Seroter's reading list: agents need documentation, not memory

rseroter · x · 2026-10-06

Richard Seroter's daily reading list #881 focuses on AI agent memory and storage. Key items:

Conversational over autonomous: an article argues the future of software may be conversational, pushing back on the "autonomy is inevitable" narrative.
Memory vs documentation: multiple posts argue "agents don't need memory, they need documentation," plus Google's WikiSkill for giving agents a record of past failures without stuffing the prompt.
Engineering practice: implementing long-term agent memory with AlloyDB and Memorystore for Valkey (short-term session + long-term persistent architecture); a call for default hard budget caps on agent spend.
Evals & platform: building a continuous eval loop, and agents as the new platform engineering customers needing service discovery and governance.

The collection signals that agent memory architecture is a rapidly fluid area right now.

More from coding & agent