#51 - AI Native Ops / Company as a Monorepo / Overbot

Link to Folder with Everything

2026.08.19.officehours.51

Recording of the Video

Watch on Google Drive

Summary (by Gemini)

Discussing AI native operations and monorepo company structures for increasing business velocity via loop based agent workflows.

AI Native Operational Framework

Organizations operate efficiently by treating the company as a monorepo where canonical documents guide agent outputs. This loop based system drastically increases output speed compared to manual methods.

Git Based Agent Architecture

Teams manage workflows by storing processes in version controlled directories that function as centralized knowledge bases. Using Git and Markdown ensures interoperability and allows agents to enforce standardized operating procedures.

Overbot Open Source Release

The decision was made to release the Overbot platform on GitHub in September to enable automated organizational scaffolding. This approach shifts human roles toward fleet management rather than managing individual tasks.

Next steps

• [Michael Mayernick] Follow up: Connect with Clyde to discuss the role of ontologies, knowledge graphs, and hypergraphs in agentic systems.
• [Michael Mayernick] Share competitive analysis: Follow up with interested users who provide their email address by sharing the competitive landscape documentation.
• [Clyde] Provide email address: Submit email address in the chat or on the website to receive updates regarding the public repository release.
• [Michael Mayernick] Share repo URL: Send the repository link to Rahul once it becomes public.
• [Rahul Singh] Distribute repo URL: Send the repository link to all attendees after receiving it from Michael.
• [Michael Mayernick] Publish essay: Release the article regarding the shift toward human owned queues on Monday.
• [Rahul Singh] Distribute recordings: Provide the meeting recordings to all participants.
• [Michael Mayernick] Post website link: Provide the website URL for the Inside Native Inside publication.
• [Rahul Singh] Send Luma links: Share the discussed resources through the Luma blast message.
• [Michael Mayernick] Share email: Provide contact details for continued collaboration and discussion.

Details

• Meeting Introduction and Concept Overview: Rahul Singh opened the office hours by introducing Michael Mayernick and discussing their shared interest in the future of companies and the impact of AI, specifically highlighting Michael Mayernick's concept of "running your company out of a folder".
• Michael Mayernick's Background and Transition to AI-Native Operations: Michael Mayernick shared their background in software engineering, including leading the voice AI project for the McDonald's drive-through system, and explained how their current venture focuses on building an AI-native company where agents perform 90% of the work.
• Workflow Parallelization and Dependency Management: The organization manages work by creating batches that are sent to a dispatcher agent, which then assigns tasks to appropriate models and manages dependencies across 20 or more parallel lanes, keeping work moving to avoid expensive human blocking.
• Applying Monorepo Principles to Company Structure: Michael Mayernick described the core insight of treating an entire company as a "monorepo," where all documents, marketing strategies, and engineering processes are stored in a single folder, allowing agents to reference and ensure the interoperability of various departmental outputs.
• The Importance of Canonical Steering Documents: To maintain alignment without constant manual course correction, Michael Mayernick emphasized the use of "canonical" or "steering" documents—such as mission statements and ICP (Ideal Customer Profile) docs—that agents reference to guide output quality and strategy.
• Transitioning to Loop-Based Workflows: Approximately three months ago, Michael Mayernick moved from a manual, artisanal engineering approach to a loop-based system, noting that while it required overcoming the habit of wanting to review every line of code, this shift increased output speed by 50 to 100 times.
• AI Agents and Organizational Culture: Michael Mayernick and Fran Palokaj discussed how AI agents enforce standard operating procedures more consistently than human employees, as agents can be required to follow defined protocols and verify that outputs align with established company values.
• Implementing Technical Workflows for Non-Technical Users: To collaborate with a non-technical co-founder, Michael Mayernick's company implemented Git-based pull requests that are styled to look like familiar, tracked-change documents in Google Docs, bridging the gap between technical infrastructure and accessible user interfaces.
• Memory on Write versus Memory on Read: Addressing a question from Clyde regarding ontologies and knowledge graphs, Michael Mayernick distinguished between "memory on read" (traversing graphs to find answers) and "memory on write" (using static, canonical documents to steer output), arguing that the latter provides more predictable results for company operations.
• Managing Hallucinations and Context Saturation: Michael Mayernick explained that agent performance degrades in long, context-saturated chat threads, so they utilize fresh context windows for specific tasks and rely on human-reviewed pull requests to update canonical steering documents, effectively mitigating "hallucinations".
• Live Demonstration of Directory Structure: Michael Mayernick displayed the agents.mmd file, which acts as a master navigation guide for the company, and showed how the repository is structured into departments like "People" and "Voices," with processes that are "typed" to ensure required information is always present. The system includes automated generation of investor updates based on recorded activities, "checkpoints" that serve as save points for agent state, and a UI layer that enables team members to track every change in the Git history.
• Open Source Release of Overbot: Michael Mayernick announced that the Overbot product—comprising templates, a command-line interface, and the scaffold for this organizational structure—will be released on GitHub in early September. He also demonstrated Overbot's terminal user interface, which scaffolds company structures by guiding the user through creating a new company, defining the business purpose, and identifying relevant departments, roles, and skills.
• Platform Ownership and Open Systems: Michael Mayernick advised against using closed platforms that require transferring organizational memory to third-party companies, advocating instead for open systems built on Git that ensure the organization maintains ownership of its own data and files.
• Defining Canon for Organizational Alignment: Rahul Singh and Michael Mayernick agreed that defining "canon"—the absolute, law-like set of organization books—automatically shapes company culture, with Rahul Singh noting the utility of using "archetypes" to define the core essence of technical systems. Phemi Kgomongwe asked whether strict canons might throttle model intelligence; Michael Mayernick argued that canons are superior to unstructured prompt memory because they force alignment with defined company identity, values, and voice.
• Review Mechanisms for Drift and Updates: The organization employs two review flows: an automated feedback loop where agents compare finished work against canon to detect drift, and type-checking mechanisms that block unauthorized updates, requiring human intervention for changes to the company mission or vision.
• File Formats and AI Efficiency: Rahul Singh and Michael Mayernick discussed the necessity of structured file formats like Markdown and YAML for AI agents — compared to platforms like Google Docs, Markdown allows for efficient, lightweight processing that prevents wasted round trips during AI interactions. The repository also incorporates a local SQLite instance for full-text and vector search within the local environment.
• AI Safety and Governance: Rahul Singh and Michael Mayernick discussed the potential for Overbot to act as an "AI lawyer" or compliance department, emphasizing that in AI-native operations, understanding why an agent makes a decision is crucial for debugging and safety, particularly as the technology scales to physical robotics.
• Managing Human-Agent Workflows: Michael Mayernick described a shift in management philosophy for AI-native companies, where humans move from managing tickets to managing "fleets" of agents — if a manager is describing tasks in detail, they are essentially competing with an agent, and should instead move up a level of abstraction to keep agents saturated with work.
• Business Velocity and Economics: The participants analyzed the business side of AI — peers achieving significant ARR growth through velocity, "AI token maxing" and the cost-effectiveness of premium subscriptions, rotating between subscriptions to maximize inference, and the question of whether traditional SaaS products can survive being rebuilt within AI-integrated workflows. They also explored autonomous agent commerce, where agents self-provision services, manage budgets, and make purchases via systems like Stripe projects.
• Meeting Closing: Rahul Singh concluded the meeting, expressing excitement for the open-sourcing of Overbot, with links to Overbot.ai and Michael Mayernick's Inside Native publication.