Most finance teams have started using AI to ask questions. Fewer have used it to build. This workshop is about the second thing. Charlie, who leads finance at Cube, and Rob Dandorph, Cube's VP of Operations, will show how they use AI to build real, persisting tools: dashboards, reports, and automated workflows that hold up month after month. Not a screenshot of a chat, but working software you deploy once and refresh forever. This is the first session in a series built specifically for finance teams, and you'll leave with prompts and templates you can paste into your own AI the same day. We'll cover:
What vibe coding is, and why finance should care
Vibe coding is a way of building software without writing code. You describe what you want in plain English, the AI builds it, you give feedback, and it refines. Describe, generate, refine. That's the whole loop, and it doesn't require an engineering background. For finance, that matters more than it sounds. Finance is the operating system of the company: the close, the budget, the P&L, the scenarios the board sees. Yet the tooling has barely changed in decades. Vibe coding closes that gap, letting you build the reporting and workflows you always wanted without waiting in an engineering queue, so the manual assembly gets automated and the time goes back to analysis.
What you can actually build
Automated month-end reporting with written commentary on variances. A Monday morning variance alert that lands in Slack on its own. Agent workflows that map new GL accounts or flag mismatched data. Interactive dashboards you filter by department, period, or scenario instead of rebuilding in a spreadsheet. We'll walk through the building blocks that make this possible, using Claude's names for them: skills (reusable capabilities like "build a PowerPoint deck" the way you defined it), artifacts (live, interactive pages loaded with fresh data), code (full workflows that run and deploy across systems), and agents (work that runs in the background without being prompted). Other platforms have their own versions of the same ideas, so what you learn carries over.
How to work with AI the right way
The fastest way to a good result is to slow down at the start. We'll share the approach Charlie uses on every build: state the goal clearly, tell the AI not to build anything yet and to ask clarifying questions first, then have it summarize the plan in plain English for your approval before it writes a thing. We'll also cover using the right model for the right job, a stronger model for planning and a faster one to execute, which keeps quality high and cost low.
Accuracy: why the data layer decides everything
This is the part finance cares about most, and rightly so. AI can make calculation errors, miss line items, and produce numbers that look right but aren't. We ran a study on exactly this. Given a clean, detailed Excel model, an AI built a variance dashboard that was roughly 60% accurate. Connected to Cube through the Cube MCP Server instead, the same build was 100% accurate, because Cube gives the AI a structured single source of truth to read instead of a spreadsheet to interpret. The takeaway is simple. Vibe coding on messy data is still messy. Vibe coding on Cube is reliable.
The demo: a budget versus actuals dashboard you refresh, not rebuild
The main event is a live look at the actual budget versus actuals dashboard Charlie's team uses every month, running on scrambled demo data. We'll show the monthly toggle, the rules that flag any expense meaningfully over or under budget, and the drill-down from department to category to individual vendor. The auditability is the star: hover over any number to see the exact journal entries behind it, pulled straight from Cube. We'll also show the notes, the export to Sheets safeguard, the one-click refresh, and a Claude Code workflow that DMs the right department head about a flagged expense and writes their reply back into the dashboard automatically. You'll leave with a prompt, dropped in the chat during the session, that builds the shell of this dashboard for you in minutes.
Making it real: safe, auditable, and built to last
Building it once is the easy part. Making it survive is what separates a webinar demo from something your team relies on. Rob will cover the guardrails: one source of truth, citations on every number so you can trace it back, version control, a delegation of authority so nothing ships without review, read access before write access, and a review mode that catches jobs that fail silently. We'll also cover what your security and IT teams will want to know before you deploy, with language you can copy and send.
Plus: prompts to take with you, and what we build next
Throughout the session we'll drop prompts and instructions into the chat that you can copy and paste into your own AI, including the budget versus actuals build and the Claude Code automation. Because this is the first in a series, we'll close by asking what you want us to build next, followed by live Q&A.
Charlie Erlikh
Sr. Finance and Operations Manager,
Cube Software
Rob Dandorph
VP of Finance and Operations,
Cube Software