NewsletterThe ThreadUsersFeeds

Siddhant Khare

Siddhant Khare builds developer tools and products for AI agents and publishes research on inference performance, agent security, and verification.

Sort by:
What agent pull requests carry for their reviewer
November 20th, 2026•siddhantkhare.com
A skill that never loads is not a broken skill
November 16th, 2026•siddhantkhare.com
Can a flaky test make a coding agent undo a good fix?
November 8th, 2026•siddhantkhare.com
Your agent's clock is not the model's clock
November 6th, 2026•siddhantkhare.com
The cluster scaled to zero before its first request
November 1st, 2026•siddhantkhare.com
I asked Codex to pick ten photos from my trip
November 1st, 2026•siddhantkhare.com
Leave your coding agent for 5 minutes and most of its cache is gone
October 31st, 2026•siddhantkhare.com
Your coding agent loses its cache while it waits for you
October 18th, 2026•siddhantkhare.com
A clever memory pattern in Copilot’s harness
October 15th, 2026•siddhantkhare.com
Repeatable output is not verified evidence
October 4th, 2026•siddhantkhare.com
Inspect the context before a security agent runs
October 4th, 2026•siddhantkhare.com
The coordinates matched. The tensors did not.
October 1st, 2026•siddhantkhare.com
Stop making the model write hashes
September 30th, 2026•siddhantkhare.com
Before GPU inference, prove the inputs
September 28th, 2026•siddhantkhare.com
Reliability is not safety: Compiled context raised observed correctness and unsafe acceptance
September 24th, 2026•siddhantkhare.com
Context should be a build artifact, not a prompt assembled at runtime.
September 20th, 2026•siddhantkhare.com
A cache hit is not proof that you skipped the work
September 13th, 2026•siddhantkhare.com
The compiled run had a lower request-latency sum. I still could not claim break-even.
September 5th, 2026•siddhantkhare.com
A benchmark result without lineage is just a screenshot
September 4th, 2026•siddhantkhare.com
The fastest passing system was not the cheapest one
September 3rd, 2026•siddhantkhare.com