OpenAI dominated the AI headlines last week with the release of new open weights models (their first open release since GPT-2) and GPT-5. There’s been some user backlash against GPT-5 replacing all the models in ChatGPT, but as developers we can still pick the GPT we want to use. Not wanting to feel left out, Anthropic released Claude Opus 4.1 and the Qwen team impressed with Qwen-Image.
This week in AI++ we have some great open-source examples of agents, we dive into UX and accessibility for agent interfaces, and run models like gpt-oss locally with Ollama.
Phil Nash
Developer relations engineer for Langflow
🛠️ Building with AI, Agents & MCP
The new OpenAI models are out, but can you run them in Langflow? Of course you can!
I love digging into example applications to see how they are built, and LangChain has produced two such agents to take inspiration from: an asynchronous coding agent and a deep research agent framework. Check out the Langflow and JigsawStack take on a deep research agent too.
When developing agents there is a lot to consider in how the application works, but it’s just as important to consider how a user is going to interact with it. This collection of patterns captures many of the common scenarios giving names and examples. On a similar note, this article from Mike Gower in IBM design on Usable AI Chat discusses some of the differences between chatting with a person compared to an AI.
Runno is a project that lets you run code written in many different languages in a web assembly sandbox. It’s a safe way to run untrusted (generated?) code. Ben also wrote an MCP server for Runno, so you can provide your agents with a safe place to run code without messing around with containers. Check it out.
Have you ever asked questions like “How will MCP help everyday internet users?” or “Are MCP servers secure?“
If you have, you’ll definitely enjoy this in depth FAQ on MCP from the fine folks at Vercel.
🗞️ Other news
The more people build successful agents, the more we can all learn from them.
From “An agent should always disclose that it's an agent” to “An agent cannot be held accountable” these are some principles you might want to consider in your own agents.
It’s really fascinating seeing what works and what doesn’t with prompts and this collection of user generated prompts that others try and then rate is a good way to find much inspiration.
🧑💻 Code & Libraries
- AI Elements is a component library built on shadcn/ui that provides a bunch of useful components for AI conversations.
- Open Lovable is an open-source Lovable clone
🔦 Langflow Spotlight
If you want to build completely local agents, using an open model like gpt-oss-20b, say, then combining Langflow with Ollama is one of the best ways to go about it. Get to know the Ollama component and you’ll only be held back by the power of the model you can use locally.
Check out this great walkthrough that showed agentic capabilities from Qwen2 via Ollama in Langfllow from back in March!
🗓️ Events
11am PDT, August 13th – Dominik Kundel from OpenAI joins David and Carter on The Flow to breakdown everything about gpt-oss.
Can't make the live stream? Subscribe to The Flow wherever you like to get your podcasts.
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.