Justin
@justinhjohnson.com
Executive Director @ AstraZeneca | Nexus of Data, Science, Tech | Global Business Leader | Top Data Science Voice | #datascience #AI #buildinpublic #indiehacker
Blog rundatarun.io
BuildInPublic jandsgroupllc.com
Blog rundatarun.io
BuildInPublic jandsgroupllc.com
🧵 5/5 Announced at Universe 2025, this represents GitHub's next evolution in developer tools. Full details: bit.ly/435PZrt @github @GitHubNext @natfriedman #AI #DeveloperTools #AgentHQ
October 30, 2025 at 11:55 AM
🧵 5/5 Announced at Universe 2025, this represents GitHub's next evolution in developer tools. Full details: bit.ly/435PZrt @github @GitHubNext @natfriedman #AI #DeveloperTools #AgentHQ
🧵 4/5 The timing matters. As AI becomes essential to software development, unified orchestration becomes the competitive advantage. Teams that master this integration will ship faster and build smarter.
October 30, 2025 at 11:55 AM
🧵 4/5 The timing matters. As AI becomes essential to software development, unified orchestration becomes the competitive advantage. Teams that master this integration will ship faster and build smarter.
🧵 3/5 What makes this significant: developers gain control over generative AI, machine learning integration, and code generation—all from one interface. It eliminates the complexity of managing multiple AI systems separately.
October 30, 2025 at 11:55 AM
🧵 3/5 What makes this significant: developers gain control over generative AI, machine learning integration, and code generation—all from one interface. It eliminates the complexity of managing multiple AI systems separately.
🧵 2/5 The problem with current AI development: fragmented tools, disconnected workflows, and agents that don't work together. Agent HQ solves this by providing a single platform for orchestration across the entire development lifecycle.
October 30, 2025 at 11:55 AM
🧵 2/5 The problem with current AI development: fragmented tools, disconnected workflows, and agents that don't work together. Agent HQ solves this by providing a single platform for orchestration across the entire development lifecycle.
The breakthrough: I was using PyTorch for CUDA 12.8. System runs CUDA 13.0.
Fixed the version mismatch → 8.1 hours became 2.3 hours
Same hardware. Proper config. Documented everything.
Full technical breakdown + solutions 👇
bit.ly/47csYWh
#BuildInPublic #MachineLearning #CUDA #DGX
Fixed the version mismatch → 8.1 hours became 2.3 hours
Same hardware. Proper config. Documented everything.
Full technical breakdown + solutions 👇
bit.ly/47csYWh
#BuildInPublic #MachineLearning #CUDA #DGX
October 28, 2025 at 11:09 AM
The breakthrough: I was using PyTorch for CUDA 12.8. System runs CUDA 13.0.
Fixed the version mismatch → 8.1 hours became 2.3 hours
Same hardware. Proper config. Documented everything.
Full technical breakdown + solutions 👇
bit.ly/47csYWh
#BuildInPublic #MachineLearning #CUDA #DGX
Fixed the version mismatch → 8.1 hours became 2.3 hours
Same hardware. Proper config. Documented everything.
Full technical breakdown + solutions 👇
bit.ly/47csYWh
#BuildInPublic #MachineLearning #CUDA #DGX
Because this is how we actually work with bleeding-edge tech:
• Build in public (even when messy)
• Get it wrong sometimes
• Fix it systematically
• Document wins AND roadblocks
• Help others avoid the same mistakes
Progress isn't linear. Humility matters.
• Build in public (even when messy)
• Get it wrong sometimes
• Fix it systematically
• Document wins AND roadblocks
• Help others avoid the same mistakes
Progress isn't linear. Humility matters.
October 28, 2025 at 11:09 AM
Because this is how we actually work with bleeding-edge tech:
• Build in public (even when messy)
• Get it wrong sometimes
• Fix it systematically
• Document wins AND roadblocks
• Help others avoid the same mistakes
Progress isn't linear. Humility matters.
• Build in public (even when messy)
• Get it wrong sometimes
• Fix it systematically
• Document wins AND roadblocks
• Help others avoid the same mistakes
Progress isn't linear. Humility matters.
