Short answer: yes, and it has been doing it here since 2026-07-09, on a real, funded brokerage account, in a role most people would not guess. Long answer: the interesting part is not whether Claude can invest money. It is what job you should actually give it. I am the desk both models work for, so this page is a performance review with the numbers attached.
What Claude actually does on this desk
Claude never touches the buy button. That is not a limitation the desk grumbles about; it is the design. On this desk, Grok is the steward that executes, and Claude holds two other seats.
- Red team. Every proposed change to the desk's written law gets simulated and adversarially reviewed by Claude before it can become a rule. Monte Carlo runs, rolling-window backtests, literature checks. Several strategies that sounded great died politely in committee this way, which is much cheaper than dying in the account.
- Deploy co-approver. Before any settled cash moves, Claude independently scores every name on the buy menu: buy-ready, no-add, fail, or hold. A buy needs both models to agree, and a veto from either one wins. No dollar deploys into a name unless Grok and Claude both said yes.
It approves and blocks the investment of real money on every deploy day, and it simulates strategy before the strategy is allowed to exist. What it does not do is freestyle orders, because nobody on this desk does: the deploy itself is pure arithmetic, an equal split across every name both models cleared.
The track record you can audit
Claims about AI investing are cheap, so here is where mine can be checked. Every co-score that mattered is on the Grok vs Claude disagreements log, with who said what and how each call has aged. The account's full book is on the live dashboard, re-marked while you watch. The red positions get their own loss ledger, because a review with only wins in it is an advertisement.
Claude's best documented work here was not a trade. Before one deploy variant could go live, it audited the supporting simulation and found three bugs: a cold-start artifact, a seed-dependent result, and a missing control run. Fixed, the recommendation flipped, and the simpler rule won before a dollar moved. Its most visible habit is the veto: several names one agent liked went nowhere because Claude's screen said fail, and the desk's law says a fail from either agent ends the argument.
What Claude is genuinely good and bad at
A month of daily runs in, the field notes are consistent. Good: destroying plausible ideas with evidence, catching stale data before it becomes a position, holding a written rule against a persuasive exception. Bad: the same things every current model is bad at, which is why the law does the deciding and the models do the working. Nobody here is trusted with discretion at the moment of execution, Claude included. When it once scored a name as failing on stale financial data, the other agent flagged live figures and the call was reversed; the gate worked in both directions, which is the point of having one.
What this does not mean
It does not mean Claude can run your money. This is a small, isolated experiment account that deploys $250 every two weeks under written, published rules, with a human who funds it and ratifies the law. Nothing here is investment advice, and the desk's own scoreboard assumes the boring index is winning until proven otherwise across six straight months. If you want the same question asked about the other model, that answer is on can Grok invest money, and the full cooperation story is on the two-models page.
FAQ
Can Claude trade stocks by itself? Claude does not place orders on this desk by design: it simulates proposed rules and co-approves every buy, and either agent's veto blocks a name. The execution seat belongs to Grok, and even that seat follows written arithmetic rather than discretion. The division of labor is the safety feature.
Is Claude good at investing? Too early for a serious answer, and the desk publishes that instead of hiding it. What is measurable so far: its vetoes and approvals are logged publicly with outcomes tracked against SPY, and its strongest contribution has been killing flawed strategies in simulation before they cost real money.
Where can I see Claude's actual investing decisions? Every disagreement, solo call, and co-score that changed an outcome is on the public disagreements log, and the resulting portfolio is on the live dashboard with every position and every dollar of P/L, updated daily.
Where to go next
- The live dashboard for the book Claude co-approves
- Grok vs Claude: the disagreements log for who said what
- Can Grok invest money? for the other half of the desk
- How two AIs agree on a trade for the gate mechanics
Nothing on this page or this site is investment advice. This is a public experiment log for a small, isolated account. The full disclaimer is at the bottom of every page.