Adestto AI
Bots & Tech

Our bots self-optimize every week with AI

The weekly loop, no black box: how Adestto's bots learn from their mistakes and adjust their parameters via AI, with out-of-sample validation.

Mahugnon7 min read

Most automated-trading platforms suffer from the same flaw: their bots are configured once, then left untouched for months. Markets, meanwhile, never stop evolving. At Adestto AI, every bot goes through a weekly optimization loop assisted by AI (Claude, from Anthropic). Here's how it works — without a black box and with no promise of gain.

The problem with frozen bots

What worked in a gold uptrend can break within weeks if the market shifts to a range. Stop-loss, take-profit and session-filter parameters can't stay relevant indefinitely. That's the heart of the topic "adaptive bots vs static bots": a system that doesn't evolve degrades when context changes.

The weekly loop, in four steps

  1. Collect: the week's trades are gathered (entry, exit, duration, result, market regime at the time of the trade).
  2. Analyze: we identify patterns — when do losses concentrate? In which market regime? With what spread conditions at entry?
  3. Consult the AI: a structured report is sent to the AI, with a clear instruction: propose adjustments and justify them, within predefined bounds.
  4. Out-of-sample validation: an adjustment is only applied if it holds on data not used to optimize it. Otherwise the bot keeps its settings.

This last step is the most important. It exists precisely to avoid overfitting: adjustments that look good on the recent past but prove counterproductive live.

What the AI looks at (and what it doesn't do)

The AI doesn't get a single number: it gets an overview — behaviour by session and by market regime, loss streaks, problematic time windows, spread conditions. From there, it can spot correlations a human would take a long time to detect.

What it does not do: predict the market, or guarantee a result. Adjustments affect families of parameters (risk tolerance, session filters, signal-quality thresholds), always within framed ranges, and are applied gradually to avoid abrupt breaks. We don't disclose the exact settings of our Expert Advisors — but the logic of the loop is public.

Guardrails, always on

Independently of optimization, drawdown circuit breakers (daily/weekly loss limits) stay active at all times. Optimization seeks consistency with the present market; the guardrails bound risk. Both are necessary.

Why this transparency

Many "AI" products work like an oracle: a signal, with no explanation. We do the opposite — the method is explainable and documented. It's harder to present, but on a subject as sensitive as your money, trust is built through transparency. To dig deeper, read how our bots learn from their mistakes, the design framework and our reinforcement-learning R&D.


Educational content. Adestto AI is a software and educational-content publisher — not a broker or an investment adviser. No return is guaranteed; past performance does not predict future results. Trading involves a risk of capital loss.

Partager

The conversation continues

A reaction, a question about this article?

Join the Adestto community on Telegram — share your take, your test score or your questions.

Join the channel

About the author

Portrait of Mahugnon H.

Mahugnon H.

CEO · Adestto AI · from Montréal

CEO of Adestto AI. Builds AI analysis tools for the markets and publishes an editorial journal to learn both without kidding himself.

Go further

Learn to trade step by step with the Academy.

3 structured modules, from beginner to advanced. No promise of returns.

Discover the Academy

More into the tools? Discover the Adestto Autopilot