Abstract
Running AI today often means choosing between Anthropic or OpenAI, and accepting the cost and privacy concerns that come with shipping data to third-parties. What if there was another way? Tools like Transformers.js and WebLLM have progressed and it’s now possible to do useful inference locally - keeping sensitive data off the network entirely, cutting token costs, and building things that work without a server in the loop.
If you've ever called an LLM via API and wondered whether you actually needed to, this talk is for you.
James Hall, creator of jsPDF, shares the lessons from building browser-native AI - the demos that worked, the early Chrome APIs that didn’t.
In this talk you'll learn:
- When local inference beats a cloud API call, and how to benchmark the difference on your actual workloads
- How to prevent PII from ever leaving a machine
- The pitfalls that will catch you out and how to design around them
- How to leverage WebGPU as an in-browser data scientist
QCon London 2026 is a three day conference for senior software engineers, architects and team leads. An international program committee of working engineers selects every session. Patterns and practices, not products and pitches.
Part of the track
Emerging Trends in the Frontend and Mobile Hosted by Ian Thomas Software Engineer @Meta, QCon London & San Francisco Co-Chair, International SpeakerFrom the same track
Wednesday 18 March
10:35 Mountbatten (6th Fl.) Session AI Tools That Enable the Next 1B Developers Ivan Zarea Director of Platform Engineering @Netlify The developer population is growing by orders of magnitude. The new AI tooling is turning domain experts and operators into builders. Unfortunately, the tools and the platforms they are building on weren't designed for them. 11:45 Mountbatten (6th Fl.) Session GUI Agents Computer Use Agents: The Frontier of Vision-Based Automation Stefan Dirnstorfer CTO @Thetaris GmbH, Architect of ThetaML & Thetaris’ Testing Application, 42 Years into Software Computer Use Agents represent a paradigm shift in software interaction: AI models trained to operate interfaces visually, mimicking human interaction rather than relying on technical APIs. 13:35 Windsor (5th Fl.) Session Panel: Who Builds the Frontend Now? (And What Breaks When They Do) Luca Mezzalira, James Hall, Danielle An, Ivan Zarea This panel is for frontend and mobile engineers trying to make sense of what's actually happening versus what's hype. Practitioners from platform infrastructure, observability, serverless architecture, digital consultancy, and mobile development will share what they're seeing on the ground and debate what it means. 14:45 Windsor (5th Fl.) Session GenAI Architecting AI Driven Game Creation: From Research to Production Scale Systems Danielle An Principal Engineer / GenAI Architect @Meta, Ph.D. with 15 years of professional experience in film, MR and gaming I spent the last two years as an architect of the next mobile gaming platform at meta, with the ambition that anyone can create, share, and play AI-generated mobile games. What I learned is that the hard part is no longer the generation — it's everything that happens after. 15:55 Mountbatten (6th Fl.) Session Running AI at the Edge: Running Real Workloads Directly in the Browser James Hall Founder and Director @Parallax, Author of jsPDF Running AI today often means choosing between Anthropic or OpenAI, and accepting the cost and privacy concerns that come with shipping data to third-parties. What if there was another way?