Evals Are the New Unit Tests: Notes From Ramp
The interesting story is not the demo. It is the second month, when evals-first development either earns its keep or gets quietly rolled back.
Priya Raman3 min read
The archive
All articles
100 stories from the Inference Daily desk.
long-context workflows: The Pattern Behind the Best Agents This Quarter
Beyond the launch posts, Qwen 3 is reshaping how engineering teams approach customer support. We talked to the people actually using it in production.
Mira Castellanos3 min read
When the Board Asks About AI: A Practical Answer From Ramp
The interesting story is not the demo. It is the second month, when small-model orchestration either earns its keep or gets quietly rolled back.
Jonas Halvorsen3 min read
LLMs·
The Quiet Architecture Shift Behind Phi-4
Beyond the launch posts, Gemini 3 Pro is reshaping how analysts approach pricing analysis. We talked to the people actually using it in production.
Elena Brost3 min read
Why structured outputs Beats Bigger Models for Real-World Agents
Figma is the latest in a string of teams treating computer-use agents as the default, not the experiment. Here is what they got right — and what they are still figuring out.
Elena Brost3 min read
Evals Are the New Unit Tests: Notes From Snowflake
The interesting story is not the demo. It is the second month, when RAG-as-a-service either earns its keep or gets quietly rolled back.
Yuki Tanabe3 min read
LLMs·
Why Llama 4 Feels Different, Even When the Numbers Don't
HubSpot is the latest in a string of teams treating computer-use agents as the default, not the experiment. Here is what they got right — and what they are still figuring out.
Daniel Okafor3 min read
Beyond Vibes: Measuring GPT-5.1 in Production
Beyond the launch posts, Grok 4 is reshaping how sales teams approach contract review. We talked to the people actually using it in production.
Mira Castellanos3 min read
How Booking.com Cut a onboarding From Hours to Seconds
Snowflake is the latest in a string of teams treating tool-first agents as the default, not the experiment. Here is what they got right — and what they are still figuring out.
Yuki Tanabe3 min read