How we run Holdex on GitHub and AI agents
Every contract, process, and HR record lives in a repository. AI agents execute on top of it.
Table of Contents
Most companies separate their technical work from their operational work. Code goes in GitHub. Contracts go in email. Procedures go in Notion. HR records go in a spreadsheet, or a Google Doc, or wherever they landed.
Over time the fragmentation compounds. Nobody has the full picture. Nobody can automate against it. And when AI tools arrive, there is nothing structured for them to act on.
We chose a different substrate.
GitHub is not just for code
At Holdex, GitHub is the operating system for the company — not just the product.
HR agreements are markdown files in private repositories.
Each contractor has a private hr-member-{name} repo containing their filled-in
agreement, version-controlled and reviewed via pull request before merging.
The standard terms they reference live in a public repo, versioned by git tags.
When those terms change,
a GitHub Actions workflow sends personalised emails to every active contractor
automatically — one per person, HR in CC.
Onboarding follows the same pattern. When we bring on a new developer, a script creates their HR repository, configures labels, opens the standard issue set, and commits the agreement template. The process is repeatable and auditable, not dependent on anyone remembering the checklist.
Client work sits in the same structure. GitHub as the source of truth for client engagements covers the customer-repo setup in detail: Goals at the top, Problems underneath, time tracked against the pull requests that produced the work. No Slack thread that disappears. No Notion page that drifts.
Why markdown and not a dedicated system
There are purpose-built HR platforms, contract tools, and onboarding software. The problem is always the same: they are islands.
Data in a dedicated system is human-readable at best. It is almost never machine-actionable in a predictable way. You can export a CSV. You can call an API that changes with every release. You cannot point an AI agent at it and say: read this, understand it, act on it.
A markdown file in a git repository is different. It has a full revision history. Every change is a commit with an author, a timestamp, and a message. It is reviewed by a human before it merges. And any capable model can read it cold and understand the current state.
That last property is what makes the system compound. Every automation we build is a permanent asset. Every markdown file is history that a future agent — or a new team member, or an audit — can read without needing context handed to them.
The AI layer
The markdown layer is where the real leverage comes from.
When a contractor agreement needs to be generated, an agent pulls the template, fills in the placeholders, opens a pull request for review, and commits the result. When the standard terms release, a workflow assembles the recipient list from the config file, fetches the release notes, and sends personalised emails.
Our partners and senior developers do not perform those steps. They set direction. The agents handle execution.
This is not a prediction about where AI is going. It is the current operating model. The same GitHub Actions and scripts that have run our HR and client workflows for years now run alongside AI agents — reading the same repositories, writing to the same pull requests, operating against the same structured data.
The compounding effect is real. The Goal → Problem → Solution framework we use for product development is the same one we use for every internal process. One tool. One structure. No context switch.
What this means for the founders we work with
The founders we work with are building in DeFi, fintech, and AI infrastructure. They hold their own products to an institutional standard of rigour, and they expect the same from the teams alongside them.
When you work with Holdex, you work with a team that has built the same systematic infrastructure for their own operations that they build for yours. Every decision is version-controlled. Every process is auditable. Every automation is a repeatable asset.
That is not a claim about culture or values. It is something you can inspect.
Work with us
If you are building in DeFi, fintech, or AI and want to understand how we work, reach out.