AI Memory: why context windows are not memory
Notes on building durable AI memory — what is worth storing, how retrieval decays, and why bigger context windows do not make an assistant remember.
Research notes
Working notes from Karnveer Singh on the questions behind the systems he builds — AI memory, autonomous agents, pattern recognition, startup research, human-AI collaboration, market intelligence and future computing.
Autonomous agents: execution is the hard part
Why the difficulty in autonomous AI agents is not reasoning quality but execution, recovery and bounded authority.
Pattern recognition as a working method
How noticing repeated structures across markets, codebases and conversations becomes a repeatable engineering method.
Startup research: signal before spend
A validation method for founders — find demand evidence before writing code, and kill ideas on evidence rather than on mood.
Human-AI collaboration: one voice across many models
Why a portable reasoning identity beats model loyalty, and how a boot prompt keeps behaviour consistent across AI systems.
Market intelligence with AI: mapping demand, not opinions
Using AI and automation to map real demand signals — pricing, competitors, distribution and unmet needs — instead of surveys.
Future computing: software that creates new ways of working
Why the next generation of software should not only solve problems, but change what work looks like.
Related: projects · about Karnveer Singh