Money Matters

I.The Files Begin to Move

In 2026, Ashland & Pike announced that it would spend half a billion dollars building proprietary technology. Margaret Halloran, then chief legal officer of a corporation outside Cincinnati, forwarded the announcement to her team with the subject line “half a billion dollars for autocorrect.” It expressed a common view among the people being asked to pay for the first generation of tools: producing text more quickly was useful, and it was less clear why that should require a new account of what a law firm was.

Halloran’s company bought legal advice, negotiated commercial agreements, and kept an expanding collection of systems for both. Its legal memory was scattered among them, so that a contract sat in one system, an amendment in another, and the explanation for accepting an unusual term might be in a message sent by someone who had since left the company. Outside firms held parts of the record. So did business teams that did not think of their correspondence as legal information. A lawyer’s value included knowing which document mattered and whom to call when it did not explain what had happened.

The first systems could summarize the documents they were given. They were less reliable at establishing which agreement controlled, whether a party had changed its name, or whether two records described the same obligation. Those failures mattered more in practice than a well-written paragraph, because a mistaken relationship could turn a good account of the wrong contract into advice that looked dependable. The lawyer in the loop spent much of the time checking the structure beneath the answer.

The work that changed the market began with this less glamorous problem. Companies wanted records they could connect and check. They also wanted to change firms without reconstructing their legal history each time. In its 2029 renewals, Halloran’s company insisted on usable exports, stable identifiers, and records that could be connected to its own financial systems. She expected the clauses to strengthen her position in the next fee negotiation, since a firm that no longer held the files would have that much less control over whether the client could leave.

Once the record could move, a different supplier could act on it. The contract could be connected to the invoice, the shipment, the amendment, and the decision made when a similar transaction had gone wrong, until the accumulated choices of lawyers became a record that could be inspected and reused. Some of what had appeared to be individual judgment proved to be a reasonably consistent set of company preferences, applied repeatedly by people who had learned them through experience. The systems picked up those positions and the department’s bad habits with the same appetite, and a habit became a rule only when someone with the authority to promote it did so.

The systems began carrying more than a history of transactions. They carried the company’s settled positions: which risks it would accept, which departures required permission, which counterparty had already been granted an exception. The lawyers checked recommendations against that history from inside a harness of limits, records, and permissions that said what the system could do on its own, what it had to show someone, and where it had to stop. As the checks became dependable, the lawyer in the loop concentrated on the cases where the recommendation failed or the circumstances had changed.

The routine work ran over the lawyer, who watched it, and the judgment came back to her when the system reached a case it could not frame. A lawyer who once negotiated an exception on Tuesday might now spend Tuesday changing the policy under which a thousand exceptions would be handled. Framing the question, stopping the process, and signing the result stayed with a person. The company’s commercial choices still had to be made, and someone still had to answer for them, but fewer and fewer transactions required those choices to be made afresh.

No department had to close for these changes to take hold. Procurement wanted records that could move. Finance wanted them reconciled with payments. The legal team wanted fewer hours spent reconstructing an agreement. Each request was defensible on its own. Together they gave a new supplier enough of the company’s history to begin competing for the work.

II.The First Decisions

While companies were connecting their records, a separate experiment was establishing whether a machine’s answer could carry institutional authority. The National Arbitration Association began automating small commercial disputes in 2027. A customer had been charged twice, a delivery had failed, or a seller and buyer disagreed about a return, and the amounts were often too small to justify professional help, so that many of these claims had never become cases at all.

Hannah Stein, who ran the initiative, had spent a decade in a state court administrator’s office. When she moved to the association, she brought less reverence for existing procedure than familiarity with what happened to the people it could not accommodate. The association had four hundred arbitrators. The surrounding commercial platforms generated claims on a scale that made hiring a few more beside the point.

The platforms had their own reasons to help. A customer who trusted the remedy was more likely to complete the purchase, and a merchant who could offer a reliable dispute process had an advantage over one who could not. Automated arbitration could be paid for as part of a commercial service. The unfortunate person who needed it no longer financed it alone. Delgado’s free decision was possible because resolving her claim had become a small cost inside a much larger business.

