Raw transcripts beat summaries by 20-25% recall; agent-written skills lift accuracy 31%
aakashgupta · x · 2026-10-28
Aakash Gupta shares key takeaways from a podcast with OLX Classifieds CPO Mikhail Shcheglov, whose product team has run on OpenClaw and Hermes for 5 months.
Key findings
- Storing meeting transcripts raw instead of summarized: summaries made recall 20-25% worse in their test, so everything is stored raw.
- Letting Hermes write its own skills off repeated tasks produced a 31% accuracy lift in controlled testing.
- His knowledge graph covers 54% of the company's product, business and customer context — tracked as a personal CPO KPI. At that level the agent already makes backlog decisions, and stakeholders must pitch feature requests to it before reaching a PM.
- The agent also runs his email, calendar, recruiting funnel and design system.
Heavier questions: what happens to PM headcount when one PM covers four domains, and what does a CPO screen for now when hiring?
More from coding & agent
- After auditing his own C code with AI, dev no longer trusts unaudited C/C++ — burny_tech · 2026-10-28
- Snowflake Doubles Down on Agentic RL: Overfit First, Then Scale to 65.7 on SWE-bench — StasBekman · 2026-10-28
- Hook-level agent security wins: Microsoft, Anthropic and OpenAI all shifted from gateways to runtime hooks — jacob_posel · 2026-10-28
- Parsing Claude's Raw SSE Stream into a Forensic Timeline of Agent Tool Calls and Diffs — Additional-Ladder-24 · 2026-10-28
- Supabase launches Compute in private alpha: run AI agents and long-running jobs next to your Postgres data — dshukertjr · 2026-10-28
- Agency Owner Predicts Agents Go Mainstream in 2027 — Built Into Products, Not Self-Set Up — tjrobertson-seo · 2026-10-28