# Your second brain will not scale URL: https://www.associativetrails.com/writing/2026/08/27/your-second-brain-will-not-scale/ Published: 2026-08-27 · AI · Sheet 35 Summary: A folder of markdown files will run a one-person business. It will not run a practice of fifty. On Monday, Garry Tan (of Y Combinator) gave systems of record two options: become an AI harness, or be replaced: > Prediction: systems of record will need to become AI harnesses or face replacement by agents — Garry Tan (https://x.com/garrytan/status/2091742825042030681), X, 24 August 2026 He and his company have been banging this particular drum for a number of months: - Garry open-sourced GBrain, his "brain layer your AI agent has been missing (https://github.com/garrytan/gbrain)" in April and has been adding to it ever since. - Y Combinator's Summer 2026 Request for Startups included two similar ideas: the Company Brain (https://www.ycombinator.com/rfs#company-brain) and the AI Operating System for Companies (https://www.ycombinator.com/rfs#ai-operating-system-for-companies). - He spoke about GBrain and the org-as-markdown mapping at the AI Engineer World's Fair (https://www.youtube.com/watch?v=eBUyTS7SzV4) in July. - And as of 27 August, the day I'm writing, the above X post has over 698K views. It's clearly an idea that is generating a lot of interest, so it's worth unpacking. ## What is a "harness"? Large Language Models (LLMs) are generalists. They have read the public internet. They have not read your project records, your staff capabilities, or know which client always pays late. So, while an LLM can produce incredible documents and slide decks based on general public knowledge, you need to give it the "context" of your own business for it to produce the really useful stuff. An AI "harness" is an interface that lets a human interact with an LLM. At the moment, the most high-profile examples of these are chat interfaces. But they are becoming more sophisticated - Claude Cowork and ChatGPT Work still expect you to chat with the LLM, but they can read files from your computer, search the web, and take control of other applications on your computer. For complex tasks, they can strategise - build a plan, send out sub-agents to execute bits of it, and pull together the results into a coherent output. ## Why are harnesses so useful? For a solo/small business, these commercial harnesses are revolutionary. A bunch of text files can contain all the context the AI needs to produce meaningful work that is targeted to you. I keep a Second Brain-style (https://www.buildingasecondbrain.com/) repository of markdown files in Obsidian, and Claude Cowork uses that as context. It means that when I ask Claude to analyse an academic paper, summarise a YouTube video or help me plan a blog post it knows exactly what my business does, what I have been reading recently, and who I am targeting. For solo/small businesses, the main thing stopping them getting the most out of AI is resources - they lack the time or people to set up a Second Brain, GBrain or something similar. But the tools and advice are freely available, when they get round to it. ## What about larger businesses? The constraint does not disappear as a firm grows. It migrates. Imagine ten engineers all producing design reviews. Left alone, each develops their own prompts, standards and shortcuts. Six months later nobody can explain why similar projects receive different recommendations. Consistency is now the main problem. This is where shared standards are needed, something that cannot really be done with an ad-hoc collection of markdown files on each user's personal drive. And as this problem is solved and businesses continue increasing in size, the constraints continue migrating. Unverified AI output, ungoverned data collection, and unaccountable decisions turn into audit exposure, compliance risk, and reputational damage. At this stage, your "Company Brain" needs controls, an audit trail, and integration with risk frameworks. That's not to say that these constraints are being ignored. Some of them are raised in the comments below Garry's post: - @paulbeneteauu cited Salesforce's "Headless 360" exposing the platform via API/MCP/CLI, and made the sharpest operational point in the thread: the hard part is not the API surface, it is moving business rules that used to live in the UI down into the platform core, "or agents will bypass them without knowing." - @mtramos asking who keeps the record honest if the system of record also runs the agent. - @truevined: tools can move the work but responsibility cannot move with them. He wants deterministic records, visible approvals, and a named person on the hook when an agent fails. "The interface can change. Accountability can't." ## So why do it? Even so, there are all sorts of other benefits in standardising your staff's access to AI agents via a harness: - It can keep a record (https://www.associativetrails.com/writing/2026/05/07/agents-cannot-do-osmosis/) of all chat sessions and summarise the decisions made, list action items and write it all to an audit log. This needs user consent, and proper framing: a record of the work, not a record of the worker. - Outputs can be automatically evaluated against desired business outcomes or formatted in a certain way. - The harness can brief you at the end of each session on what was produced and why, so you walk into the meeting holding the reasoning rather than just the conclusion. This helps with comprehension debt (https://www.associativetrails.com/writing/2026/07/23/borrowing-against-the-day-a-client-asks-why/). The briefing belongs to the person who has to stand behind it, and nobody else needs a copy. - When working with the AI to write documents, slide decks or meeting notes, they can be automatically checked with a panel of experts (https://www.associativetrails.com/writing/2026/06/04/introducing-roundtable/). - It can swap the model underneath without touching the context. If a new model launches, a provider changes its terms, or a US export-control order switches a particular model off (https://www.infoq.com/news/2026/06/claude-5-release/), the markdown files and the manifest stay exactly where they are, and you point them at something else. Model churn is not your problem. (https://www.associativetrails.com/writing/2026/07/16/model-churn-is-not-your-problem/) - It can screen for private or client-sensitive information before anything leaves the company firewall, stripping names, financial figures or personal data out of the prompt before it reaches an external provider's servers. Despite the hurdles, it's clear that company-specific harnesses, Claude Cowork-style interfaces with instant access to all of the company's data and preferred outputs, are on the way. As Harrison Chase of LangChain recently put (https://www.youtube.com/watch?v=HI2q3ci3Iuc), a harness has one job - putting the right context into the window at the right moment. General-purpose harnesses are tuned for general-purpose work. By all means, buy the harness that drafts your emails. But build the one that knows which client always pays late. ## Where to start? - If you can, it's well worth setting up a second brain-style repository of markdown files. This should be in a lightweight format that doesn't take too many tokens for the AI to process. I know AI lab employees with unlimited token budgets are espousing HTML as a richer format (https://claude.com/blog/using-claude-code-the-unreasonable-effectiveness-of-html) but the rest of us need to keep to our daily limits. HTML output is great if you need an interactive explainer or a resource you can share with other humans, but it's too bloated for agent context. - If you decide to point a commercial harness at a directory of documents, include a manifest file (CLAUDE.md or AGENTS.md) with three tiers (Canonical, Domain, Archival) which tells the AI which documents are source of truth, which are domain-specific references, and which to ignore unless explicitly asked. - If you have an existing CMS or intranet, a good first step is to update the system so all content is mirrored (and automatically kept updated) in Markdown format and provide a way (via API or MCP (https://modelcontextprotocol.io/docs/2026-07-28/getting-started/intro)) for the AI agents to search and retrieve the markdown information. The business rules, attribution and accountability remain in the intranet interface, but the information is available to the AI agents to do wonderful stuff with. - Throughout all of this, you need to be mindful of privacy and disclosure practices. If you're using a commercial agent harness like Claude Cowork or ChatGPT Work, any context you provide will be sent to their servers.