents running for hours need state, recovery, and approvals. QCon AI Boston examines these production design choices, July 8–9, 2027. Explore the program areas: buff.ly/47kUC1t

Some of the biggest improvements in AI coding tools haven't come from the model alone. Alasdair Allan, CTO at Negroni Venture Studios, on the agentic loops and structured feedback that help coding agents correct themselves.

In 2015, broken data might have meant a wrong dashboard and an annoyed executive. Lada Indra, Head of Data Platform at Pleo, on why data quality has moved from a reporting concern into the production path.