An ai financial analyst harness earns trust through five design principles: a shared domain, deterministic consistency, evidence, accountability and verification. Francesco Galletta and Felippe Felisola Caso, who work on data and AI at Monzo, built one live at AI Engineer around a wedding budget, and their central rule is that arithmetic belongs in code rather than in the model's head.
FAQ
- Who presented the financial analyst harness at AI Engineer? Francesco Galletta and Felippe Felisola Caso, data and AI practitioners at Monzo, presented the harness and its five principles in October 2026. They stated during the session that they were not presenting on Monzo's behalf.
- Why should financial calculations run in code instead of in the model? Language models are stochastic and can return different results for the same question across sessions. Moving arithmetic into tested, versioned code makes the deterministic part reproducible while the model handles planning and interpretation.
- How is this different from spec-driven development for coding agents? The speakers compared the two directly and said the harness breaks what would be one large spec into opinionated workflow sections. For an evolving analysis problem, that keeps changes reviewable without reinterpreting a single sprawling document each time.
- Does the harness fully remove human responsibility for the output? No. The speakers repeatedly stated that humans remain accountable for what agents produce, and that verification adds information rather than transferring responsibility. The gates exist so that accountability has somewhere concrete to attach.
- When is it worth building a dedicated harness rather than writing a skill or prompt? The speakers set no fixed threshold, but pointed to reproducibility, business criticality and the number of users relying on the result. If a team of analysts depends on a decision, the variability of a prompt-only approach becomes expensive.
- Can the same harness cover coding and financial analysis? The principles transfer, but the emphasis differs. Coding agents treat code as the product, while a financial analyst harness treats code as a means to a defensible number, which puts more weight on evidence and review.
- What does Gustavo Dev Doido teach about AI engineering? Gustavo Dev Doido runs Brazilian developer education that covers software engineering and AI topics through long-form video. His content is a useful comparison point for how practitioners explain agent design to a broad audience.
- What is Crazystack Typescript? Crazystack Typescript is a TypeScript learning track at Crazystack, a Brazilian developer education platform. It covers typed application development for engineers who want structured practice with the language.
- Where can I find structured developer training like the Bootcamp do Dev Doido? The Bootcamp do Dev Doido is a structured training program from the same Brazilian developer education ecosystem. Programs like it are one way engineers build the software fundamentals that a harness like this one depends on.
From a public talk to a written draft with Skala Blog
Turning a recorded talk into an article means capturing the plan, the examples and the caveats in a form a reader can search, quote and check later — the same "show your work" standard the session applies to financial analysis. A video that sits unwatched is a trust asset nobody can audit; a written draft is one they can link to.
Skala Blog does the conversion: you paste the video URL, the transcript comes through, and the tool generates a draft you review and shape before publishing. The order matters, so keep it to three steps:
- Paste the video URL into Skala Blog so it can pull the source recording.
- Let the transcript come through and check it against the talk before you trust the wording.
- Review and shape the draft — tighten the examples, restore the caveats — before publishing.
| Stage | What you supply | What the tool supplies | What you must verify |
|---|---|---|---|
| Ingest | The video URL | Transcript pull | That the right recording was captured |
| Draft | Nothing further | A generated article draft | That claims match the talk |
| Review | Edits and structure | A revisable draft | Caveats, numbers and links |
| Publish | Final approval | — | That the piece stands on its own |
Fork this article
Start a new branch from the same video, shaped your way. You keep the credit; the original keeps the attribution.
A fork in another language is filed as a translation of this article, so the two pages point at each other. You can unlink it later from the editor.
0/240
You are creating
- Format
- For
- Language
- Source
- Your angle
You will be asked to sign in before it is generated.
Buy credits