Annette Garcia came to accounting by first discovering where she did not belong. She entered the University of Florida as a pre-med student, took one look at the likely destination and decided a hospital was not for her. Accounting offered a less cinematic but more expansive proposition: learn to read the numbers and an entire business becomes legible. Financial statements could be maps. The career paths branching away from them seemed plentiful.
The decision proved practical in precisely the interesting way. Garcia earned a master's in accounting while beginning at Kaufman Rossin, where audit gave her permission to be professionally nosy. She reviewed the financial statements of large private companies in manufacturing, distribution, real estate, healthcare, finance and produce. One week might reveal the habits of a warehouse; another, the economics of a property business. Each industry had its own processes, tolerances and judgment calls. Audit was a tour of how companies explain themselves when explanation has consequences.
A few years in, she wanted to move inside one of those machines. At OpenStore, she worked with engineers on internal tooling and reporting, pulling data across software systems. The technical part of the job, rather than frightening an accountant back toward a reassuring spreadsheet, caught her attention. Accounting had rules. Software had systems. Between them sat a large, untidy translation problem.
That progression matters because Garcia did not approach AI as a tourist from the software world. She had seen the statements after the work, then the systems underneath the statements. She understood that a clean total can conceal a heroic amount of chasing, reconciling and deciding. The nuisance is not always arithmetic. It is context scattered among ledgers, policies, emails and the memories of people who know why last February was unusual. Any useful agent would have to enter that thicket without flattening every exception into a generic rule.
Garcia joined Basis in 2024 as its fifth full-time employee, looking for an earlier startup, more risk and harder problems.
A job arrives before its title
Basis presented all three. The company was building AI agents for accountants, a field with plenty of unsettled territory and very little inherited etiquette. Garcia joined in 2024 as the fifth full-time employee. She liked the seriousness of the team, the appetite for hard work and what she called the chance to make order out of chaos.
The role that emerged did not fit tidily into an org chart. It took more than a year to name the team Accounting Product Operations, or APO. Garcia describes its members as translators of accounting jargon, knowledge and workflows. The product group decides what Basis should build and how an accountant should encounter it. Engineers construct the infrastructure. APO supplies the professional instincts: where evidence should appear, what a familiar workflow feels like, how an instruction becomes judgment and where a human should review the work.
The resulting job is part agent manager, part context engineer, part intelligence architect and part quality assurance engineer. Garcia's compact description is also revealing: APO resembles an engineering team, except its code is written in English rather than Python. The language has to be self-contained. A reader with no prior context, whether human or machine, should be able to understand it without returning with a fistful of questions.
It is a demanding invitation to accountants joining the team. Garcia says they are being asked to pull apart the career path they expected and enter work that is still being defined. Their accounting knowledge remains essential, but it is no longer the whole job. They must explain tacit habits, examine outputs, debate product choices and learn how agents fail. The reward is proximity to the frontier; the price is surrendering the comfort of an established ladder. Garcia, who joined an early startup because she wanted more risk, is hardly selling a quiet life.
“Writing is number one.”Annette Garcia on the essential skill for her team
This makes clear writing less a corporate virtue than an operational control. An ambiguous sentence can become an ambiguous action. A missing condition can travel all the way into a workbook. Garcia's team writes for an unusual colleague: capable, literal in surprising places and unable to rely on the office folklore that helps humans fill gaps.
The answer is only the beginning
The popular account of AI prizes the answer. Professional accounting prizes the path. Accountants want consistency, a trail they can audit, terminology they recognize and some confidence that next month's work will follow the preferences established this month. A general chatbot might produce a lease amortization schedule twice and return two very different artifacts. Novelty is delightful in a cocktail and expensive in a close process.
Garcia's standard is more exacting. The system should feel like another staff accountant on the team, already acquainted with the firm's way of working, rather than a contractor hearing the assignment for the first and last time. That framing preserves the human relationship to the output. The agent works. The accountant reviews. Professional responsibility does not evaporate because a machine completed the first pass.
“It is not just whether an AI model can produce an answer. It is whether an accountant can understand how it got there.”Garcia on the trust problem
There is a consequential idea inside that conservative demand: AI could make more economic life worth accounting for. Garcia explains it with office furniture. An accountant might capitalize a fixed asset only if it costs at least $5,000. Chairs, tables and computers below that figure may not justify the hours required to reconcile, track and depreciate them. The threshold is partly policy and partly a rationing device for scarce attention.
If an agent can examine the transaction, apply the policy, keep the record and follow the same instructions month after month, the calculation changes. Perhaps $1,000 becomes practical. Judgment remains; drudgery loses its veto. In Garcia's view, the promise is not simply doing today's accounting faster. It is doing more accounting, more deeply, while giving people room to consider larger questions.
A tolerance for being wrong
The personality required for this work is not the one lazily assigned to accountants. Garcia asks for agency, curiosity and comfort with beginning at a very low point on a steep learning curve. There is no manual because the profession itself is still discovering how to supervise agents. Intuition arrives by doing the work, observing failure and revising the instruction.
A colleague, Tony Greenfield, once recalled Garcia asking him in an interview why debits and credits are the way they are. It is a splendidly inconvenient question. Working systems often depend on conventions that everyone has agreed to stop interrogating. Garcia appears unable to grant them permanent immunity. At a company trying to teach the conventions to machines, that habit becomes useful.
She is equally direct about failure. Sometimes there is no clear right answer, so a person proposes what seems right, fails, fails again and eventually wins. Patience is not passivity in this telling. It is the stamina to keep hashing out a problem with other people until the group has found something truer. The process sounds a little like audit and a little like engineering, which is exactly the territory Garcia occupies.
The phrase Basis uses for that habit is “seek truth together.” In Garcia's account, the together is doing real work. Product needs the accountant's empathy for the user. Engineering needs the domain logic that cannot be guessed from a tidy interface. Accountants need the engineers' understanding of what the system can repeat, remember and test. The collaboration is not a ceremonial meeting between departments. It is the workbench on which the agent is assembled.
New muscles for tax and audit
Basis began with client accounting services, the recurring core of bookkeeping and close work. Garcia's next challenge is to help extend the platform into tax and audit. Each field brings a different species of difficulty. Tax has an enormous regulatory surface that keeps changing. Audit depends on professional judgment: what to test, how much evidence is enough and what supports a conclusion.
Those are not merely larger versions of the same workflow. They require new muscles from the team and finer decisions about when the agent should proceed, ask or stop. The ambition is broad, but Garcia's method remains grounded in the habits she learned earlier: inspect how the work is actually done, understand why the practices differ and write the logic clearly enough to withstand review.
Her career has the satisfying shape of a ledger whose entries only balance in retrospect. Pre-med clarified what she did not want. Accounting made business legible. Audit exposed variation. OpenStore revealed the pleasure of technical systems. Basis gathered those experiences into a job that had to be invented around them. The path was not a straight line; straight lines are generally drawn after the journey, by people with excellent rulers and suspicious memories.
Garcia's work is ultimately less about making a machine resemble an accountant than deciding which parts of accounting deserve to be taught with care. Consistency. Evidence. Language. The humility of review. The willingness to ask why the rule exists before encoding it. AI may be the new colleague in the room, but the standard is recognizably human: do the work in a way another person can trust.