Devloop: Closing the loop
Spec-driven code and review loop that runs until all acceptance criteria are met, and summons you for the last mile where human judgment is actually needed.
Satya Borgohain's Blog
This is my blog's RSS feed
Latest posts
Happy Paths Are Edge Cases Now
What applying the known-unknowns quadrant to evals reveals.
Kensa: An Open Source Agent Eval Harness
Kensa lets your coding agents eval any agent.
Why Build the Agent Verification Layer
Thoughts on why agents will always require verifiers at scale.
Physician Disagreement in Healthcare Evals
When you ask two doctors to grade the same AI response they disagree almost a quarter of the time. We wanted to know why.
The Half-life of Benchmarks
Every AI benchmark expires but measuring reliability across longer time horizons reveals what sets the absolute limits.
Blurt: Talk to your Agents
Blurt is a free, on-device, voice-to-text CLI tool for coding agents.
Human Review is the Bottleneck
Agent-generated code creates a review bottleneck that planning can solve.
10x Coding Agents
Lessons from pushing coding agents to their limits over the holidays.
Directive for Claude
A preamble for Claude that prioritizes clarity over pleasantries.
Multimodal RAG
How to build a scalable, multimodal RAG using SQL.
The Paradox of Infinite Context
Why infinite context makes retrieval more important, not less.