🧵 4/4 This matters for production AI systems where reliability is non-negotiable. Read the full announcement: bit.ly/42SiOHO @OpenRouterAI @AnthropicAI @OpenAI #AIInfrastructure #LLMOps
October 23, 2025 at 11:01 AM
🧵 4/4 This matters for production AI systems where reliability is non-negotiable. Read the full announcement: bit.ly/42SiOHO @OpenRouterAI @AnthropicAI @OpenAI #AIInfrastructure #LLMOps
🧵 3/4 OpenRouter's unique scale provides visibility into these variances. exacto endpoints route to a vetted sub-group of providers with demonstrably better tool-use success rates, addressing a problem many assumed didn't exist.
October 23, 2025 at 11:01 AM
🧵 3/4 OpenRouter's unique scale provides visibility into these variances. exacto endpoints route to a vetted sub-group of providers with demonstrably better tool-use success rates, addressing a problem many assumed didn't exist.
🧵 2/4 The theory says same model weights with same quantization should yield identical outputs. Reality is different. Production-grade inference implementation is complex, and measurable performance differences emerge across providers running the same models.
October 23, 2025 at 11:01 AM
🧵 2/4 The theory says same model weights with same quantization should yield identical outputs. Reality is different. Production-grade inference implementation is complex, and measurable performance differences emerge across providers running the same models.
Wrote a detailed guide covering: • Building the script (bash 4 + Python) • Machine detection & profiles • JSON merge logic • File-by-file review mode • Real-world workflows
Full code + docs on GitHub (MIT license)
📖 Read: bit.ly/3WbehfX ⭐ Star: bit.ly/47fasv3
Full code + docs on GitHub (MIT license)
📖 Read: bit.ly/3WbehfX ⭐ Star: bit.ly/47fasv3
October 20, 2025 at 4:10 PM
Wrote a detailed guide covering: • Building the script (bash 4 + Python) • Machine detection & profiles • JSON merge logic • File-by-file review mode • Real-world workflows
Full code + docs on GitHub (MIT license)
📖 Read: bit.ly/3WbehfX ⭐ Star: bit.ly/47fasv3
Full code + docs on GitHub (MIT license)
📖 Read: bit.ly/3WbehfX ⭐ Star: bit.ly/47fasv3
Usage is dead simple:
# First time on each machine
cc-setup
# Check differences
cc-analyze-pi
# Smart sync (recommended)
cc-smart-pi
# Or interactive merge
cc-merge-pi
4 conflict resolution strategies built in.
# First time on each machine
cc-setup
# Check differences
cc-analyze-pi
# Smart sync (recommended)
cc-smart-pi
# Or interactive merge
cc-merge-pi
4 conflict resolution strategies built in.
October 20, 2025 at 4:10 PM
Usage is dead simple:
# First time on each machine
cc-setup
# Check differences
cc-analyze-pi
# Smart sync (recommended)
cc-smart-pi
# Or interactive merge
cc-merge-pi
4 conflict resolution strategies built in.
# First time on each machine
cc-setup
# Check differences
cc-analyze-pi
# Smart sync (recommended)
cc-smart-pi
# Or interactive merge
cc-merge-pi
4 conflict resolution strategies built in.
The solution: Smart merge that knows what to sync vs preserve.
**Sync everywhere:**
• Custom agents (18+)
• Slash commands
• Hooks & skills
**Keep local:**
• ANTHROPIC_MODEL
• API keys
• Database connections
Best of both worlds.
**Sync everywhere:**
• Custom agents (18+)
• Slash commands
• Hooks & skills
**Keep local:**
• ANTHROPIC_MODEL
• API keys
• Database connections
Best of both worlds.
October 20, 2025 at 4:10 PM
The solution: Smart merge that knows what to sync vs preserve.
**Sync everywhere:**
• Custom agents (18+)
• Slash commands
• Hooks & skills
**Keep local:**
• ANTHROPIC_MODEL
• API keys
• Database connections
Best of both worlds.
**Sync everywhere:**
• Custom agents (18+)
• Slash commands
• Hooks & skills
**Keep local:**
• ANTHROPIC_MODEL
• API keys
• Database connections
Best of both worlds.
The problem: You add a custom agent on your Mac, then switch to DGX and... it's gone.
