For the complete documentation index, see llms.txt. This page is also available as Markdown.

Coding Agent vs Finance Agent

Short answer: A coding agent writes and builds software. A finance agent discovers, allocates, and executes capital. Same architecture — model + harness + tools + memory — different object. Codex and Cursor are AI coding agents. Questflow is an AI Finance Agent: Codex for finance.

The analogy that actually holds

Coding agent
Finance agent

Object

Code, tests, PRs

Capital, risk, orders

Loop

Explore → Plan → Implement → Verify

Signal → Thesis → Execution → Post-mortem

Identity file

CLAUDE.md / AGENTS.md

invest.md

Danger

Broken production

Broken account

Safety

Hooks, tests, review

Kill switch, limits, scoped permissions

Proof

CI green, review

Live performance ledger

Scarce input

Taste in a codebase

Investor judgment

Coding agents crossed the line when they stopped autocompleting and started owning tasks. Finance agents cross the same line when they stop chatting about charts and start closing the loop inside your rules.

Where the analogy breaks

Markets punish you in a way compilers do not. A failing test is free. A failing thesis costs money. That is why Questflow insists the alpha source is human fund manager judgment, not a prompt that "generates a strategy." AI scales judgment. It does not invent it.

Also: code is usually one repo. Capital is fragmented across brokers, wallets, and venues. A financial harness has to read that whole picture before it writes an order.

What to use when

  • Building software → Codex, Cursor, Claude Code

  • Allocating capital with someone else's proven framework → Questflow Funds

  • Wanting a personal AI to vibe-trade your own untested ideas → that is a different product category (see Questflow vs Minara)

Related: What is an AI Finance Agent? · Questflow vs Codex · What is a financial harness?

Nothing here is financial advice.