Twice per year, I manually rebalance a single Vanguard Traditional IRA into a fixed target allocation across 5 index ETFs. My take is that this kind of personal, human-in-the-loop automation should be doable by the end of 2026 with a reasonable amount of effort.
This market resolves YES if, by December 31, 2026, I successfully automate this workflow using an AI-driven computer agent with human-in-the-loop approvals.
I commit to:
Attempting this in good faith
Spending at least 6 hours and no more than 12 hours total on setup and experimentation.
Using general-purpose AI tools such as Claude Code CLI or similar agent frameworks
My current experience level:
I have vibe-coded and shipped a very simple calculator app to the iOS App Store, and I have used Claude Code CLI to build a Python web scraper with basic macOS automated notifications. I am not a professional programmer or engineer.
Definition of Success:
The market resolves YES if, during a real rebalance attempt, the AI agent:
Logs into Vanguard using credentials stored in my password manager and pauses for me to enter 2FA
Navigates to the correct IRA account and reads current ETF allocations
Compares allocations to a predefined target and proposes a rebalance plan
Waits for my approval, then places each ETF trade, requesting approval before submitting each one
Completes the workflow without manual navigation or data entry from me beyond 2FA and approvals
I am the final judge of success, acting in good faith.
Allowed tools:
The automation may use general-purpose AI tools, browser automation, and Mac OS-level scripting available to a consumer user
The automation may include access to my own API (OpenAI or Anthropic) key and accountsIt may not rely on non-public Vanguard APIs, insider access, or sandbox-only environments
If the market resolves YES, I will provide a short write-up or screen recording demonstrating the workflow, excluding sensitive information.
Resolution:
YES if the above succeeds at least once by Dec 31, 2026
NO otherwise
1,000@AlexanderTheGreater I don't believe I could realistically build this today on my own. My assumption is that without an AI agent or AI CLI actively helping me design, debug, and adapt the automation, I wouldn’t be able to get through everything within a 6–12 hour limit as a non-engineer. The question is whether AI tools will be good enough over the next ~12 months to let a person like me clear those hurdles with modest effort.
This market relies on me judging in good faith when I’m ready to attempt it and whether the stated criteria are met.