Writing
DeepSWE - the benchmark that makes SWE-Bench look tired
DeepSWE is not interesting because it crowns a new model. It is interesting because it asks whether our coding benchmarks can still be trusted.
Pi Coding Harness - the agent you are supposed to modify
Pi is not trying to be another polished AI IDE. It is a small coding harness for people who want to own the workflow.
How to make Contributions ?
A practical guide to making real open source contributions - not just fixing typos, but shipping actual value.
Claude Design - any good?
Anthropic's new design tool, tried on two real projects with a $200 Max sub. Verdict in two layers - short answer up top, the long write-up below.
Claude Code Regression - Can we stop writing AI CLI with AI CLI itself?
Can we talk about absolute tragedy of Claude Code?
Gemini 3 Pro. Was the hype worth it?
My thoughts on Gemini 3 Pro after a week or two of real usage
DeepSWE - the benchmark that makes SWE-Bench look tired
DeepSWE is not interesting because it crowns a new model. It is interesting because it asks whether our coding benchmarks can still be trusted.
Read the piecePi Coding Harness - the agent you are supposed to modify
Pi is not trying to be another polished AI IDE. It is a small coding harness for people who want to own the workflow.
Read the pieceHow to make Contributions ?
A practical guide to making real open source contributions - not just fixing typos, but shipping actual value.
Read the pieceClaude Design - any good?
Anthropic's new design tool, tried on two real projects with a $200 Max sub. Verdict in two layers - short answer up top, the long write-up below.
Read the piece
Claude Code Regression - Can we stop writing AI CLI with AI CLI itself?
Can we talk about absolute tragedy of Claude Code?
Read the pieceGemini 3 Pro. Was the hype worth it?
My thoughts on Gemini 3 Pro after a week or two of real usage
Read the piece