I was super happy to come back and speak again at AI Builders Amsterdam, this time showing that "the real second brain was the skills we made along the way". We explored what a second brain can be in context of an LLM and I shared how we've been optimising ours at @getstream_io! (With a minor detour to Mesopotamia, 8000 BCE.) I love the energy and curiosity of people I meet at AI Builders events (this was my third) and it's great to connect with people about topics we all care about. Can't wait for the next one! #secondbrain #ai #agentskills #aibuilders

Had an *amazing* time on Tuesday speaking at the @ClaudeDevs Code community meetup in Amsterdam! So many great discussions about improving a second brain system with interesting, engaged people building really cool projects. Doing more talks again recently has reminded me how much I love public speaking as a way to connect with people. And it turns out that "the real second brain was the skills we made along the way" 🤷‍♂️ It was great to discuss what we're doing to help AI agents like @claudeai to help you get the best results when you build with @getstream_io, and even better to take everyone through 44,000 years of second brain history! Thanks Roman for inviting me and to you and Jenny for running such a great event! Can't wait to be back 🙌 #ai #claude #secondbrain #skills

Super fun time yesterday at the AI Builders Amsterdam meetup! I spoke about why your (yes, your) docs make me angry, and helped 90 people improve their docs from the perspective of an LLM as well as a human. Thanks AI Builders for having me and @getstream_io for sponsoring!

Package theft happens every day, but what if we could do something about it? 🤔 I built a live security camera with @getstream_io Vision Agents SDK that uses @GoogleDeepMind Gemini + @NanoBanana (thanks @thorwebdev 😆), @elevenlabs, @DeepgramAI and the @X API, with a custom @ultralytics YOLOv11 model trained using @Meta SAM3 with the @roboflow platform to single-handedly save the universe. Example code in 🧵. Imagine... what else you could build with this?

🚨 WANTED: John caught "stealing" a package! AI-powered security #VisionAgents

openai realtime had a consistently lower time-to-first-token, though this is likely because we connect to openai via webrtc and gemini via websocket

here's the repo with the full writeup and the code sample you can run yourself: github.com/GetStream/Vision-… in the next reply i'll compare openai realtime and gemini live and here's a video of a different attempt 🤦‍♂️

we had fun trying to make a real-time sports commentator by joining together @roboflow inference with @OpenAIDevs Realtime and @GoogleDeepMind Gemini with the @getstream_io vision agents sdk this helped us understand the exact state of video AI at the end of 2025, where we are and what's coming next we wrote this up, with metrics comparing OpenAI vs. Gemini in our setup - link to the full article and more of our outtakes in the 🧵 this video was our first attempt, sharing it first so you can enjoy pinpointing exactly where in the video my soul leaves my body

when you see the sphinx, either answer 3 riddles or distract it with a blåhaj and run

aww enough people have posted about this that it was able to find its error online :/

why stop there

true

looks like you're starting everywhere

stop with the homework i'm too dumb for that

i think this speaks for itself here:

here, we can share this nice pile

cloudflare has been so unreliable! we should roll it back to before the rebrand

bröther look into the light

just make sure you don't end up with something like codex's model selection in cursor

. @OpenAI @cursor_ai can I get the gpt-5 codex venti caramel macchiato with cocoa dusting pls

ONE Bob Dylan album

brb quitting my job

brb quitting my job

brb quitting my job

This reminds me that Brian David Gilbert did a great video about this piped.video/watch?v=-wpHszfn…

The open-source vision agents SDK from @getstream_io now supports @openrouter, so you can switch between models and providers just by changing the model name - everything else in your app stays the same!

.@OpenAIDevs dominates @geoguessr but it's trained on Google Street View - same as the game. Can it *actually* recognise real-world locations? We need a side-by-side comparison with a game location. I did some investigative journalism with @getstream_io ...did I go too far?

Because @OpenAIDevs realtime API is so good at guessing locations on @geoguessr here's a compilation, including my actual location at the end 🫢 Heavily inspired by @vedal and @DougDougFood's collab as you can probably tell from the first guess 😅 Made with @getstream_io vision agents OOTB just by changing the system prompt!

The code is actually super simple, just copied this demo and changed the instructions github.com/GetStream/Vision-… ⭐️

Flippin eck @OpenAIDevs gpt-realtime is REALLY, REALLY good at @geoguessr Cooked up this bot using our Vision AI SDK, almost as good as @georainbolt 😜

Today in "I made my AI code reviewer summarise in the form of dark poetry"... what's the change in this PR?

the work I'm possibly proudest of this year is this one-line configuration change I made to our @coderabbitai repo integration

Wanted to test the emotional range of the new @OpenAI @OpenAIDevs gpt-realtime model... it's certainly a wider range than before 😆 it's super easy with a @getstream_io video call, it's a drop-in replacement with the new model name!

I wrote this @googleaidevs #geminilive #Python #AI plugin where I have the power to interrupt a bot on a @getstream_io video call!

Built an #MCP demo with @getstream_io Python AI SDK, @DeepgramAI STT, @elevenlabs TTS and @OpenAI LLM. You can use Stream's AI plugins to define your pipeline and build an end-to-end bot. This one has MCP (with #fastmcp) too! Code available on GitHub.

Also, who needs figma or illustrator when you have keynote to draw your technical diagrams? 😆