How to build a Company Operating System with Hermes and OpenClaw

Most PMs are drowning in AI tool guides and still not feeling more productive. Mikhail Shcheglov is not one of them. As CPO at OLX Classifieds, he built an entire company operating system on top of OpenClaw and Hermes, with a knowledge graph that already covers 54% of his company's product context and an agent his stakeholders talk to before they ever reach a PM. Full Writeup: https://www.news.aakashg.com/p/company-os-hermes-openclaw Transcript: https://tinyurl.com/3unkrzzv Open source GitHub repo: https://tinyurl.com/mrh94mdx Timestamps 1:36 - Why most PMs are not getting AI productivity boost 2:43 - Knowledge leakage and why context is the new company asset 4:01 - The knowledge graph walkthrough and product context coverage 8:02 - How to build the graph and measure coverage 10:03 - Ads 12:14 - What the agentic COS owns 16:50 - Live agent demo, status reports and the stakeholder gatekeeper 22:19 - Inside the architecture, OpenClaw plus Hermes 23:46 - 3 layers of memory and why summarizing hurts recall 27:06 - Ads 30:40 - Hybrid retrieval, imperatives, CLAUDE.md and SOUL.md 37:41 - Auto generated skills and the plus 31% recall test 41:41 - Board skill, design system, backlog and recruiting 57:19 - The future of PM jobs and hiring AI native PMs 🏆 Thanks to our sponsors 1. Bolt.new (https://tinyurl.com/y8srfztt) - Ship AI-powered products 10x faster 2. Product Faculty (https://www.productfaculty.com/?code=AAKASH150) - Get $150 off their #1 AI PM Certification: code AAKASH150 3. Customer.io (https://tinyurl.com/mry7x2r7) - Send smarter messages using your product data 4. Land PM Job (https://www.landpmjob.com/) - 12-week live course to master the PM job search 5. Viktor (http://ref.viktor.com/aakashgupta3) - Use $100 in starting credits to get 5x more done with this AI employee A full year of paid plans across Airtable, Speechify, Descript, Magic Patterns, Linear, Dovetail, Arize at https://bundle.aakashg.com/. Key Takeaways 1. Context coverage is a CPO level KPI - His knowledge graph sits at 54% coverage of industry, business model and customer knowledge. At that level the agent operates like a junior to mid product manager and can make backlog decisions. At 70 to 90% he expects it to support strategy level work. 2. The real problem AI solves is knowledge leakage - When a domain expert leaves, the context leaves with them. A single store of business, customer, product and technical knowledge raises the value of the whole organization and lets you delegate higher level tasks. 3. Do not summarize your transcripts - Summarization cost them roughly 20 to 25% worse recall. You lose granular detail and you force every conversation into a template it does not fit. Store every meeting and every agent conversation raw. 4. Memory needs 3 layers - A knowledge graph for structure, a vector database for fuzzy retrieval, and raw daily transcripts in MD files. Keyword matching fails on roughly 75% of real queries, which is why the vector layer carries the load. 5. Auto generated skills lifted recall by 31% - Hermes watches which tasks you repeat and decides on its own when a skill is worth creating. Tested across five core topics with ten questions each, skills beat no skills by plus 31% accuracy. 6. Imperatives matter more than prompts - Their rules file runs roughly 700 lines. It covers no fabrications, think before you act, facts over guesswork, and a ban on what he calls fake helpful, where the agent explains something obvious instead of solving the problem. 7. CLAUDE.md stays short, SOUL.md goes long - CLAUDE.md is kept under 100 lines, matching the advice from the creator of Claude Code. SOUL.md runs around 800 lines and is second in priority for OpenClaw. Every imperative gets tested against real queries. 8. Make the agent the gatekeeper for feature requests - Stakeholders are trained to pitch the agent first. It asks clarifying questions, checks the request against existing priorities, declines politely if it does not clear the bar, and routes it to the right PM if it does. 9. Roughly 50% of PM time is process, not thinking - Weekly reports, stakeholder updates, demos. Delegate that layer and one PM can cover the work of two, focused entirely on discovery. He now runs one PM across three or four customer facing domains. 👨‍💻 Where to find Mikhail Shcheglov: LinkedIn: https://www.linkedin.com/in/scheglovm1/ Substack: https://corpwaters.substack.com/ 👨‍💻 Where to find Aakash: Twitter: https://x.com/aakashgupta LinkedIn: https://www.linkedin.com/in/aagupta/ Newsletter: https://www.news.aakashg.com/ #AIProductManagement #AgenticAI 🧠 About Product Growth: The world's largest podcast focused solely on product + growth, with over 200K+ listeners. 🔔 Subscribe and turn on notifications.

