Subaru

AI engineer | swag owner

[Subaru @0xSubarru]

Anthropic’s Margot van Laar just dropped one of the cleanest free breakdowns on prompt engineering for real agent systems

0% → 00:00 - why “better prompting” starts with evals, not vibes
25% → 03:40 - how to clean a broken production prompt after a model migration
50% → 07:10 - why structure, output contracts, and tools beat endlessly adding instructions
75% → 10:20 - how to build an agentic workflow instead of one giant mega-prompt
100% → 13:20 - why Generate → Evaluate → Repair is the loop that actually scales

Most people still treat prompts like magic text

She treats them like production systems

Evals → Structure → Tools → Agent Loop

That same mindset is exactly what matters in quant

Claude can generate the strategy

Python can test it

But the system still needs to verify, reject, and repair before anything reaches capital

Watch the full 14-minute session today

Then read the full AI trading system below ↓

[kvro @0xkvro]

https://t.co/xU1TXhObIR

[Subaru @0xSubarru]

Anthropic just showed Claude Code doing the part most people still treat as manual engineering work

And that same workflow is exactly why Claude is so interesting for quant

Give it one hypothesis

Let Claude Code turn it into a full research loop:

Idea → Generate Variants → Write Python → Backtest → Stress Costs → Walk-Forward → Reject / Keep

The important part is not asking Claude whether a strategy is “good”

Claude should build the experiment

Python should compute the result

The validation layer should try to break it

And the human should still control the capital

That is where Claude Code becomes more than a coding assistant

It becomes the operating layer around an entire quant lab

Watch the demo

Then read the full Claude Code AI Trading Lab guide below ↓

[kvro @0xkvro]

https://t.co/xU1TXhObIR