QCon Software Development Conferences

Agents fail in production not because of the model but because of what's missing around it. Vinoth Govindarajan (@OpenAI) on control planes, single-writer execution, and approval boundaries that keep agents predictable under load. QCon AI Boston: bit.ly/3PynDm9

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 (@Spotify) on rebuilding it as a multi-agent pipeline. QCon AI Boston, June 1-2: bit.ly/42vfpOE

QCon San Francisco is offering diversity scholarships for 2026 (Nov 16-18). Applications are open. Apply here: bit.ly/4tWBB09

From failed prompt-chains to the actor model: Manju Rajashekhar shares the agent patterns that actually survive production. QCon AI Boston, June 1-2. 🔗 bit.ly/4erDKMt

JPMorganChase runs inference at a scale where own vs. rent has real consequences. Dio Rettori on which use cases drive capacity needs and where the biggest technical challenges are. QCon AI Boston, June 1-2. Current pricing ends May 12. 🔗 bit.ly/4ewaZ1e

Premium models with low latency, or cheap inference with unpredictable quality? Aditya Mulik @Walmart Global Tech on a third path: batch multi-agent pipelines on pay-as-you-go models, with MCP grounding closing the quality gap. QCon AI Boston, June 1-2: bit.ly/4tenrq5

Last day to register for the May 7 InfoQ Certified Architect cohort. 5 weeks, small group, hands-on. ICSAET certification and an article on InfoQ. June 10 cohort still open if May doesn't work. bit.ly/4dvYznZ

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/4wj36CO

@mcuban is not wrong. But consistency is only part of the story. The real challenge shows up when AI is used to make decisions in high-stakes systems. In those environments, domain knowledge and human judgment help, but they are not enough. You still need decisions that are applied consistently and are auditable, reproducible, and explainable against policies and regulations. That is a problem that has been around for a long time. It just moved layers. It used to be built into enterprise systems. Now those same guarantees have to hold at the AI layer. Been working on this space for close to two decades. Going deeper on it at @QCon AI Boston

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/4n6s6sw

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/4n7Q9aJ

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. Registration closes May 6. bit.ly/48HwIzp

"Writing helps you crystallize what you think you know." Luca Mezzalira on why the InfoQ Certified Architect cohort includes publishing on InfoQ, and why writing three books taught him more than any other format. May and June cohorts open: bit.ly/4tJPte4

How do you move architectural decisions out of a central team without losing consistency? @lucamezzalira on what the InfoQ Certified Architect cohort works through, and why some attendees applied the approach the same week. May and June cohorts open: bit.ly/4mXloFc

QCon San Francisco 2026 tracks are out. 12 tracks, 60+ speakers. Architecting for agents, guardrails and safety nets (evals), staff+ engineering skills, DevEx as an architecture concern, and more. Nov 16-20. Current pricing ends May 12. 🔗 bit.ly/4uicYKT

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/42EgweI

Forter gave 200 people two weeks to build their own AI agents. They built 80. 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/41TycCX

I'm 2 weeks into facilitating the April @InfoQ architecture cohort, and something keeps coming up that I want to share. Most architects know what they're building. Fewer can articulate why, and fewer still can surface the tradeoffs of a decision in a way that lands across an entire organisation. That gap is where good architecture goes quiet. The cohort works through tools that force that clarity. Architecture canvases to make the thinking visible. ADRs and RFCs to externalise the reasoning and invite challenge. Decentralisation patterns to understand where decisions should actually live inside a sociotechnical system. None of these are documentation exercises. They are communication mechanisms. They change how you frame a decision before you've even opened your mouth. What emerges after 5 weeks is something I see consistently: participants start understanding not just what they are building, but why, and how to surface the tradeoffs honestly. Because in architecture there are no perfect decisions. There are only the least wrong ones for your context, your team topology, and the constraints you're operating under. Being able to communicate that, to leadership, to peers, to teams, is what separates an architect from a senior engineer with opinions. Chinmay Sawaji at Klaviyo said it directly: "I can now articulate trade-offs in a way that improves my proposals at work." That's not a curriculum outcome. It's a side effect of having to be precise with people who won't fill in the gaps for you. The May cohort starts 7 May. I'd like to see some of you in it 😉 Details in the comments 👇

25 trillion tokens across 1.5M developers. Three places teams stall out: context construction, model routing, feedback loops. Brian Turcotte @KiloCode on what the data says about AI agent adoption, the stuff benchmarks miss. QCon AI Boston, June 1-2. 🔗 bit.ly/3OzNAkQ

People join the online architect cohort expecting lectures. It’s nothing like that. Luca Mezzalira on what actually happens: architects from finance, health, and media solving each other's problems. May and June cohorts open: bit.ly/3OGq7hL