It’s easy to scroll LinkedIn, see someone talking about how they automated half their job with OpenClaw or Claude Code, and feel like you’re way behind.
You’re not.
Half the attendees in our recent webinar "Build Your First Finance Workflow with AI" had never vibe coded anything before. An hour later, they walked away with valuable know-how on what’s possible and how to get started.
That alone puts them ahead of most finance teams. You can join them by watching the recording.
Here are some of the highlights from the session.
Claude vs. Claude Code vs. Claude Cowork
Claude is quickly becoming a go-to LLM for finance teams, and it comes in a few different flavors. Choosing the right one is the first real decision you need to make.
So Claude has three forms now. There's so then there's chat, Cowork, and Code. Chat is really if you just have questions for it. Like, according to the accounting standards, what's the right way to do this? Cowork is for more for, like, building things. Any sort of artifact, whether it's a, like a Google Doc, whether it's a slide, whether it's like your own software app, it's really for building. And then Code is arguably the more advanced one, more frequently used by engineers. So for me, my recommendation for for the half of you that have never built anything, start with something like Cowork or Replit. And then for those that are looking to take the next step, you can go into, like, a Claude terminal. You can go into Cursor. You can go into, more advanced use cases.
Most teams default to Chat because it’s the most familiar. That’s a good starting point, but using Chat exclusively leaves a lot of value on the table.
Here’s a simple framework for deciding which to use:
- Chat is best for one-off questions. You’re not trying to build anything—you just need an answer. Example: What’s the right revenue recognition treatment for this scenario?
- Cowork is a good entry point into vibe coding. Describe what you want to build in plain language, and Claude will create an app for you. From there, you can iterate by describing changes you want to see. Example: Build a budget vs. actuals tracker with variance flags and a summary view for execs.
- Code is more flexible than Cowork, but it’s also most effective if you have some coding knowledge. Leave this one to the engineers if you prefer to stay out of your machine’s terminal. Example: building something that needs to plug into your data warehouse or handle more complex logic.
Balancing quality and speed
Because LLMs can answer any question in seconds, waiting a few minutes for them to build an app can feel like an eternity.
But you don’t have to twiddle your thumbs as the LLM does its thing. Both John and Albert mentioned they’ll check email or do something else while the AI works in the background.
John was saying, I thought this was supposed to be ultra fast. Can you talk about speed? Yeah. I consider it was pretty fast. I mean, it's building a piece of software for us in, you know, I don't know, five minutes, but I totally hear you. So how I think about it is my order of operations is quality, speed, cost. So I think, like, this session is meant to be more of, let's just make sure we get the right output. And so with quality and, like, with the most advanced models, you're giving up a bit of speed and you're giving up a bit on cost. I think once you're comfortable with how you get the best quality and you're using it a lot, then I would start to think about speed, John. And then I'd start to think about optimizing for cost as kind of the third one to focus on. But just for this exercise, we're just focusing on quality.
John’s vibe coding priorities in order:
- Quality
- Speed
- Cost
You want the app to be good. If that takes a bit of extra time—whether from the LLM itself or from a few rounds of iteration—that’s time well spent.
Once you’re consistently getting high-quality results, then it makes sense to focus on speed. And after that, cost.
The cost piece mostly comes down to token usage. That will vary by team, so it’s worth understanding how your org tracks usage and what kind of spend makes sense for experimentation vs. production-grade workflows.
An app to help AEs with deal hygiene
When thinking about what to vibe code, start with recurring pain points. What’s something you or your business partners do every week that’s manual and annoying?
Deal hygiene fits that sweet spot. AEs want to spend their time talking to prospects, not cleaning up data in HubSpot.
So Albert built a live sales tracker to make it easier to stay on top of deal status and reduce friction for AEs to provide updates.
