Day 2 #QConAI Boston 🚀 💡 How engineering teams are measuring & growing AI maturity 💡 Multi-agent platforms built for real org complexity (Spotify, OpenAI) 💡 AI infrastructure: GPU markets, LLM ranking systems &cost transparency 💡 What 25 trillion tokens tell us about how developers actually adopt AI

Halfway through Day 1 at #QConAI Boston, a few things are standing out: 🔹 Context is becoming an architecture problem, not just a prompt problem 🔹 Trust and safety at scale needs real infrastructure, not just policies 🔹 KV cache is one of those levers teams notice when inference costs start to hurt

A theme showing up across #QConAI Boston today: production AI is mostly a systems problem now. The model still matters, but the harder questions are around cost, latency, context, compliance, security & scaling AI work beyond a few specialists: boston.qcon.ai/schedule/bos... #AIConference

🎤 Kicking off Day 1 #QConAI Boston with keynote from Martin Spier, Engineering Leader OpenAI: "Keeping ChatGPT Fast in the Agentic Era." Martin dives into the infrastructure and systems thinking behind keeping one of the world's most-used AI products fast at scale: shorturl.at/MctwH #AIConference

Day 1 of #QConAI Boston 2026 is ON! 🎉 Two days of practical, production-focused AI, real systems, real teams, real constraints. Let's explore what practitioners are doing to ship AI safely, scale inference reliably, and build the agentic systems that actually hold up in production: boston.qcon.ai

Big technical transitions get carried by Staff+ engineers. What that work looks like in practice: the Staff+ Engineering track at QCon San Francisco 2026. bit.ly/3Q7EpJ2

Someone has to make the platform call, ship the first agent, and build the eval strategy that holds under real traffic. The InfoQ Certified AI Engineering Program is for staff engineers, engineering managers, and ML platform leads. Starts July 25, 2026. → bit.ly/4tT2HEp

Platform and value-stream design are hard to check from inside your org. The Organizational Architect cohort with Manuel Pais opens June 19: bit.ly/3PoYwlS

Your AI prototype worked. Then real users showed up with questions, load, and edge cases you didn't anticipate. The InfoQ Certified AI Engineering Program starts July 25, 2026. 5 weeks online with Hien Luu, author of MLOps with Ray. 🔗 bit.ly/43q0yoT

When ChatGPT feels slow, users blame the model. The real bottleneck is usually something else entirely. Martin Spier (Head of ChatGPT Performance @OpenAI) on full-stack AI latency and agents that investigate their own regressions. QCon AI Boston, June 1–2: bit.ly/3Rce3WH

~90% of enterprise AI agents stall at POC. The problem: no systematic way to know when an agent is ready. Zhou (Jo) Yu (CS Professor @Columbia, Founder @Arklex.ai) on multi-turn simulation and eval methods that break the demo loop. QCon AI Boston, June 1–2: bit.ly/4dWzxi3

Training days are available at QCon San Francisco 2026 (November 19–20), right after the conference. Hands-on workshops from senior practitioners. Smaller groups, focused topics, time to actually work through the material. Add to your pass or attend standalone. Spots are limited: bit.ly/4uSEFup

Agents fail in production not because of the model but because of what's missing around it. Vinoth Govindarajan (MTS @OpenAI) on control planes, single-writer execution, and approval boundaries that keep agents predictable under load. QCon AI Boston, June 1–2. Find out more: bit.ly/4dqq4Pz

Spotify's ad platform: audience targeting, creative gen, campaign goals, and budget optimization. A single agent for all of it = slow and hard to debug. Pratik Rasam, Senior Engineer @Spotify, on rebuilding it as a multi-agent pipeline. QCon AI Boston, Jun 1-2: bit.ly/4dFpFdi

They built an agent system with a prompt-chain framework. It demoed well. Then real workloads hit: tool calls lasting hours, workflows that needed to survive restarts. They threw it out and rebuilt on the actor model. Manju Rajashekhar on the patterns that survived: bit.ly/4wf7P8l

Fabiane Nardon, Data Platform Director @TOTVS, on the data layer beneath reliable agent systems, and the token economics that quietly kill most projects. QCon AI Boston, June 1-2. Current pricing ends May 12. 🔗 bit.ly/4cPGlP2

