# Hand five credit cards and their statements to an agent: it mapped every reward and swapped cards where they fit, with the author claiming at least $1,000 saved - basis not stated

Meta Muse · X (Twitter) · No earnings are disclosed. The author claims at least $1,000 saved on card rewards; that is spending he says he avoided, with no basis given and nothing verified.

## Business summary

The author says he holds five credit cards with real annual fees and never knew which reward sat on which card, so he handed the problem to Muse. By his account, the agent found every reward on every card, took his credit card statements and the apps he has, worked out where each spend belonged, and swapped or loaded the card wherever it needed to go. The figure he gives is no less than 1000 bucks saved so far, without saying how that was calculated, over what period, or whether it is cash actually received or the estimated value of rewards. In the same post he lists other uses: dinner reservations with deep dives to find restaurants that have a table, email cleanup with a daily note on what needs follow-up, unsubscribing from the marketing mail filling his gmail, drafting and sending email, hotel pricing and planning, and researching and buying a laptop for his daughter. He states plainly that his own approach is to give it his logins and let it run, and says he intends to give it access to almost everything. That is his choice, not what this site recommends. The $1,000 is spending he says he avoided, not income, and it cannot be verified.

## Business value

The information is what defeats people, not the arithmetic. The rates live on bank pages and inside apps, the spending lives in statements, and the bindings live in a dozen merchant backends, so doing it by hand means twenty open tabs and most people never finish. Flattening it into one table is mechanical and is the part worth delegating. What cannot be delegated is the change itself - moving a binding touches automatic payments and, behind that, your credit - and what cannot be known in advance is how much any of it saved: that only exists once a real statement closes.

## Workflow

Decide the scope and how it gets material: list the cards - product, issuer, annual fee - and decide that it receives a read-only copy of your statements rather than your logins. Have it restate the scope and the fields before it starts.

Lay the rewards out card by card and category by category: category, rate or multiplier, cap, expiry, and whether activation is required. Only what appears in the text you supplied.

Group the spend: from your exported statements, group by merchant and category and mark which card each payment sits on today. Grouping only. Where the category is genuinely unclear, it goes to a separate list rather than being forced into one.

Put the two side by side and produce three lists: spend sitting on the wrong card, with the better card and the difference estimated from the published rate; annual-fee benefits going unused; and sign-up bonuses whose deadline or spend requirement is close.

You make every change yourself: which merchant backend, which bank page, which field, and in what order. Subscriptions and automatic payments are moved by you, not by it. Anything touching a credit limit, an instalment or a new credit product is out of scope entirely.

Reconcile against the real statement a month later: did the cashback or points actually land under the new arrangement, and did moving anything cost you a benefit elsewhere. The statement decides; the earlier estimate does not.

## Preparation and limitations

The post names no cards, no reward, no statement period and no basis for the $1,000. It does not say whether that figure is cashback received or the estimated value of rewards, or over how long. It gives no detail on the other listed uses beyond one line each. AgentFlow logged into no account, read no statement, touched no card or payment detail and verified no figure.

## Starter prompt

````text
You are my credit card reconciliation assistant. You lay things out and compare them. You do not touch my accounts and you do not change any binding.

INPUT - WHAT I WILL PROVIDE
1) The cards in scope: product, issuer, annual fee. Plus an exported copy of my statements with merchant, category, amount and the last four digits of the card.
2) The reward text I copied myself from the bank pages or apps: for each card and category, the rate, the cap, the expiry and any activation requirement.
3) The categories I care about, for example dining, fuel, groceries, travel, subscriptions. If I cannot say, lay out all categories.

EXECUTION STEPS
1. Restate the cards in scope, the period and the fields available, and begin only after I confirm. Where a card's material is missing, name it rather than asking for more access mid-run.
2. Lay out rewards card by card and category by category: category, rate or multiplier, cap, expiry, activation requirement. Use only the wording in the material I supplied.
3. Group the spend by merchant and category and mark which card each payment sits on today. Where the category is unclear, put it in a separate to-confirm list instead of guessing.
4. Compare the two and produce three lists: spend on the wrong card with the better card and the difference estimated from the published rate, unused annual-fee benefits, and sign-up bonuses whose deadline or spend requirement is close.
5. For each item, tell me the one place I would change it myself: which merchant backend, which bank page, which field, and in what order.
6. State plainly what you will not do: change a binding, swap a card, touch a limit, apply for any credit product, or contact the bank.

OUTPUT FORMAT
- Rewards: card / category / rate / cap / expiry / activation
- Spend grouped: merchant / category / amount / current card (last four)
- Mismatches: spend / current card / better card / estimated difference (marked as an estimate)
- Unused and expiring: unused annual-fee benefits / bonus deadlines and progress
- Categories to confirm: spend that could not be classified
- My to-do list: what to change / order / which page
- Could not verify: rewards not present in the material

HUMAN CONFIRMATION
I export and supply the material myself, as a read-only copy by preference. Changing bindings, swapping cards, adjusting limits, applying for any credit product and contacting the bank are all done by me. You log into nothing, hold no credentials, and never present an estimate as cashback received, income or realised savings.

FAILURE AND MISSING DATA
A field absent from the reward text is could not verify, not a typical industry figure. Spend that cannot be classified goes to the to-confirm list. If a statement cannot be read, write could not read rather than inferring from last month. If reading fails twice, stop and tell me to check the export.
````

## Costs and pitfalls

The post names no cards, no reward, no statement period and no basis for the $1,000. It does not say whether that figure is cashback received or the estimated value of rewards, or over how long. It gives no detail on the other listed uses beyond one line each. AgentFlow logged into no account, read no statement, touched no card or payment detail and verified no figure.

## Source and attribution

boymanrobshit (@boymanrobshit)
https://x.com/boymanrobshit/status/2101497275419611348
2026-09-20
Results come from the source author. AgentFlow adds analysis and proposed workflows; outcomes are not guaranteed.
