Who is naming these model releases? Gemini’s new image editing model is called Nano Banana (and they dropped a great tutorial on how to use it as a developer). Meanwhile, Microsoft AI launched two models in the new MAI series, presumably pronounced “my” because they’re not OpenAI models. At least OpenAI, the creators of gpt-4o-mini and o4-mini, were a bit more sensible with their recent release of gpt-realtime along with the generally available Realtime API. And that’s only some of the newly model and API releases in the last couple of weeks.

In AI++ today there are competing views on whether multi-agent systems are the right solution, some diverse examples of Langflow-powered applications and many lessons on building agents. For fans of the game werewolf, you can also find out which model is best at playing.

Phil Nash
Developer relations engineer for Langflow

🛠️ Building with AI, Agents & MCP

​Introducing the MCP Registry

The official MCP registry is an open catalog and API for publicly available MCP servers. You can register MCP servers that you've built, and there is an API to access the data. I noticed there’s no MCP server for the MCP server registry though...

On a related note, this article breaks down and implements an MCP server over STDIO using only standard Java libraries.

To build multi-agent systems or not

There seems to be splits in the community over how to build powerful agent systems. Some people espouse multi-agent systems, like in LangChain’s Open SWE coding agent, or these lessons building an AI data analyst. On the other side are the Cline team, who describe mutli-agent orchestration as a “seductive trap”, and Cognition who simply say don’t build mutli-agents.

Building with Langflow

Check out this project to build an AI-powered, Notion note organizer, get into smart model routing, and build your own GPT-5, or combine Langflow, Next.js, and multimodal models to create an alt text generator to help improve accessibility.

Lessons in agent building

Have you got some spare reading time? This is a huge guide, three whole books-worth, on building agentic systems. And here’s a five-post series on context engineering. On the (slightly) shorter side here’s an article on building a CLI coding agent with Pydantic-AI, some lessons on how to build agents with small language models, and a quick article on how a small, focused agent can be more predictable than trying to one-shot a feature with an LLM.

🗞️ Other news

• A team had agents play werewolf against each other, find out which model was the best! 🐺
• If you thought prompt injection was dangerous, what if I told you that it could be turned into an AI-powered virus?
• This is a visualization that shows you step-by-step how an LLM actually works.
• The easiest way to compare model prices: https://pricepertoken.com/
• ​The PM’s guide to AI agent architecture is a thorough approach to understanding and building AI products
• How do you start to evaluate agents?
• And once you’ve built your agent, what are the KPIs you should be tracking?
• The latest episode of The Flow is out on YouTube, Spotify, and Apple Podcasts​

🧑‍💻 Code & Libraries

• There’s a lot of buzz around MCP-UI, sharing interactive UI could be the next stage for MCP interactions
• ​LangChain & LangGraph 1.0 alpha has been released for testing
• This repo is called awesome-agent-failures and that’s all you need to know
• Vercel built an open-source vibe coding platform that you can fork and build on

🔦 Langflow Spotlight

The Prompt Template is one of the core components of Langflow and it features in almost every flow I build, so you should get to know it. It’s a seemingly simple component, initially just a template input.

When you add a mixture of static text and curly brace surrounded variables you can add as many external inputs as you need. You can bring together message history, user queries, and any other context that will be useful to your agents.

🗓️ Events

September 10th – AI-Driven Development Day is a free online event digging into AI workflows, prompting techniques and agent strategies that make you a better developer with these new tools. Tejas and Phil from the Langflow team will be presenting. If you can't make the event, register to receive the recordings.

September 18th–19th – Combine web, AI and community at CascadiaJS 2025 in Seattle. The Langflow team will be there, and you can get tickets for a magical 50% off with the promo code LANGFLOW_50.

September 20th – The AI party keeps rolling with the Cascadia AI Hack Day in Seattle. Hang out for a day to see what you can build using AI, Agents and MCP.

October 15th–16th – join the Langflow team at AI for the rest of us in London. Promising two days of inspiring and practical sessions that demystify jargon, share real world stories of AI in action, and include developer focused sessions and hands-on workshops, this will be a fantastic community event.