Agents built with LangChain Agent Builder can produce near-correct outputs on the first pass, requiring only minor tweaks and easy updates over time.

Tom Spencer shares that they have deployed and demoed multiple agents, including an email assistant built with Agent Builder, on their own LangSmith s...Show more

When querying HubSpot pipeline projections via a connector, the model misread a field format and returned only 1% of expected deal value, highlighting...Show more

After struggling to write MCP servers correctly for OpenAI, Cameron celebrates that the new developer mode now simplifies custom connector integration...Show more

Tool usage is constrained by credit-based pricing, leading to limited monthly tool calls and development friction.

Tom Spencer recounts struggling to find turnkey cloud providers for deploying Polygon MCP servers to make chat-based market data tools accessible on m...Show more

Tom Spencer notes Amazon announced layoffs of 30,000 employees, representing roughly 2% of its workforce and largely affecting middle management.

Cameron Rohn recalls Primogen’s tweet showing a spike in AWS outages, illustrating potential reliability impacts following staff cuts.

Tom mentions his bet on Google’s AI as paying off, with Gemini achieving 600–650 million DAUs versus OpenAI’s 800 million.

Tom humorously shares his anecdote of traveling to every Texas city except Cameron’s to avoid him.

Despite acquiring a powerful NVIDIA DGX, Cameron humorously admits he hasn't been able to leverage its full capabilities since last week.

Participated in a fireside chat at Enable's AI summit where Cameron shared his transition from managed services to AI-focused infrastructure discussio...Show more

Saw the community rapidly creating fine-tunes on the newly released Minimax M2, illustrating early grassroots innovation around open models.

Tom faced a real-world challenge optimizing a light agentic tool at work, driving his interest in prompt templates and LangChain’s Agent Builder.

While demoing an AI-generated Australian beach promo, the team realized the output resembled generic coastal landscapes rather than the specific targe...Show more

In minutes, Cameron built and ran a demo agent that uses speech input, location insights, and video templates to auto-generate a promo reel for Venice...Show more

Tom ran DeepSeek OCR on his NVIDIA DGX for local inference and is exploring using it as a testbed for future fine-tuning experiments.

The jump from 256 MB to 512 MB of RAM at historically high cost illustrates how revolutionary hardware leaps feel until they become the new normal.

Tom shares that running the OCR dataset of 1,600 PNG images on his local device took about 13 hours, illustrating the compute resource demands for lar...Show more

I unboxed and connected my DGX Spark via RDP, using a guitar wah-wah pedal as a makeshift stand to show its compact size on video.