That arrangement brought a conflict with it. The institution paying for a dispute system might care greatly about keeping transactions moving and less about questions that threatened the terms on which those transactions occurred. The early safeguards mattered: notice, an accessible record, a reasoned award, and a route to challenge it, which together distinguished the new service from a customer-support decision that had acquired authority only because no one answered the next message.

Stein kept a folder of thank-you notes. She also kept the reversals. In its first eighteen months, the association vacated eleven hundred awards, most because a shipping timestamp had been interpreted in the wrong time zone. The technical defect was repaired in a weekend. Recovering payments and correcting what had already happened took longer. Software could be fixed much faster than the consequences of using it.

It was nevertheless an improvement for people who previously had no economical way to obtain a decision. A professional comparing the new process with a carefully argued case could find deficiencies, while a buyer comparing it with abandoning the claim altogether often found a reason to use it again. Stein objected when I dwelt on the reversals without asking about the claims that had once gone unheard. “You can’t leave them out because there’s no letter in a file,” she said.

Making claims cheap also created work. People brought grievances they would previously have swallowed. Companies could pursue small debts that once cost more to collect than they were worth. The machinery expanded both access to a remedy and exposure to enforcement. Whether someone experienced it as liberation depended in part on which side of the unpaid balance they occupied.

In Maryland, Judge Antoine Reyes built on the arbitration example. Mary-Tech, introduced in 2030, allowed parties to submit defined categories of commercial disputes to an automated process with reasons and an appeal route. The initial boundaries were narrower than the builders thought technically necessary; they were what legislators, insurers, clerks, and contracting parties could be brought to accept.

Success enlarged the constituency for expansion. Companies routed disputes to Maryland, creating pressure on competing jurisdictions. Delaware eventually changed its rules. Legal recognition, commercial adoption, and technical improvement reinforced one another. A provider could invest more once its decisions were enforceable; a jurisdiction could authorize more once it had a record of decisions to inspect.

The bar associations challenged the arrangements in court. Their economic interest was evident, but their objection could not be disposed of by pointing to it. A license attached obligations to an identifiable professional, and if a system took over the work, someone still had to determine who carried those obligations, who could be sanctioned, and who would pay when things went wrong. The providers were often more precise about the work they could perform than the responsibility they would accept.

By the early thirties, the systems had customers who wanted them kept, institutions that would recognize their decisions, and records against which their performance could be checked. The litigation continued. Inside companies, a different process was getting underway: the next annual budget.

III.The Department Closes

The firms’ business problem became visible before their professional one had been resolved. A team assembled for a matter could be sold again only by doing more work, whereas a provider could apply a common system across many matters and charge for its use each time it ran, so the same investment kept earning without anyone billing another hour. As customers connected procurement, finance, and compliance, the same account of their obligations served all three. The legal product became part of a larger service that the client used every day.

Some firms built capable products. Reorganizing the partnership around them was harder. A tool that saved an associate a day could improve the margin on a fixed-fee matter. On an hourly matter, it could eliminate revenue. Sustained investment had to be funded by partners whose clients, earnings, and remaining careers were distributed unevenly across practices. Partners who agreed on the technology could disagree about whether their own share of the cost would ever return to them.

By 2032, Halloran’s finance team could compare the alternatives on actual company work. The providers were cheaper and, on the measures the company used, made fewer errors. The spreadsheet she remembers was persuasive partly because it arrived after years of changing what could be measured, years in which the record had grown more complete, the policies more explicit, and the company’s expectations of its lawyers more standardized than anyone had set out to make them. The experiment no longer asked whether a machine could be a good lawyer in the abstract. It asked whether retaining a particular group of lawyers improved this company’s outcomes enough to justify the cost.

The presentation included an option to keep a smaller team during a longer transition. Halloran recommended moving faster. She believed additional supervision would buy little. That was a decision, although she sometimes described the spreadsheet as having left her without one.

The transactional group closed on a Thursday in March of 2033. There were sixty-two lawyers on the floor that morning, and what Halloran remembered, when I asked, was ordering the good sandwiches and revising a memo, several times over, that the system eventually rewrote more kindly than she had managed to. She chose to read her own version aloud. A contracts lawyer named Teresa interrupted during the first paragraph: “Margaret, we’ve all read it.”