The QCon San Francisco hotel block at the Hyatt Regency is already 65% booked, six months out. Staying at the venue keeps you in the conversation between sessions. Nov 16-18. Early bird May 12. 🔗 bit.ly/4dn2iFk

When an agent approves a loan or classifies risk, the output can't be probabilistic. Alex Porcelli on how a financial services enterprise runs decision models as agent skills, processing millions of evaluations daily. QCon AI Boston, June 1-2. 🔗 bit.ly/3P3UsqH

Four hours a week sounds like a lot. Architects who've done the InfoQ Certified Architect cohort say it replaced unstructured thinking they were already doing, with external feedback they couldn't get internally. May cohort starts May 7. Registration closes May 6. 🔗 bit.ly/3PlfzF9

Rachel Shalom, Distinguished Engineer @Dell, on embedding models, diagnosing failure modes and when fine-tuning is worth it, from Dell's agentic search system. QCon AI Boston, June 1-2. Current pricing ends May 12. 🔗 bit.ly/3PaPBUM

QCon San Francisco 2026 tracks are out. 12 tracks, 60+ speakers. Architecting for Agents, Guardrails and Safety Nets (evals), Staff+ Engineering Skills, Developer Experience as an architecture concern, and more. Nov 16-18. Current pricing ends May 12. 🔗 bit.ly/4t8DntM

Forter gave 200 people two weeks to build their own AI agents. They built 80. Engineers, PMs, analysts. Ben Maraney, Principal Engineer @Forter, on the minimum viable platform and why their MCP server was the biggest force multiplier. QCon AI Boston, June 1-2. 🔗 bit.ly/4d7zEb5

"I go to companies that are over-engineering their systems because they're not asking the right questions." Luca Mezzalira on what the in-person Certified Architect cohort at QCon San Francisco covers: ADRs, team topologies, trade-off mapping, and Architecture Decision Forums. 🔗 bit.ly/4tTNR0M

Red Hat AI's 500-engineer org is adopting agents across their SDLC. They started measuring speed. Their engineers pushed back. Catherine Weeks on what they measure now: including a security pipeline that took 4 iterations to go from 58% false positives to 22%. 🔗 bit.ly/4cs8bkk

Partial schedule for QCon AI Boston (June 1-2) is live. Sessions from DoorDash, LinkedIn, Apple, Red Hat on agents, context engineering, AI safety at marketplace scale, and more. Early bird ends May 12. 🔗 bit.ly/4euJi8L

"Code is becoming a commodity. Good code for your context isn't." Luca Mezzalira on why every senior engineer now needs to think like an architect. A full week of the cohort covers it. May and June cohorts open: bit.ly/4u3YT3T

AI agents don't know your company. Ajay Prakash, Senior Staff @LinkedIn, on how CAPT gave 500+ skills to engineers, PMs, and designers via MCP, with a live demo. QCon AI Boston, June 1-2. Early bird ends May 12. 🔗 bit.ly/48He2jc

Direct LLM-to-SQL doesn’t hold up at enterprise scale. Francesca Lazzeri (Microsoft) shares a hybrid architecture that combines LLM reasoning with deterministic computation for reliable query generation. QCon AI Boston, June 1-2. Register: bit.ly/4dQ8iHl

Reliable LLM outputs in production require more than prompting. Ricardo Ferreira (Redis, ex-AWS, Elastic, Confluent) covers context engineering as an architectural discipline at QCon AI Boston, June 1-2. Register: bit.ly/484fBYf

Early in your career? The QCon AI Boston volunteer program offers free access to sessions and recordings in exchange for onsite support. June 1-2, Boston. 🔗 bit.ly/4tRwvC1

Moving an AI agent from prototype to production introduces state management and framework lock-in problems most tooling wasn’t built for. Deepak Chandramouli and Bhumik Thakkar @Apple share how they solved it with Ray. QCon AI Boston, June 1-2. Early bird ends today. 🔗 bit.ly/4elqp88

Registration for the April architecture online cohort closes Monday, April 13. Join Luca Mezzalira and a small group of senior peers to pressure-test your technical strategy. Starts April 15. Secure your spot: bit.ly/4tFcazC

Early in your career? The QCon San Francisco 2026 volunteer program gives you free access to sessions and recordings in exchange for onsite support. Conference: Nov 16-18, San Francisco. Apply: bit.ly/4sZ6sIX

