Vincent Codes Finance → interactive research workflow
Let the agent move fast.
Keep the researcher in control.
Vincent’s Claude Code tutorial is not just a tool walkthrough. It describes a control system: add context, plan before writing, constrain permissions, run and inspect, repair failures, manage context, then preserve the work. Move the levers and watch that system change.
▶ Start at the project-control setup · 17:14Analysis run · FOMC → VIX
A reviewed plan, bounded permissions, and live execution feedback keep the research loop inspectable.
Enough room remains; compact at a logical boundary rather than mid-task.
Transcript-grounded evidence
Every lever comes from the workflow.
Open any timestamp to hear Vincent explain the source decision. The model connects those decisions; it does not claim that the scores are experimental measurements.
Context, permissions, review
Claude.md carries project rules into every interaction; broad permissions create risk, and the researcher remains responsible for reviewing mistakes.
Plan before implementation
Planning first avoids wasted work and tokens. The researcher simplifies an overbuilt plan before allowing edits.
Run, inspect, interrupt, repair
Execution returns errors to the agent; live inspection catches an unwanted dependency edit and a missing line in the generated figure.
A notebook is not yet verified
The generated notebook still needs to be run. When it errors, the workflow reads, fixes, and reruns it.
Context creates pressure
History and tool calls accumulate. Clearing or compacting at a logical boundary avoids lossy compression at a bad moment.
Preserve the session
A final Git commit documents what changed and gives a future agent a reliable record of prior work.
What must remain true
The agent may accelerate implementation, but the researcher still defines the question, approves the plan, inspects outputs, and owns mistakes.
What would change the model
Measured task results—time, token use, defects caught, reproducibility failures—would replace these illustrative scores with evidence.
Built as an unsolicited, transcript-grounded demonstration of what YouTube Conversation can create from a public captioned tutorial. Vincent Codes Finance did not sponsor or endorse it.