How Beginners Build a Self-Improving AI Trading Agent
For Beginner algorithmic traders with no existing strategy · Based on Lewis Jackson Self-Improving Trading Agent Framework
// TL;DR
If you're new to algorithmic trading and don't have an existing strategy, this framework scaffolds a basic baseline strategy for you, then improves it automatically through weekly Hermes learning cycles. You provide the target asset, starting capital, and realistic success/failure definitions. The oneshot prompt handles everything — environment setup, strategy creation, cloud deployment, and Hermes installation. Each week, Hermes changes one variable (e.g., tighter stop-loss, different entry signal), measures the result, and promotes improvements. You learn by watching systematic iteration in action.
Can I build a self-improving trading agent with no coding or trading experience?
Yes. The Lewis Jackson framework is designed so the oneshot prompt handles all technical setup through Claude Code. You paste one prompt, answer guided questions, and the agent builds your entire system. No existing strategy is required — during Phase 2, select path B and the agent scaffolds a baseline strategy appropriate for your chosen asset.
What you do need: a clear target asset (e.g., Bitcoin, Ethereum), realistic success and failure definitions given your capital, and a Railway.app account for 24/7 hosting. The framework will flag impossible goals — if you have $100 in starting capital, targeting $10,000/month returns will be rejected.
How does the framework create a strategy for me if I don't have one?
During Phase 2, when you select the "no existing strategy" path, the onboarding agent builds a basic but functional baseline strategy tailored to your chosen asset. This includes entry and exit rules, position limits, slippage tolerance, gas reserves (for crypto), and scoring weights aligned with your success definition.
This baseline isn't meant to be perfect — it's meant to be the starting point for the self-improvement loop. Hermes will analyze the baseline's performance each week and propose single-variable changes to move it closer to your goal. The first strategy is the raw material; Hermes does the refining.
What's the minimum I need to define before starting?
Four things, in writing, before you touch any tooling:
1. Data accuracy: How will you ensure the data feeding your agent is clean? (Use a single reliable API source, don't mix providers.)
2. Reliability: Your agent runs on Railway 24/7 — confirm you'll create and maintain a Railway account.
3. Success definition: A specific, measurable goal. Example: 8% monthly return with Sharpe ≥ 0.8 and max drawdown ≤ 10%.
4. Failure definition: A specific threshold. Example: any 30-day period with return below 0% or drawdown exceeding 15%.
Be conservative as a beginner. You can raise targets later once the improvement loop has compounded a few winning cycles.
What does the first month look like for a beginner?
Week 1: Paste the oneshot prompt, complete onboarding, deploy to Railway. Hermes runs its first read-only cycle — observing and scoring without making changes.
Week 2: Review the read-only output. Verify Hermes correctly understood your strategy and goals. If everything looks right, enable live mode by editing the YAML file.
Week 3: Hermes proposes its first variable change (e.g., tightening the stop-loss by 5%). The change is applied automatically. You observe results through check-in commands.
Week 4: Hermes evaluates Week 3's change. If it moved toward your success definition, it becomes the new baseline. If not, Hermes proposes a different single-variable change.
You now have a system that's been refined once with clean attribution. Over months, these incremental improvements compound.
What's the smartest way for a beginner to set success and failure definitions?
Start conservative. A common beginner mistake is setting targets based on what they hope to earn rather than what's realistic. If your starting capital is $500, a 5-8% monthly return target with a max drawdown of 10-15% is realistic for most crypto assets. Failure might be defined as any rolling 30-day period with negative returns.
Remember: the framework improves toward your target over time. Setting a modest initial target lets Hermes achieve it faster, giving you a validated baseline you can then raise. An impossible target wastes every improvement cycle.
Start now by retrieving the latest oneshot prompt from the 01 Systems community (Classroom > YouTube Video Prompts) and opening Claude Code in your terminal.
// FREQUENTLY ASKED QUESTIONS
How much starting capital do I need for this framework?
There is no hard minimum, but your success definition must be realistic for your capital. The onboarding agent will flag targets that are mathematically impossible (e.g., targeting $1M/month on $10). With very small capital, focus on percentage-based targets rather than dollar amounts. The framework's value is in systematic improvement, which works regardless of account size — the learning compounds whether you're trading $500 or $50,000.
What if the scaffolded baseline strategy loses money in the first week?
That's expected and by design. The baseline is a starting point, not a finished product. The self-improvement loop exists specifically to refine the strategy over time. Hermes analyzes what went wrong, forms a hypothesis, and changes one variable. Your failure definition prevents catastrophic losses by defining a hard stop. The first few cycles are the rawest learning phase — each subsequent cycle builds on a better baseline.
Do I need to understand the Sharpe ratio to use this framework?
Not deeply. The Sharpe ratio measures risk-adjusted returns — a higher number means better returns relative to the risk taken. The onboarding agent will suggest appropriate Sharpe floors based on your asset and capital. You can accept the suggested defaults. Over time, watching Hermes optimize toward a Sharpe target teaches you intuitively what the metric means in practice.