AI infrastructure decisions are being shaped by constraints most teams can’t see from the inside. Jordan Nanos @SemiAnalysis breaks down the hardware and economics at QCon AI Boston, June 1-2, Boston. Early bird ends April 14. 🔗 bit.ly/4vfbBhu

Justifying architectural trade-offs to stakeholders requires more than technical intuition. The InfoQ Certified Architect cohort covers the Architecture Canvas, ADRs, and RFC frameworks. Starts April 15. Find out more: bit.ly/4vh0lkz

AI doesn't just speed up delivery; it exposes technical debt. DORA data from 5,000 professionals shows AI surfaces dysfunction faster than teams can pay it down. Get practitioner-led strategies at QCon AI Boston and QCon San Francisco. Early bird ends April 14: bit.ly/4dG7Ew9

Most certifications end with a quiz. This one ends with a capstone assessment applied to a real architectural challenge. 5 weeks, live sessions, led by Luca Mezzalira. Starts April 15. Find out more at: bit.ly/4tByZEs

Higher AI adoption, no corresponding improvement in delivery. A lot of teams are stuck in that gap. Lizzie Matusov @Quotient shares a research-driven maturity model at QCon AI Boston, June 1-2. Early bird ends April 14: bit.ly/4c43QCg

A lot of field-tested architectural knowledge never leaves the company where it was built. In this 5-week cohort, you'll write a peer-reviewed article for InfoQ.com documenting your production trade-offs. Starts April 15. Find out more: bit.ly/3NQ4T0J

One of the most practical talks from InfoQ Dev Summit Boston last year: Phil Calçado on the gap between AI experimentation and production. Worth watching before QCon AI Boston, June 1-2. Early bird ends April 14. Watch: bit.ly/4v56dxz

AI makes code faster to write. It also makes every downstream bottleneck more expensive. Watch the Nicole Forsgren keynote on the real constraint from QCon San Francisco 2025. Worth watching before QCon San Francisco 2026, Nov 16-18. Early bird ends April 14. bit.ly/4v1zLMh

At some point in a senior technical career, the feedback loops disappear. You become the person making the calls. The InfoQ Certified Architect cohort is built around that. Starts April 15: bit.ly/4tifGj7

As AI takes on more engineering work, what does the job actually become? Michelle Brush, Engineering Director, SRE @Google, addressed it head-on at QCon San Francisco 2025. Worth watching before QCon San Francisco 2026, Nov 16-18. Early bird ends April 14. 🔗 bit.ly/4s9VnDr

Trust and safety at marketplace scale means low tolerance for errors in both directions. Bruna Pereira @DoorDash on building AI moderation that holds up in production. QCon AI Boston, June 1-2, Boston. Early bird ends April 14. ⬇️ bit.ly/4tiUoSn

LLM ranking at LinkedIn scale means strict latency budgets and hundreds of candidates per query. Sundara Ramachandran shares how they built it at QCon AI Boston, June 1-2. Early bird ends April 14. 🔗 bit.ly/4uWvTw5

Evaluation for agentic AI is harder than it looks, and rebuilding it per product doesn't scale. Susan Chang (Elastic) shares two years of production learnings at QCon AI Boston, June 1-2. Early bird ends April 14. 🔗 bit.ly/4bIwKYs

Architecture, platform engineering, AI agents — the InfoQ Certified Architect program covers how they connect. 5 weeks online. April, May, June cohorts open. 🔗 bit.ly/4byYPmc

Agent extensions can add tools, but they can't deeply shape behavior. And most are tied to a single vendor. If you've been hitting those limits, Niko Matsakis (Rust lang design team, Amazon) is building something different. QCon AI Boston, June 1-2. Early bird ends April 14. 🔗 bit.ly/4lN4oAO

QCon AI Boston (June 1-2) has a volunteer program for students and early-career professionals. Free conference access in exchange for onsite support. 🔗 bit.ly/4rPW7NS

JP Morgan Chase at QCon London: companies end up multi-cloud by accident, acquisitions, shadow IT, GPU chasing, then firefight their way out. Fix: treat it as a product. www.infoq.com/news/2026/03... #multicloud #platformengineering #opentelemetry #fintech #qconlondon