# We Let Our Financial Planners Write the Agents Matthew Ragone, Range | AI Engineer NYC 2026 | 08:54 Source: https://www.youtube.com/watch?v=gpmze-6ueGU Channel: AI Engineer (https://www.youtube.com/@aiDotEngineer). Summarised by AIE Talks. Page: https://aietalks.com/talks/we-let-our-financial-planners-write-the-agents Published: 2026-10-12 Tags: human-in-the-loop, testing, workflows ## TL;DR - Financial planners at Range build agents without writing code because they know the firm's decisions, checks, and limits. - An agent can draft a cash flow or insurance analysis while a planner is busy, but a planner reviews every recommendation before it reaches a client. - A reliable skill comes from choosing a judgeable workflow, teaching it with real cases, testing it repeatedly, and assigning an owner and a separate reviewer. ## Summary Matthew Ragone describes how Range's financial planners built agents for their own work without writing code. The change was moving from chatbot assistance to drafts that happen while a planner is occupied with another client. Agents can prepare cash flow and insurance analyses, but planners review and approve every recommendation before it is sent. Ragone gives a five-step process: choose a repeated workflow whose quality is easy to judge, teach the agent with real cases, package the planner's judgment into a skill, test it five to ten times, then ship it with an owner and a separate reviewer. The instructions must include the questions the agent should ask and the actions it must never take. Six months in, Range had 24 skills across the company, including 17 in planning. The harder problem now is deciding who owns the instructions and how quality will be checked as more people rely on them. ## Key ideas ### The useful shift is work happening while the planner is busy [01:28](https://www.youtube.com/watch?v=gpmze-6ueGU&t=88s) Ragone says typing speed was never the main constraint for a financial planner. A planner can only be in one place at a time, so making a prompt faster does not create much capacity. At Range, an agent drafts a cash flow analysis or insurance analysis for one client while the planner is on a call with another. After the call, the planner opens the draft, reviews it fully, changes what is needed, and sends the recommendations. The agent never sends anything directly to a client. ### A skill captures how Range does a specific piece of work [02:39](https://www.youtube.com/watch?v=gpmze-6ueGU&t=159s) The agent has to work where the firm's work already lives, rather than in a separate chat window that requires copying and pasting. It also needs to know Range's process, including what planners check first, what they always flag, and how they phrase recommendations. Ragone calls this combination a skill: the firm's way of doing one specific piece of work, written so an agent can follow it. The people who know the work should build the skill, and none of the five steps requires code. ### Start with work that is repeated and easy to judge [03:16](https://www.youtube.com/watch?v=gpmze-6ueGU&t=196s) Ragone advises against starting with the flashiest or most complex workflow. The first choice should be something the team does often and where a planner already knows what good work looks like. He also recommends workflows with substantial manual effort, even if they are relatively simple. At Range, cash flow analysis and insurance reviews fit that description. If a person cannot judge the output quickly, the team cannot test the skill properly or trust its drafts. ### The planner's reasoning matters more than a list of clicks [04:22](https://www.youtube.com/watch?v=gpmze-6ueGU&t=262s) To teach a skill, planners should walk through several real cases from start to finish. They should explain why they open a document, what they are looking for, and what would make them stop and ask a question. A sequence such as opening a particular file is easy to record, while the judgment behind that action makes the workflow specific to the firm. The skill also needs the questions it must ask before starting and the limits it must follow. At Range, it drafts recommendations but never sends them to a client. ### Repeated testing exposes missing questions and confident mistakes [05:47](https://www.youtube.com/watch?v=gpmze-6ueGU&t=347s) Ragone recommends running a new skill five to ten times on real cases before anyone else sees it. Each run can reveal an edge case, a question the skill should have asked, or a draft that sounds confident while being wrong. The team updates the skill, saves it, and runs it again. He compares this work with the larger cost of handing an unfinished skill to many people across the company. ### Ownership keeps a skill aligned with changing practice [06:14](https://www.youtube.com/watch?v=gpmze-6ueGU&t=374s) A shipped skill needs an owner who keeps it current and a reviewer who checks it before the team relies on it. Processes and tax rules change, so instructions that were right in March may be wrong by December. The reviewer should be someone other than the author, following the same separation used when planners review each other's plans. This turns an individual's shortcut into something a team can depend on. ### Subject matter experts made the skills more specific [06:57](https://www.youtube.com/watch?v=gpmze-6ueGU&t=417s) After six months, Range had 24 skills across the company, with 17 in planning. Sales, product, and marketing began building their own agents without a mandate. Ragone says the resulting skills became more specific because the person who knows insurance best builds the insurance skill, while the person who knows tax law best builds the tax skill. Peer review also carried over naturally from planning work to skill review. ### The remaining problem is governance beyond the engineering team [08:10](https://www.youtube.com/watch?v=gpmze-6ueGU&t=490s) Ragone says the question of whether a model can draft a cash flow analysis has already been settled at Range. The harder questions concern who writes down the firm's method, who tests it before release, and who owns it after many colleagues depend on it. As agents spread beyond engineering, these responsibilities become part of the work. He is direct that there is not much field data yet, and presents Range's process as its current answer. ## Notable quotes - "What fixes it is work that happens while I'm somewhere else." (01:28) - "The agent doesn't send anything. A planner reviews every single recommendation before a client sees it." (02:15) - "The steps are easy to write down. The judgment is what makes it your firm's work instead of generic advice." (04:43) - "Every run teaches you something the skill didn't know yet." (05:47) - "A couple of things changed once subject matter experts started building for their own niche. The skills got a lot more specific." (07:26) ## Tools & references mentioned - Range - Claude ## Who should watch - You are a subject matter expert who wants to build an agent for repeated work without becoming a programmer. - Your team is drafting regulated or client-facing material and needs a clear review boundary before anything is sent. - You are rolling out internal agents and need owners, reviewers, and a testing process for instructions that will change over time. ## Editor's note Matthew Ragone says repeated runs expose missing questions and drafts that sound confident while being wrong. Kitaru records each agent run, including model calls and tool results, as a session. Replay runs the changed agent against that recording with the same inputs and tool responses, so the team can check whether a prompt, model, or code change fixes the recorded failure without touching real systems. Written by the AIE Talks editors (the Kitaru team), not by the speaker. ## Related talks - [How to Build Planning Agents without Losing Control](https://aietalks.com/talks/how-to-build-planning-agents-without-losing-control) (Yogendra Miraje, FactSet, 15:58) - [Your Finance Agent's Bottleneck Is You](https://aietalks.com/talks/your-finance-agents-bottleneck-is-you) (Ramana Siddanth Emani, Auditoria AI, 13:42) - [From Copilot to Colleague: Trustworthy Agents for High-Stakes](https://aietalks.com/talks/from-copilot-to-colleague-trustworthy-agents-for-high-stakes) (Joel Hron, Thomson Reuters, 19:45) - [Trust, but Verify: Knowledge Agents for Finance Workflows](https://aietalks.com/talks/trust-but-verify-knowledge-agents-for-finance-workflows) (Mike Conover, Brightwave, 21:10) - [Agents Don't Do Standups: Building the Post-Engineer Engineering Org](https://aietalks.com/talks/agents-dont-do-standups-building-the-post-engineer-engineering-org) (Mike Spitz, PFF, 17:50)