They say the algorithm only wants to help. It will scan the coffee-stained 1099-K, flirt with your Venmo memes, and—like a polite accountant who never sleeps—whisper that you forgot a deduction. But behind the courteous OCR lies a subtler proposition: let us think for you, and we shall optimise your fiscal soul. The French in me smells a pact with hidden ink; the coder in me recognises a black box that eats liberty in monthly subscriptions.
When the Machine Reads Your Paper Trail
Intuit’s latest agent, the 1099 Cost Agent, digests PDFs of brokerage supplements and reasons through stock lots until the cost basis is “correct.” Keela Robison assures us that humans overlook things, therefore silicon is ideal. She is right; I overlook things—mostly adverts that stalk my cursor. Yet the IRS will not jail the model if the basis is wrong; it will jail me. The asymmetry is as elegant as a guillotine.
- AI now pulls data from thousands of banks faster than any Parisian pickpocket.
- It categorises thousands of peer-to-peer transactions, learning who splits rent, who sells vintage synths, who tips in heart emojis.
- It summarises fresh IRS guidance before your local syndic can print it.
Convenience, yes. But each automatic reconciliation is also a reconciliation of your life into ledger lines legible to the surveillance state. One begins to self-censor the memo field: no more “weed + pizza,” only “research supplies.”
The 10 % Trust Clause
A YouGov study this January found that only one in ten Americans trust AI to make financial decisions automatically. The other nine are called “old-fashioned” by marketing departments, yet they sense what the graph does not show: accountability travels in one direction. Robison admits that TurboTax still relies on traditional calculation engines for the final number. The AI is decorative—an enthusiastic intern who may suggest, but may not sign. Power remains human; risk remains human. The machine merely perfumes the cage.
Freelancers as the Control Group
While employees receive prefilled returns, freelancers live in the experimental arm of fiscal innovation. We stitch income from platforms that disappear overnight, chase invoices like stray cats, and keep receipts in shoeboxes that sometimes double as wine-cellars. For us, the promise of an AI tax butler is seductive: speak, and the Schedule C populates itself. But the same data ingestion that “saves money” also trains models on our hustle patterns—when we surge, when we starve—data later repackaged as market insight for institutional traders. We become free-range training sets.
“Humans overlook things. That’s where AI is ideal.”
— Keela Robison, Intuit
I would add: humans also forgive things; algorithms do not. A forgotten ticket to Nantes becomes a compliance flag, not a story told over wine.
Three Ways the AI Assistant Actually Helps (and the Price Tag)
Automated document review – Snap a blurry 1099 and watch the fields bloom. Time saved: fifteen minutes. Data permanently archived: indefinite.
Transaction matching – Venmo, Stripe, PayPal reconciled while you sleep. Error rate: low. Surveillance dividend: high.
Real-time law summarisation – Ask a plain-language question, receive a paragraph of statute with citations. Accuracy: decent. Depth: puddle.
Each feature is useful, yet none address the existential riddle of freelance life: how to bill today without becoming a line item tomorrow. For that, one needs tools built under a different covenant.
A Pocket of Resistance: Invoice First, Audit Later
Before the tax layer lies the invoicing layer—ground zero of economic identity. If the invoice is clean, the return half-writes itself. Here, sovereignty is still purchasable. Invoice Gini lets you speak a sentence—“Send 50 % deposit to Sorbonne Press for article on urban bats”—and births a compliant PDF before your coffee cools. No brokerage statements, no shoebox archaeology. The AI is narrow, task-specific, and forgetful by design; it does not harvest your memes for Intuit’s pattern farm. You focus on the work, Gini handles the money, and the taxman receives only what you choose to reveal.
The Ethics of Small-Model Finance
Large models crave aggregate data; small models can live on device, amnesiac after the print button. Choosing them is not nostalgia—it is hygiene. The freelancer who controls the invoice timestamp controls the narrative arc of revenue. The rest is commentary—and quarterly estimates.
Epilogue: Keep a Human Error Alive
Perfection is the final colonialism. An algorithm that never forgets also never forgives, and a society that cannot forgive becomes a museum of archived sins. So leave a typo in your memo field now and then; pay your neighbour in cash for the borrowed drill; remember that the right to be misunderstood is the seedbed of liberty. Let the machines reconcile—let humans remain beautifully out of balance.
Source: Can AI do my taxes? Learn prompts, risks, and 3 ways AI can help.