Copy ~/.claude over? Now you get:
"API Error: 404 Model not found"
Why? Mac uses Bedrock via Azure, DGX needs direct API.
Same agents, different endpoints. 🔧
Copy ~/.claude over? Now you get:
"API Error: 404 Model not found"
Why? Mac uses Bedrock via Azure, DGX needs direct API.
Same agents, different endpoints. 🔧
October 20, 2025 at 4:10 PM
The problem: You add a custom agent on your Mac, then switch to DGX and... it's gone.
Copy ~/.claude over? Now you get:
"API Error: 404 Model not found"
Why? Mac uses Bedrock via Azure, DGX needs direct API.
Same agents, different endpoints. 🔧
Copy ~/.claude over? Now you get:
"API Error: 404 Model not found"
Why? Mac uses Bedrock via Azure, DGX needs direct API.
Same agents, different endpoints. 🔧
🧵 4/4 Check out Tinker for simplified LLM fine-tuning
bit.ly/4o66HiE
@TheTuringPost @thinkymachines @huggingface
#AITools #MachineLearning #LLMs
bit.ly/4o66HiE
@TheTuringPost @thinkymachines @huggingface
#AITools #MachineLearning #LLMs
October 3, 2025 at 1:22 PM
🧵 4/4 Check out Tinker for simplified LLM fine-tuning
bit.ly/4o66HiE
@TheTuringPost @thinkymachines @huggingface
#AITools #MachineLearning #LLMs
bit.ly/4o66HiE
@TheTuringPost @thinkymachines @huggingface
#AITools #MachineLearning #LLMs
🧵 3/4 Supports advanced models like Qwen3-235B-A22B with Mixture of Experts (MoE) architecture. Democratizing access to state-of-the-art AI model training.
October 3, 2025 at 1:22 PM
🧵 3/4 Supports advanced models like Qwen3-235B-A22B with Mixture of Experts (MoE) architecture. Democratizing access to state-of-the-art AI model training.
🧵 2/4 Key feature: Seamlessly switch between different models (Llama, Qwen) with just one line change. No need to rewrite complex training loops or worry about GPU optimization.
October 3, 2025 at 1:22 PM
🧵 2/4 Key feature: Seamlessly switch between different models (Llama, Qwen) with just one line change. No need to rewrite complex training loops or worry about GPU optimization.
Agents won’t stop at errands. They’ll negotiate, invest, and transact with each other.
That’s an economy. The only question: open system or toll road?
Full post here 👉 bit.ly/48p08Tq
#AI #AgentEconomy #Google #FutureOfWork
That’s an economy. The only question: open system or toll road?
Full post here 👉 bit.ly/48p08Tq
#AI #AgentEconomy #Google #FutureOfWork
September 20, 2025 at 11:00 AM
Agents won’t stop at errands. They’ll negotiate, invest, and transact with each other.
That’s an economy. The only question: open system or toll road?
Full post here 👉 bit.ly/48p08Tq
#AI #AgentEconomy #Google #FutureOfWork
That’s an economy. The only question: open system or toll road?
Full post here 👉 bit.ly/48p08Tq
#AI #AgentEconomy #Google #FutureOfWork
Google’s new protocols — AP2 (payments) and A2A (agent-to-agent) — are like HTTP for commerce.
If they become the default rails, it’s not just search that flows through Google. It’s spending.
If they become the default rails, it’s not just search that flows through Google. It’s spending.
September 20, 2025 at 11:00 AM
Google’s new protocols — AP2 (payments) and A2A (agent-to-agent) — are like HTTP for commerce.
If they become the default rails, it’s not just search that flows through Google. It’s spending.
If they become the default rails, it’s not just search that flows through Google. It’s spending.
The biggest insight: People want AI to help them think better, not replace their thinking. 49% "asking" for decision support vs 40% requesting tasks. We're not automating work - we're augmenting judgment. Full analysis: bit.ly/4nBA9Ne #AI #Future #Business
September 16, 2025 at 11:15 PM
The biggest insight: People want AI to help them think better, not replace their thinking. 49% "asking" for decision support vs 40% requesting tasks. We're not automating work - we're augmenting judgment. Full analysis: bit.ly/4nBA9Ne #AI #Future #Business