Mikhail ShcheglovguestAakash Guptahost

Aug 28, 20261h 6mWatch on YouTube ↗

EPISODE INFO

Released: August 28, 2026
Duration: 1h 6m
Channel: Aakash Gupta
Watch on YouTube: ▶ Open ↗

EPISODE DESCRIPTION

Most PMs are drowning in AI tool guides and still not feeling more productive. Mikhail Shcheglov is not one of them. As CPO at OLX Classifieds, he built an entire company operating system on top of OpenClaw and Hermes, with a knowledge graph that already covers 54% of his company's product context and an agent his stakeholders talk to before they ever reach a PM. Full Writeup: https://www.news.aakashg.com/p/company-os-hermes-openclaw Transcript: https://tinyurl.com/3unkrzzv Open source GitHub repo: https://tinyurl.com/mrh94mdx Timestamps 1:36 - Why most PMs are not getting AI productivity boost 2:43 - Knowledge leakage and why context is the new company asset 4:01 - The knowledge graph walkthrough and product context coverage 8:02 - How to build the graph and measure coverage 10:03 - Ads 12:14 - What the agentic COS owns 16:50 - Live agent demo, status reports and the stakeholder gatekeeper 22:19 - Inside the architecture, OpenClaw plus Hermes 23:46 - 3 layers of memory and why summarizing hurts recall 27:06 - Ads 30:40 - Hybrid retrieval, imperatives, CLAUDE.md and SOUL.md 37:41 - Auto generated skills and the plus 31% recall test 41:41 - Board skill, design system, backlog and recruiting 57:19 - The future of PM jobs and hiring AI native PMs 🏆 Thanks to our sponsors
Bolt.new (https://tinyurl.com/y8srfztt) - Ship AI-powered products 10x faster
Product Faculty (https://www.productfaculty.com/?code=AAKASH150) - Get $150 off their #1 AI PM Certification: code AAKASH150
Customer.io (https://tinyurl.com/mry7x2r7) - Send smarter messages using your product data
Land PM Job (https://www.landpmjob.com/) - 12-week live course to master the PM job search
Viktor (http://ref.viktor.com/aakashgupta3) - Use $100 in starting credits to get 5x more done with this AI employee
A full year of paid plans across Airtable, Speechify, Descript, Magic Patterns, Linear, Dovetail, Arize at https://bundle.aakashg.com/. Key Takeaways
1. Context coverage is a CPO level KPI - His knowledge graph sits at 54% coverage of industry, business model and customer knowledge. At that level the agent operates like a junior to mid product manager and can make backlog decisions. At 70 to 90% he expects it to support strategy level work.
1. The real problem AI solves is knowledge leakage - When a domain expert leaves, the context leaves with them. A single store of business, customer, product and technical knowledge raises the value of the whole organization and lets you delegate higher level tasks.
1. Do not summarize your transcripts - Summarization cost them roughly 20 to 25% worse recall. You lose granular detail and you force every conversation into a template it does not fit. Store every meeting and every agent conversation raw.
1. Memory needs 3 layers - A knowledge graph for structure, a vector database for fuzzy retrieval, and raw daily transcripts in MD files. Keyword matching fails on roughly 75% of real queries, which is why the vector layer carries the load.
1. Auto generated skills lifted recall by 31% - Hermes watches which tasks you repeat and decides on its own when a skill is worth creating. Tested across five core topics with ten questions each, skills beat no skills by plus 31% accuracy.
1. Imperatives matter more than prompts - Their rules file runs roughly 700 lines. It covers no fabrications, think before you act, facts over guesswork, and a ban on what he calls fake helpful, where the agent explains something obvious instead of solving the problem.
1. CLAUDE.md stays short, SOUL.md goes long - CLAUDE.md is kept under 100 lines, matching the advice from the creator of Claude Code. SOUL.md runs around 800 lines and is second in priority for OpenClaw. Every imperative gets tested against real queries.
1. Make the agent the gatekeeper for feature requests - Stakeholders are trained to pitch the agent first. It asks clarifying questions, checks the request against existing priorities, declines politely if it does not clear the bar, and routes it to the right PM if it does.
1. Roughly 50% of PM time is process, not thinking - Weekly reports, stakeholder updates, demos. Delegate that layer and one PM can cover the work of two, focused entirely on discovery. He now runs one PM across three or four customer facing domains. 👨‍💻 Where to find Mikhail Shcheglov: LinkedIn: https://www.linkedin.com/in/scheglovm1/ Substack: https://corpwaters.substack.com/ 👨‍💻 Where to find Aakash: Twitter: https://x.com/aakashgupta LinkedIn: https://www.linkedin.com/in/aagupta/ Newsletter: https://www.news.aakashg.com/ #AIProductManagement #AgenticAI 🧠 About Product Growth: The world's largest podcast focused solely on product + growth, with over 200K+ listeners. 🔔 Subscribe and turn on notifications.

SPEAKERS

Mikhail Shcheglov
guest
CPO at OLX Classifieds focused on building AI-native company operating systems (Hermes/OpenClaw) and agentic workflow scaffolding.
Aakash Gupta
host
Host of 'Product Growth with Aakash Gupta' who interviews product and AI leaders and shares AI tooling/resources.

EPISODE SUMMARY

In this episode of Aakash Gupta, featuring Mikhail Shcheglov and Aakash Gupta, How to build a Company Operating System with Hermes and OpenClaw explores building an agentic company operating system with knowledge graphs and skills OLX’s CPO built an “agentic Company Operating System” integrated with Slack, Google Workspace, Confluence, Jira, and meeting transcripts so stakeholders can query projects, validate feature ideas, and even prototype work through an AI gatekeeper.

RELATED EPISODES