QCon AI Boston 2026: The Future of AI Engineering Awaits
QCon AI Boston 2026, happening from June 1–2 at Boston University’s George Sherman Union, is just two weeks away, and excitement is building as the event nears a sellout. This year’s conference features an impressive lineup of over 40 sessions diving deep into AI engineering challenges. Particularly intriguing are six highlighted sessions centered around evolving AI capabilities beyond prototype stages—exploring what AI engineering looks like after extensive real-world testing.
Keeping ChatGPT Fast in the Agentic Era
Martin Spier, Performance Engineer at OpenAI, will kick off Day One with an insightful keynote focused on a persistent misconception: AI application latency isn’t merely a problem of GPU capacity. Spier will dissect the lifecycle of a single user request, which traverses multiple layers including client work, conversation loading, context assembly, tokenization, routing, inference, streaming, and observability. An issue in any one of these stages can become a bottleneck, severely affecting performance.
In a world where agentic coding enables faster shipping, it also brings the risk of accumulating performance regressions more rapidly. Spier will elucidate how OpenAI is innovating in performance engineering, using telemetry and tools that empower agents to conduct investigations autonomously.
Context Engineering at LinkedIn
Enter Ajay Prakash, Senior Staff Software Engineer at LinkedIn, who will share a compelling approach to build a robust organizational context layer for AI agents via CAPT—a MCP-based solution. Prakash will explain how coding agents often operate without understanding the unique frameworks, services, and data systems that define a company’s workflows.
His session will recount LinkedIn’s journey of deploying MCP across its engineering departments, addressing initial failures, and evolving the system. Remarkably, this initiative resulted in a 70% faster issue triage and the establishment of over 500 community-authored skills.
More Read
The Agent Harness: Control Planes and Production Strategies
In an era of apparent autonomy, Vinoth Govindarajan from OpenAI will explore a crucial topic: the reliability of AI agents hinges on the harness encasing the model. His session will introduce vital elements including control planes, auditing mechanisms, and approval paths.
Through the lens of the case study OpenClaw, Govindarajan will clarify how managing event flow, session state, execution constraints, and audit trails critically enhances the reliability of AI applications. These considerations extend beyond mere model features—highlighting the importance of systems-level thinking in AI engineering.
Building Reusable Evaluation Frameworks
Susan Chang, Principal Data Scientist at Elastic, will discuss the benefits of building a centralized evaluation framework for AI agents. With nearly two years of running a user-facing AI agent in production—far longer than many current experiences—Chang will share her methodologies for evaluating agent performance and integrating feedback loops into product improvements.
This session offers invaluable insights for any team navigating the challenges of aligning evaluation patterns with the specific failure modes of their systems, making it essential for AI practitioners.
Building a GenAI Platform at DoorDash
Siddharth Kodwani and Swaroop Chitlur from DoorDash will address a common pain point: teams redundantly rebuilding the same foundational Large Language Model (LLM) infrastructure. Their talk will provide a behind-the-scenes look at how DoorDash consolidated various functionalities—retry logic, fallback mechanisms, cost tracking, and more—into shared platform components that streamline operations.
Rodwani and Chitlur will not only delve into the construction of components like the LLM Gateway and Batch Inference Platform but also pose critical questions regarding the balance between shared infrastructure benefits and potential overhead.
From Prompt to Production: Engineering an Autonomous SDLC
Finally, Andrew Swerdlow, Sr. Director of Software at Roblox, will tackle the challenge that arises when more generated code does not necessarily translate to quicker software delivery. Swerdlow’s powerful insights focus on redesigning the Software Development Life Cycle (SDLC) to optimize production quality and speed.
By leveraging autonomous agents to manage code migrations and maintenance, Roblox is pioneering an approach named Exemplar Alignment, which grounds agents in expert engineering judgment. Ultimately, this talk will grapple with the pressing issue of quality assessment—how to measure software quality when agents generate code but humans oversee outcomes.
QCon AI Boston 2026 promises a rich tapestry of insights focused on the real-world challenges of AI engineering. With a comprehensive schedule showcasing industry pioneers, attendees can expect to deepen their understanding of AI technologies and their intricate engineering demands. For complete session details and to secure your spot, visit boston.qcon.ai.
Inspired by: Source
