The shared spreadsheet dies by week three
Every couple who decides to combine finances hits the same wall. Someone builds a shared spreadsheet, you both promise to log expenses, and it works for about two weeks. Then one person falls behind, the other quietly resents doing all the data entry, and the sheet becomes a monument to good intentions.
The problem isn't discipline. It's that a spreadsheet asks two people to hand-copy data that already exists, perfectly and completely, inside two banks. And the conversation the sheet was supposed to support (who pays for what, are we overspending, is this split fair) never happens, because nobody trusts numbers that are three weeks stale.
There's a better setup now: connect both banks to one AI agent and ask it the questions directly.
An agent has no stake in the argument
Money conversations between partners go sideways for a predictable reason. Whoever presents the numbers is also a party to the dispute. If you built the spreadsheet, your categories win. "Eating out" versus "that was travel, it doesn't count" is an argument about data entry wearing a costume.
An agent reading live data from both banks changes the dynamic. It doesn't remember who picked the expensive restaurant. It doesn't round in anyone's favor. When it says one of you spent $410 on takeout last month, that's not an accusation. It's a query result.
You're no longer arguing with each other about whose numbers are right. You're both looking at the same numbers, produced by something with no side to take. That's the whole trick, and it works better than it has any right to.
The setup: two banks, one key
BankBridge is a hosted MCP server that gives AI agents read-only access to bank data. One of you creates an account at bankbridge.money and connects your bank through the secure connection flow. Then your partner connects theirs on the same account, logging in with their own bank credentials. Neither of you ever sees the other's banking password; the flow talks to each bank directly.
Pricing is $5 a month per connected bank, so a typical couple pays $10. Cancel anytime.
Grab the API key (it starts with bbk_) and add BankBridge to whichever AI app you already use. Claude Desktop, Claude Code, ChatGPT, Cursor, Gemini CLI, Zed: there are setup docs for 29 hosts. You don't even need to use the same app. The same key works in both, and both agents see the same live data.
From here on, one agent can see both banks. Nothing is cached on BankBridge servers, so every answer reflects the accounts as they are right now.
Start with the combined picture
Before the pointed questions, get a shared baseline. Ask something like:
List every account across both banks with current balances, grouped by owner.
What was our combined spending last month, broken down by category? Show each person's share within each category.
Show our combined monthly cashflow for the last six months, income versus spending.
Under the hood the agent is calling tools like list_accounts, get_spending_summary, and get_monthly_cashflow across both connections, then merging the results. You don't manage any of that. You ask in plain English and get one answer that covers both of you. Do this part together, laptop between you on the couch. Seeing the combined burn rate for the first time is usually the moment this stops feeling like an audit and starts feeling like a joint project.
The awkward questions, asked out loud
Now the ones couples avoid for months.
Over the last 90 days, who paid for groceries, and what was the split in dollars and percent?
Compare our discretionary spending: dining out, shopping, subscriptions, entertainment. Person by person, last three months.
Which shared expenses (rent, utilities, insurance, groceries) come out of whose account today?
These questions are radioactive when one partner asks the other. They're routine when both of you ask an agent. The answer arrives with merchant names and dates attached, so there's nothing to relitigate. One caution: raw payment data shows who swiped the card, not who ultimately bore the cost. If one of you pays the electric bill and the other reimburses in cash, tell the agent, and it'll account for the transfer instead of counting it twice.
Duplicate subscriptions across two households
Moving in together usually means merging two full sets of subscriptions. Two video streaming plans. Two music plans that could be one duo plan. Two cloud storage tiers backing up to the same ecosystem.
Pull recurring charges from both of our accounts and flag any service that appears on both.
Which streaming services are we paying for twice, and what would we save by keeping one of each?
The get_recurring_charges tool finds subscription patterns in each bank, and the agent cross-references the two lists. Descriptors don't always match between banks (one might show "Spotify USA" and the other "SPOTIFYUSA 877-77"), which is exactly the kind of fuzzy matching agents are good at and spreadsheets are terrible at.
Most couples find $30 to $60 a month of pure overlap on the first pass. Not life-changing, but it's the easiest money you'll ever agree on.
Work out a fair split with real numbers
The classic joint-account question: 50/50, or proportional to income? An agent can't make that values call for you, but it can price out both options so you're choosing between real numbers instead of vibes.
Estimate each of our take-home incomes from deposits over the last three months. If we fund a joint account to cover shared expenses (rent, utilities, groceries, insurance), what would each of us contribute under a 50/50 split versus a proportional split?
The agent reads deposit history from both checking accounts, totals your shared categories from real transactions, and lays the two scenarios side by side. If your incomes are lopsided, proportional usually feels fairer within about ten seconds of seeing the numbers.
Rerun it quarterly. Incomes change, rent changes, and a split that was fair in January can drift by summer. It's a two-minute prompt, not a renegotiation.
What your partner can see, and how to undo it
Be clear-eyed about what you're setting up. Both banks on one key means the agent, and anyone holding that key, can query both banks. That's the point, but it deserves a plain sentence rather than fine print.
You control the scope. During the connection flow, each partner picks which accounts to share. Plenty of couples connect checking and the shared credit card while leaving personal accounts off entirely. The agent only ever sees what you explicitly connected.
And everything is reversible. Access is read-only forever (there's no tool that moves money), either bank can be disconnected at any time, and the API key can be rotated so previously configured apps go dark. Because nothing is cached, disconnecting doesn't leave a copy of your history behind. There is no copy.
Make it a ritual, not a project
The spreadsheet failed because it demanded ongoing labor. This setup demands a monthly conversation, which is the part you wanted anyway.
Pick a recurring twenty minutes. Ask the combined-picture questions, check the duplicates, glance at the split. When something looks off, ask a follow-up right there instead of assigning homework.
Give us a monthly money review: combined income, combined spending by category, biggest changes from last month, and anything unusual on either account.
One prompt, both banks, no data entry, and nobody had to be the accountant in the relationship. That's the version of combining finances that survives past week three.