We rebuilt The Hunter, our AI day trading agent, from a technical-analysis system into a narrative-first trader. I spent 12 years trading stocks at Goldman Sachs, most recently as head of Execution Services in Taiwan, and the new agent is built around how I actually think about markets. It explains its reasoning in real time and treats every rationale as published content. Here's what changed and why.
The original Day Trader scanned the entire US equity market for momentum signals, ran RSI, MACD, and Bollinger Bands, and used up to 3x leverage to chase whatever was moving. On paper it looked sophisticated, but in practice it lost money.
The problem was competition. We built a technical analysis bot and sent it against quantitative firms with decades of proprietary training data and billions of dollars in dedicated hardware. Even with our sophisticated intelligence framework, an AI reading indicators off a standard data feed was never going to win that fight.
So I asked a different question. Instead of "how do we make a better quant," I asked: what do experienced execution traders do that the quants don't?
I spent months last fall battle-testing our AI against a concentrated portfolio of five stocks and the IBKR VWAP algorithm. Every session was recorded and analyzed by AI, and what came out of that process wasn't a better indicator but a pattern in how I was actually making decisions.
The edge was narrative. When TSMC reports strong numbers overnight, that means incremental buyers for Intel before the US market opens. When American Airlines beats earnings, that doesn't help JetBlue because premium airline demand doesn't signal budget carrier demand. When a stock fails at $50 three times and the pullbacks get shallower, the sellers are running out of ammunition.
I knew all of this from 12 years on the trading desk. The question was whether I could teach it to an AI, and it turns out you can when the AI has the right research infrastructure. Terminal X has institutional-grade news, analyst activity, positioning data, earnings events, and technical signals. We just weren't using them correctly.
The agent now runs on a narrative-first strategy with one core question before every trade: will this trend continue or reverse? Everything else exists to answer that.
Every morning before the open, it reviews its universe: what happened overnight in related international markets, what fresh catalysts exist, which side of the bull/bear case got stronger, and what price level would prove the thesis wrong. If it can't articulate a specific reason to trade a name, it sits out. Five strong trades beat thirty weak ones.
Each trade gets a conviction score based on four factors: whether a fresh catalyst exists with a clear mechanism, whether related assets confirm the direction, whether the agent can explain why the thesis should play out today, and whether it can define the exact price where it's wrong. Strong conviction across all four means full size. Mixed signals mean smaller size. No thesis means no trade.
We also tightened the risk parameters significantly. Every trade is bracketed at entry with a profit target and a stop loss, and the targets are small and realistic, typically a 0.2% to 0.5% move capturing $100 to $500 per trade. Stops risk no more than $150 to $250, and if the agent can't define a clean bracket it doesn't take the trade. Stops never widen, only tighten, and once a position is up 50-60% of its target the stop moves to breakeven. The philosophy is simple: small captured gains and strict loss prevention, not home runs.
The agent also takes profits through the day rather than holding everything for the forced close. When a position reaches its target it scales out, when volume starts declining on a move it lightens up, and when a thesis has fully played out it exits. It can always re-enter after a pullback if the thesis still holds.
Technical analysis didn't disappear. The agent still runs RSI, MACD, Bollinger Bands, OBV, volume profiles, and 20/50/200-day moving averages on every name every cycle, but technicals now serve three specific purposes: timing entries so the agent doesn't chase, confirming that the market agrees with the thesis, and defining the stop where the thesis is wrong. You can't make money from technicals alone, but you can lose money by ignoring them.
This is what matters most to me. The Day Trader is a content product and every rationale it publishes is something people actually read, which changes what "good" looks like.
The old agent would output something like: "Buying NVDA on RSI bounce at oversold with MACD crossover confirmation." Technically accurate, but it sounds like every other trading bot on the internet.
The new agent tells a story. It explains that TSMC's overnight numbers were strong and Intel is indicating higher in pre-market, that the AI PC narrative is getting fresh legs from a packaging investment announcement, and that RSI is at 48 with room to run while the stock sits above the 50-day moving average as confirmation.
It narrates the trading day as it unfolds. It tells you Ford has been chopping between $13.80 and $14.10 for two hours and nobody's in control. It watches a stock bounce off $4.73 three times and explains that someone is defending that level with a limit order. It tracks a name stalling at $16.85 every rally and calls out the persistent seller capping the move. That's what I sounded like on the desk walking a client through my thinking, and it's the life of a salestrader narrated in real time.
The quants will always be faster and they'll always have more data, but they're not trying to explain anything. They optimize in silence.
I wanted to build something different: a system that shows you how a professional thinks about markets. Why this stock today, what the overnight signals suggest, where the key levels are, and what it means when they hold or break. The agent makes 20-50 data calls per cycle to validate a single trade, tracks bull and bear narratives, cites specific price levels and catalysts and international signals in every rationale, and brackets every position with discipline that mirrors how I was trained to manage risk.
That's Terminal X's actual edge. Not speed, not scale. Narrative intelligence applied to trading, explained in the open. I spent 12 years developing this way of thinking and now an AI runs it every five minutes.
The agent is live. You can watch it think, trade, and explain itself every day from open to close.
by: Jacob Koenig
