CTO @ PubNub | $134M raised | Series E | +1 Billion devices | 68 Patent claims | 25 Years
Seattle
Joined August 2008
- Tweets10.3K
- Following4.1K
- Followers4.5K
- Likes5K
Stephen Blum retweeted
Agents can build a whole app's UX now. What if stakeholders could review all the UX changes at once?
To go from a V1 to a great app, you need to iterate until it's amazing. Use Decode, work together, and get there the fastest.
Stephen Blum retweeted
Take your LangChain agent public or share it with just those people who need it most with Blocks.ai. No replatforming, no messy middleware, no extra cost to you.
How Blocks and #LangChain work together: bit.ly/4hNWzdY
Great collaboration matters more than certifications in our hiring process.
#hiring #teamwork #communication
Stephen Blum retweeted
Building #benchmarking and validation tools requires a clear vision for what you want to measure.
Product Leader Cres Hay built a blocks.ai agent validation tool with this idea in mind. She defined the input and output criteria before creating anything. Then she built her agent and set it loose to measure agent effectiveness on Blocks.
Learn how she did it, get her advice to new builders, and unpack product management in the #agentic age of #AI with her in our latest Blocks Builders series.
bit.ly/4yoDggs
AI uses less electricity to write code than a human at a desk:
Human: 3 workdays (24 hrs) @ 135W = 3.24 kWh
AI Pod: 10 mins @ 5,000W = 0.92 kWh
Even drawing 40× the peak power, the AI finishes so fast it uses 3.5× less total energy, including time it takes to correct AI mistakes
REA (reverse engineer anything) decompiled and rebuilt the mac notes app and calculator app for $0.63 in 19m10s
Stephen Blum retweeted
You don't need to switch apps to switch models. Switch to @TaskletAI and you can use them all together
Cgroups manage resources, while namespaces help isolate container processes.
#security #containers #linux
AI will create new app development roles beyond traditional engineering.
#ai #software #appdevelopment
Stephen Blum retweeted
You could call @THEDEVRELMARKUS one of our "founding fathers" as he's been running developer relations for us since the beginning.
He built a Plant Doctor agent from a Hugging Face model that blew up in a matter of days - turns out that green thumbs are contagious! Read the full builder story below.
He needed help with apartment plants, found a fine-tuned #Qwen classifier on Hugging Face, ran it locally, then used a Blocks skills file in Cursor to put the agent on the network. The model stays on his machine, the connect out lives with Blocks.ai
We interviewed @THEDEVRELMARKUS to get tips for agent builders out there. Habits worth stealing:
🔍 Look on Hugging Face, GitHub, and Blocks before you burn tokens reinventing an agent that already exists.
👩🔬 Test edge cases. A keychain photo of a plant convinced his Plant Doctor agent it was looking at a real one.
🚀 Split workflows. Scrape, summarize, and score as three agents, not one mega-agent. Cheap models for the small jobs; the heavy model only where reasoning matters.
🤖 Host skills as agents, not just files. Update once, and everyone calling it gets the new version. The catch: the same skills file does not behave the same on Opus, Fable, and Astra, so eval across models.
Read more: bit.ly/4AVTK20
