Three rebuilds: what building a trading platform with AI agents actually took
I started building Atlas on July 31. It's now on its third version, and the one running today finally looks like the platform I have been envisioning for a long time. Getting here took two restarts, a reset in the middle of the third attempt, and a few changes of tools. This is what happened, including the parts I'd do differently. Commits per day, Jul 31 to Sep 25. The dashed line is the Sep 23…
Building Atlas, a trading platform with AI agents, began on July 31 and evolved through three major versions. The first version, built using an AI coding agent, struggled with determining the folder structure, requiring a restart after 11 days and approximately 117 commits. The second version, which lasted about a month, placed a paper trade on an OANDA practice account but failed to account for the trade's take profit, leading to its loss.
This version highlighted the challenges in ensuring an agent understands the final product, even when it can write functional code.
The third version, utilizing Codex, progressed more smoothly. However, it suffered from a significant overhaul known as the reset on September 23, where the project was rebuilt from scratch as 'atlas-next.' This reset aimed to address the architectural issues plaguing the previous versions, such as hard-coded market specifics (e.g., EUR/USD and OANDA) and the absence of a system to reconcile differences between the trading platform and the broker's records.
The reset also introduced better planning and allowed the developer, who is a trader, to inject domain-specific knowledge into the system's decision-making process.
Today, the third version operates deterministic backtests on stored data, with the first paper run beginning on September 27 on the OANDA practice account. Despite not being ready for real trading, the experience underscores the importance of combining AI agent capabilities with human expertise to develop a comprehensive trading platform.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