So one of the things that we do internally is we always want the AEs to keep our HubSpot updated. But we know that in the end of the month, there's a lot of like conversations happen super fast and we wanna have a very quick pulse on how deals are going or not. So what I'm gonna do is I'm gonna create kind of a live tracker for how are we doing towards end of month. What I'm gonna do is I'm gonna actually create like a folder and it's gonna ask me, can I change files in this folder? I'm gonna say, you know, go for it. And if you see what the file, you know, this in my computer, it just has access to the Salesforce opportunities Excel file. So what I'm gonna do is, you know, can you build an artifact? I want to see the ARR broken down by all deals. I also want to be able to ask our AEs for updates in Slack using the Slack MCP I'm expecting an artifact, but I also want to be able to connect to Slack and integrate with that. This is what I bought, which is I got a tracker where all of the deals that I had in the spreadsheet. So if I'm here, and I go to, I don't know, Sunnyvale Semiconductor one hundred forty six. Okay. I have the opportunity number. I just wanted to, all of those track one to one. And then what I can do is, Jordan has twenty deals. I wanna ask Jordan for the update on those twenty deals. I can say, okay, I wanna post to Slack. Once I post to Slack, I go to Slack and you can see here, you know, a minute ago, I am asking send using Clog. Hey Jordan, these are all your deals. Are there any updates? Hopefully this feels like a much lower friction for like quick updates way of asking AEs for their input.
From a single build prompt and a few follow-ups, he created an app that:
- Pulls in deal data
- Displays a live snapshot of pipeline health
- Pushes updates directly to Slack to get real-time input from AEs
Hours of recurring manual work gone after a bit of vibe coding.
Building a sales commission calculator
John went a similar route in his demo, showing off a sales commission calculator.
We're gonna build a sales commission calculator today off of the sales team's comp plan. And we can go in here and we can grab our comp plan. I actually asked Claude to write this comp plan. So if you don't have a comp plan yet, you can actually start there. We of course can edit it and give it some feedback, but we'll go ahead and take it as is. I'm gonna take our Q4 sales team performance, and then we've got all the deals. And again, totally made up data that Claude generated for me. And so what I'm gonna do is I'm gonna drop this in. And so we got our comp plan and we got Q4 rep performance. It's gonna build our journal entry at the end, show rep performance at a deal level and all calcs. So here's everybody's performance. And what it did was it took that comp plan that I gave it and it built out. So Priya is on her way to P club. We got Marcus as an accelerator, folks below quota. You know, some people underperforming. I have found they will hallucinate. So I created some checks and balances here just to just to make sure that it's like it's tying things out. And then at the end, we got this journal entry.
You can tell both John and Albert have spent a lot of time vibe coding, because they don’t build apps that just do one thing. Albert’s deal hygiene app gives him a pipeline health dashboard AND automatically pushes updates to Slack. John’s sales commission calculator provides a snapshot of sales team performance AND preps a journal entry.
The only limit on what you can build is what you can imagine.
This is cool, but can I trust these homemade apps?
Any generic AI output should be approached with a healthy level of skepticism. Hallucinations continue to be LLMs’ Achilles’ heel—at least those that don't know your data.
Building checks into your apps is a great way to ensure their accuracy.
This is exactly, Marina, what I would do to ensure accuracy. Like, you know, wrong way to do it or like the way that it's very hard to audit is, you know, I want an app that I drop a Excel file and it gives me the final number. It gives me the eight thousand nine hundred that you have there in the bottom right. Great way of doing it is what John built here, where you have all of the breakdown of how the math has been calculated. And I think it's not, you don't need to grab a calculator and redo all of those. If you can see, okay, I see that it's doing four and thirty seven point five ks, that is the total amount times ten percent commission. I can like see the math and see the breakdown here. And that makes me trust it much more.
You’re not behind. Just start building.
If there’s one common theme across all the vibe coding content we’ve put out recently (and there’s been a lot!), it’s that the best way to learn is to just start building. Pick a use case that we’ve covered or choose your own. Fire up Claude Cowork and have it build something for you. Iterate and let your imagination run wild.
To see several more use cases that John and Albert walked through, check out the full webinar recording.
And for a comprehensive vibe coding starter kit, download our vibe coding playbook.
.png)
_EasiestSetup_Mid-Market_EaseOfSetup.svg)
