Every AI tool on the market is racing to be smarter. Bigger models, longer context windows (the amount of text a model can process at once), better reasoning. But the friction that actually slows people down daily is much simpler: AI tools don't remember what you told them yesterday.
Open a new ChatGPT conversation and you're starting from zero. Your formatting preferences, your brand voice, the way you like code structured, the fact that you always want metric units - gone. You re-type the same setup instructions, paste the same style guides, re-explain the same constraints. Multiply that across every tool in your stack and you're spending real time each week just getting AI back to where it was in your last session.
Some tools are chipping away at this. ChatGPT has a memory feature that stores facts across conversations, and Claude Projects let you pin instructions that persist. Custom GPTs bake in a fixed system prompt. But these are blunt instruments. None of them learn from your actual behavior the way a good assistant would - noticing that you always reject formal language, or that you prefer bullet points over paragraphs, or that you rewrite every email opening they suggest.
The gap matters because it caps how much time AI tools actually save. A tool that's 10x faster at drafting but requires 5 minutes of re-prompting each session has a much smaller real-world advantage than the benchmarks suggest. The teams getting the most value from AI right now are the ones building their own prompt libraries, template systems, and workflow wrappers to compensate for what the tools themselves don't retain.
Until AI products treat user preferences as persistent state rather than per-session context, the "just ask AI" workflow will keep bumping into the same ceiling.