What to Check Before You Fund an AI Trading Agent
A pre-flight checklist for anyone about to put real money behind an automated strategy. Seven checks, all of which you can do before funding anything, and most of which take a minute.
Funding an agent is the point where a configuration becomes a financial position. These are the checks worth making first. Most take a minute, and each one catches a different way automated strategies lose money for reasons unrelated to the market.
1. Is take-profit meaningfully larger than stop-loss?
This is the fastest and most important check.
At a realistic win rate for short-horizon strategies, roughly 45 to 50%, a configuration targeting +6% while risking −15% loses money by construction. You would need a win rate above 70% to break even, and momentum entries do not deliver that.
As a rule of thumb, take-profit should be at least twice the stop-loss. If your configuration fails that test, fix it before funding. A good platform will refuse the configuration or substitute safer defaults and tell you it did.
2. Will each position be large enough to survive its own costs?
Your allocation is divided by max_concurrent_positions, not spent as a lump. Four concurrent positions on a $100 allocation means $25 a trade.
Every on-chain position costs roughly the same to open and close regardless of size. On a $200 position those fixed costs are noise. On a $12 position they are a meaningful share of the outcome before the strategy has done anything.
Work out the arithmetic before you fund: how much capital you actually need walks through it. If you are below the threshold, reduce the position count rather than adding more positions. Concentration is the correct response to a small account.
3. Have you kept gas aside?
An agent needs native currency for transactions. An agent that spends its last SOL on a token purchase cannot afford the transaction that sells it. It holds a position it is structurally unable to exit, which means no stop-loss, no take-profit, and no way out.
Budget the gas reserve separately from the money you intend to trade, and check it is still there after a few trades.
4. Can the strategy actually be tested in the mode you are using?
Not every strategy simulates faithfully. Arbitrage in particular depends almost entirely on execution, and a simulated fill tells you very little about what a real order book would have done.
If a strategy is only meaningful live, a platform should say so rather than show you paper results that will not transfer.
5. Does the platform record why positions close?
Ask. If the answer is that it records closes but not reasons, you will have no way to tell a clean stop from a failed exit when something goes wrong. We ran for months without this and it concealed a real defect the entire time, which we wrote up in what 958 live trades taught us.
6. What is the liquidity of what you are about to trade?
A stop-loss is only as good as the book behind it. On thin markets a stop set at 15% routinely fills far below that, because the level you set never existed as a tradeable price.
This is not a reason to avoid illiquid markets. It is a reason to size positions against available depth rather than against your account balance, and to expect realised losses worse than your configured stop.
7. Have you watched it in paper mode across more than one market condition?
A strategy that works while capital is rotating into small caps behaves completely differently in a flat market. Run it through at least two distinct regimes before funding, and watch how often it declines to trade. A well-built agent spends most of its time doing nothing, and that is the strategy working, not failing.
The short version
| Check | Fails if |
|---|---|
| Take-profit vs stop-loss | TP is less than 2× SL |
| Position size vs fixed costs | Costs exceed ~1% of position |
| Gas reserve | Not budgeted separately |
| Testable in your mode | Strategy only meaningful live |
| Exit reasons recorded | Platform cannot answer |
| Liquidity vs position size | Sizing ignores book depth |
| Tested across regimes | Only seen one market condition |
None of these are about picking the right strategy. They are about whether the strategy you configured is the one that will actually run. Below a certain standard of setup it is not, and the costs quietly rewrite it.
Read the risk disclosure before funding anything. Automated trading can lose money quickly, and the checks above reduce avoidable losses rather than market ones.