We Were at Dev Week NYC. Here Is Everything You Need to Know.

Last week, the CreateHER Fest team headed to New York City for DeveloperWeek NYC and AI DevSummit, one of the largest developer conferences on the East Coast. Two days. Dozens of sessions. And one very good coffee conversation at the Paris Cafe inside the iconic TWA Hotel.

We went so you did not have to start from scratch. Here is the full breakdown.

The State of AI in 2026: Honest, Urgent, and Moving Fast

The energy at Dev Week this year was different. Less hype, more accountability. The sessions that packed rooms were not about what AI can theoretically do. They were about what breaks when you try to ship it, who is responsible when it gets misused, and how to architect systems that actually hold up under real conditions.

That is exactly the conversation our community needs to be part of.

Session Breakdowns

The Model Delivery Network: Patterns for Faster Inference at Scale

Tirth Vyas, Sr. Product Manager at Runpod

Before your model can do anything, it has to load. And for many teams, that loading time is the silent killer of user experience. Tirth walked through how Runpod scaled inference for over 750,000 developers by treating model weights as infrastructure, not static files. Content-addressed storage, tiered caching, and deferred compute are the patterns turning cold starts from a bottleneck into a near-instant operation.

Key takeaway: If you are building or deploying AI products, the delivery layer is not an engineering afterthought. It is a product decision.

KEYNOTE: Beyond the Uncanny Valley: Building Resilient Defenses

Alex Lisle, CTO at Reality Defender

This was one of the most important talks of the conference. Alex made the case that traditional deepfake detection, which relies primarily on facial analysis, is already obsolete. Modern generative AI creates multimodal impersonations across audio, image, and video simultaneously. The only way to keep up is with multi-model, context-aware detection systems that are continuously updated as generation tools evolve.

Key takeaway: The arms race between generative AI and detection is real and accelerating. Understanding how these systems work is not just for security teams. It is foundational literacy for anyone building with or alongside AI.

From Intent-Based to Reasoning-Based: Migrating Assistant Actions to Agents

Akhil Sharma, Sr. SDE at Meta and Sudipta Mohapatra, Senior Software Engineer at ChartHop

The traditional intent-mapping model that most voice and assistant products were built on is being replaced. This session showed how to migrate existing assistant actions into autonomous Gemini-powered agents using the Genkit Go v1.4 framework, with function calling replacing intent classification and stateful, non-linear conversation handling as the new standard.

Key takeaway: If you have existing assistant or chatbot products, the architecture is shifting underneath you. Now is the time to understand the new model, not after your users notice the gap.

Identity Tips to Build Secure AI Agents and MCP Servers

Mrunank Pawar, DevRel Engineer at Descope

Every team has an MCP or agentic AI project right now. Very few of them are in production. Mrunank argued that authentication, access control, and agentic identity are the primary reasons, and walked through actionable patterns to close those gaps. If your agent can retrieve sensitive data and pass it downstream, you need an identity strategy before you ship.

Key takeaway: Security is not a final step in the development process. For AI agents, it is the foundation. Build it in from day one.

The Enterprise AI Agent Playbook: What Nobody Tells You Before You Ship to Production

Kothandaraman Varadarajan, Product Manager at Nordstrom

This session was the most practical of the conference. Kothandaraman covered three failure points that kill AI agent projects: vague system prompts, knowledge bases that are just piles of documents, and a gap between stakeholder expectations and actual use case definitions. He shared templates and frameworks to close all three gaps, and the room was writing the entire time.

Key takeaway: AI agents fail in production because of process and communication problems, not just technical ones. Bridging the gap between "we want an AI agent" and "here is exactly what it needs to do" is its own skill.

From Prompt Engineering to Skill Engineering: Building AI Skill Agents

Seetaram Rayarao, VP Senior Lead Engineer at JP Morgan Chase

This was the paradigm-shift talk of the conference. Seetaram introduced Skill Engineering, a new standard for building modular, portable AI capabilities that any compatible agent can discover and use on demand. Instead of loading one massive prompt with everything an agent might need, you build discrete skills that are called when relevant. Less noise, lower cost, better performance.

Key takeaway: Prompt engineering got us here. Skill engineering is where we are going. Understanding this shift now puts you ahead of most teams building with AI today.

The Bigger Picture: What Dev Week NYC Told Us About Where AI Is Headed

A few themes showed up in almost every session we attended:

• Production readiness is the real barrier. The technology works. The process, security, and architecture around it is where teams are struggling.
• Security cannot be bolted on. From deepfake detection to agentic identity, the sessions made clear that security thinking has to be baked in from the start, not added after something goes wrong.
• The architecture is changing underneath us. Intent mapping, mega-prompts, static model deployment, all of it is being replaced. The teams that understand the new patterns now will have a significant advantage.
• Community and proximity still matter. Some of the best insights from the week came not from the main stage but from hallway conversations and our own coffee meetup at the Paris Cafe inside the TWA Hotel. Being in the room, in person, with people building at the frontier is irreplaceable.

A Note on Community

We want to give a genuine thank you to our sponsor Vonage for supporting our presence at Dev Week NYC. Events like this cost resources, and having partners who believe in putting women and non-binary technologists in these spaces is what makes it possible.

If you were not able to make it this year, that is exactly what CreateHER Fest is built for. We bring the knowledge, the access, and the community to you.

Stay tuned for more from our team on the ground.

The future is being built right now. You belong in that conversation.