Sign in

Christian Bromann

@bromann.dev
247 followers 388 following 28 posts

Running on agentic reasoning and espresso. Building with @LangChain 🦜🔗.

PostsRepliesMedia
Christian Bromann @bromann.dev · 15/01/2026
Streaming custom events to React can be boring (in the best way): wire a single handler, and every tool-emitted progress/status/file event shows up instantly. 🙅 No polling loops. ❌ No fake spinners. Just render what the agent tells you‼️
000
Christian Bromann @bromann.dev · 15/01/2026
Tools shouldn’t be silent black boxes. If a tool knows what it’s doing (step name, percent complete, current operation), it should stream that as structured events. Then your UI renders facts 👉 not vibes.
100
Christian Bromann @bromann.dev · 16/12/2025
⚡ New: Manually control strict mode when using providerStrategy! When strict: true, the model is guaranteed to return output matching your schema exactly — no extra fields, no missing fields. 🧵4/4
LangChain example showing how to apply strict: true to a providerStrategy response format of createAgent.
000
Christian Bromann @bromann.dev · 16/12/2025
🤖 @anthropic.com gets first-class tool support too! Text editor, web fetch, computer use, tool search, and MCP toolset. 🎮 See it in action: I built a demo where AI agents play Tic-Tac-Toe using Computer Use! Watch Claude see the screen, analyze the board, click etc: 👉 github.com/christian-br... 🧵3/4
LangChain example showing how to use computer use tool from Anthropic with LangChain's createAgent.
100
Christian Bromann @bromann.dev · 16/12/2025
🔍 OpenAI built-in tools are now first-class citizens in @langchain.bsky.social! Web search, file search, code interpreter, image generation, computer use, shell, and MCP connector — all work seamlessly with ChatOpenAI or @langchain.bsky.social agents. 🧵2/4
LangChain example showing how to use OpenAI webSearch tool with createAgent.
100
Christian Bromann @bromann.dev · 12/09/2025
“Run it?” Not without review. 🛑 With HITL middleware, agents surface proposed tool calls, you review the args, then ✅ approve / ✏️ edit / 🙈 ignore / 💬 respond. Policies live next to your agent; state stays coherent via graph-native interrupts. Code (alpha): github.com/langchain-ai...
000
Christian Bromann @bromann.dev · 14/05/2025
Good morning #interrrupt ☀️
000
Christian Bromann @bromann.dev · 25/04/2025
I am sorry but these labels in Playwright reports are so misleading. It's neither mobile nor Safari 😅 Something like "Emulated Mobile Viewport on (patched) WebKit" may do it but is probably to long I guess 🙈
010
Christian Bromann @bromann.dev · 31/03/2025
If there is one thing we can learn from @lauromoura.bsky.social from @igalia.com in his talk on "Advancing WebDriver BiDi support in WebKit" at #SeleniumConf is that WebKit !== Safari. You will miss browser specific quirks and bugs in your application if you think testing WebKit is "good enough" 🙈
030
Christian Bromann @bromann.dev · 27/03/2025
Really enjoying Maaret Pyhäjärvi opening statement at SeleniumConf! 👏 We’re here tackling real-world challenges—automating actual browsers and mobile apps on real devices used by real people, all built on open standards. No hype! Genuinely the best ecosystem and community on testing out there 😊
040