Practical agentic win: faster backtests

A backtest is a replay machine for trading ideas: you take an old market chart, apply a rule set, and see what would have happened if that rule had been live. NinjaTrader 8’s Strategy Analyzer is built for exactly that kind of historical testing on automated strategies written in NinjaScript, its C-sharp-based scripting framework. (ninjatrader.com) The bottleneck is usually not the math. The slow part is the loop around the math: changing parameters, launching the run, waiting for the result, reading the log, and doing it again hundreds of times. (ninjatrader.com) NinjaTrader’s own docs describe strategy development as an iterative cycle of changing the strategy description, backtesting, and repeating until the system meets requirements. It also saves a backtest log for each test, which means there is already a paper trail an agent can read and act on without a custom application programming interface. (ninjatrader.com, ninjatrader.com) That is where agentic coding tools fit. Anthropic describes Claude Code as a system that can understand a project, make changes across files, run tests, and complete development tasks autonomously from the command line. (anthropic.com) In practice, that means the model does not need a direct plug into the trading platform to be useful. If the files, prompts, logs, and desktop actions are predictable enough, the agent can act like a very fast research assistant sitting at the keyboard and working through a checklist. (anthropic.com, anthropic.com) A trader posting under the name ZenomTrader said Claude Code pushed a NinjaTrader 8 autonomous backtesting pipeline from about one test every three minutes to about five tests per minute. The same post said the setup ran fully in the background and did it without an application programming interface integration. (x.com) That jump is larger than it sounds. One test every three minutes is about 20 tests an hour, while five tests per minute is about 300 tests an hour, which is a 15-fold increase in how many ideas can be checked in the same time window. (x.com) More tests do not guarantee a profitable strategy. More tests do mean a trader can reject weak parameter combinations faster, compare more variants before the market regime changes, and spend less human time on clicking through menus. (ninjatrader.com, ninjatrader.com) NinjaTrader’s docs also note that Strategy Analyzer can slow down when it downloads data from a market data provider during a backtest, and they offer a “Use Local Data Only” setting to avoid that delay. That detail matters because the speedup here is not just smarter prompts; it is also a reminder that agentic workflows get faster when the surrounding system is tuned for local, repeatable runs. (ninjatrader.com) Anthropic has been adding features aimed at longer-running autonomous work, including sandboxing in October 2025 and “auto mode” in March 2026 to reduce approval fatigue around routine actions. Those features were built for coding tasks, but the same pattern shows up here: once the machine can safely edit files, run commands, and inspect outputs, it can also babysit repetitive research loops. (anthropic.com, anthropic.com) The important part is not that one trader found a faster way to click buttons. The important part is that a desktop tool with no direct application programming interface became scriptable enough for an agent to turn a human-speed workflow into a machine-speed one. (x.com, anthropic.com) That is what early agentic adoption often looks like in the wild. Not a new model inventing a strategy, but a model removing the dead time between attempts so the person with the strategy can learn 15 times faster. (x.com, ninjatrader.com)