I run an agency based in Haiti, and we started using QuickBooks in 2019.

That was the first version, and in theory it made sense. The accounting data lived there, the actions lived there, and if you needed something financial you went to QuickBooks and did it there.

In practice, it still felt like work.

Why QuickBooks alone was not enough

Fast forward to 2025. Claude Code entered the picture, and that changed what was possible. We built the first Intuit app. We got access to QuickBooks MCP. Suddenly we were interacting with it.

At first, that was mostly me. If I wanted a report, I could just remind myself to do it. If I needed to create something, I could move fast because I was already one step ahead of the work. But the account director did not have that. She had QuickBooks, the same way everyone has QuickBooks, which means she had to log in, find the right screen, click through the right sequence, and hope she remembered where everything lived.

That is not a skills problem. That is a friction problem. And friction compounds every single week.

The Telegram bot that worked for three days

So I built the Telegram bot. Route the request through Telegram, write it to the MCP, let her work with QuickBooks the same way I was already working with it. Natural language in. Financial action out.

That worked for about three days.

Then it started to feel wrong again. Telegram was fine for routing, but it was one more app in a rotation that already had too many apps. She was not living in Telegram. She was living in Google Workspace. That is where she talked to me, to the rest of the team, where the actual day to day coordination happened. Asking her to leave that context just to do a financial action was adding a step while pretending to remove one.

Moving the financial layer into Google Workspace

So I changed it again. I built the Google Workspace app.

And that is when it finally started feeling right, because now she could use natural language inside the place she already trusted. She did not need to learn a new habit just to do financial work. She could ask for what she wanted and move.

Create an invoice. Void something. Add a new client. Pull a report.

QuickBooks stayed QuickBooks. The change was that the interface stopped making her work harder than the work itself.

What we actually built on top of that

Once the conversational layer worked, we could start building the things that matter.

HB receipt was one of the first. The kind of thing that sounds small until you actually have to do it over and over. The receipt pulls data from QuickBooks, populates a template that does not live in QuickBooks, and sends it through the email platform without the Account Director doing anything beyond asking for it. One sentence in Google Chat, and the receipt goes out.

HB reminder goes further. It does not just look at QuickBooks. It checks the inbox too, pulls the relevant financial context, figures out what needs to be surfaced, and turns that into a reminder the account director can act on immediately. Before this existed, she had to cross-reference two systems manually and decide what was urgent. Now the agent does the cross-referencing and she decides what to do about it.

The financial pulse fits the same logic. Twice a week, a one-pager gets populated with the right numbers so we can actually see what matters without spending half an hour assembling it. Monday and Friday. No one has to build it. It builds itself from what QuickBooks already knows.

The real cost is never the mistake

The part I care about is the removal of drag.

Because in finance, the cost is rarely the mistake itself. It is the delay. The report that exists too late to help. The receipt that takes too long to send. The reminder that should have been automatic but instead sat in someone's head as a thing to do later.

Every action used to ask for one more step than it should have. Open this. Click that. Remember where it lives. Do it later. Then do it again next week. That is where time disappears, and once you see it you cannot unsee it.

Building software that fits the person instead of the other way around

We built this so the account director could move with the same efficiency I had been using for myself. That was the point. Make the work faster and easier to repeat.

QuickBooks remained the source of truth. Claude Code helped us build the bridge. Telegram was the first attempt at a conversational layer. Google Workspace became the better home. The MCP sat underneath all of it.

And honestly, that is what good internal software should do. It should not ask your team to become better at using the tool. It should become better at fitting the team.