Forty-four of the sixty-two found other positions inside the company within a year, a placement figure that appeared in later accounts of the transition and that captured nothing about changes in grade, compensation, or discretion. Teresa stayed to review exceptions. She had once known the history of particular negotiations well enough that colleagues called her before deciding how to proceed. They now contacted her when a system could not finish something. She remained useful, but the institution had changed its account of what she was useful for.

The closure was reported as an employment event, and it was one, but it was also a change in where the company made decisions: some former lawyers acquired authority over policies affecting whole classes of transactions, while others inherited narrower jobs reviewing what those policies could not resolve. Teresa’s placement counted as a retained employee in either account.

Halloran remained after the transactional group closed. Her office still dealt with investigations, board obligations, contested judgments, and questions for which the company needed an identifiable person to answer. But procurement and finance increasingly worked from the same records as legal. More of the routine coordination once supplied by her department happened inside those systems.

The transition remained uneven. A company with incomplete records, unusual work, or counterparties unable to use its systems had reasons to keep the old arrangements. A firm could survive by specializing, changing its prices, or building a product clients would buy. In the middle of the decade the profession was contracting through thousands of decisions of this kind, one company and one firm at a time, until the crisis arrived and turned a gradual process into a condition of doing business.

IV.The Correction

The currency crisis of 2036 began in a financial system that had become faster, more interconnected, and less dependent on a single settlement arrangement. The dollar’s position had been weakening for years, as commodity-backed units, stablecoins, and regional payment networks each acquired a share of transactions, and businesses used automated agreements to manage their obligations across all of them. This reduced some familiar risks while connecting decisions that had previously taken place at different speeds.

The commonest account of the Correction starts with a stablecoin used to settle commodity contracts losing its peg. Stablecoins had failed before. TerraUSD had collapsed in 2022 when its design proved to have nothing behind it, and USDC had briefly lost its dollar peg the following spring when a bank holding part of its reserves failed. Each had produced rules aimed at the last failure. The coin at the center of the Correction had passed all of them. Contracts repriced. Systems hedged the new exposures, sought additional security, and exercised exit rights, and each of those actions changed the conditions faced by other systems, which responded under their own instructions. A protection in one agreement became a trigger in another. The aggregate movement exceeded the safeguards that exchanges and lenders had built around individual transactions.

The machines did not have to depart from their instructions to make the emergency worse. Many of the contracts had been designed to protect a company precisely by moving before someone else did. Once enough parties could move at once, prudent behavior at the level of the agreement became destabilizing behavior at the level of the market. The distinction had occupied economists and regulators long before the systems arrived, and automation compressed the time that had once been available to respond to it.

Lorraine Pasko, who ran payables for an importer in Savannah, watched a steel-fittings contract reprice itself four times before lunch on the crisis’s second Tuesday. Her company had automated its agreements, and the relevant clause covered currency movements, though until that week she had never once seen it invoked. At twenty to ten, she sent outside counsel three questions about holding payment, exiting the contract, and challenging the latest increase. The answer arrived the next morning, after the system had disputed the increase, obtained a reduction, and paid the corrected amount.

The company stopped sending that work to the lawyers. It still needed legal judgment, but needed it available inside the transaction before the transaction completed. Waiting for a memorandum had ceased to be a practical form of supervision.

For a few days, Pasko thought the company had come through. Then a supplier in Ohio failed to deliver. The goods were ready, but the carrier would not move them on the old terms, and the supplier was trying to arrange a payment while its lawyers worked through the agreements. Pasko’s system advised against paying a second time for a delivery already owed. The advice was defensible. It did not get the truck out of Ohio.

More suppliers stopped answering. By the end of the month, Pasko’s company had prevailed in the disputes its automated contracts had raised and had almost no one left to buy from. It closed the following year. Pasko eventually took a job helping customers implement one of the systems, and she understood better than many of the people selling it how a company could protect its contractual position while losing the business that position was supposed to protect.

The survivors were disproportionately embedded in networks that could adjust together. An automated agreement was of limited use if the counterparty could neither respond nor keep operating after it did. Large businesses pushed compatible systems onto suppliers, lenders insisted on records they could inspect, and emergency financing favored companies whose exposures could be established quickly. A technical advantage became a condition of access to the network.

The rebuilding accelerated adoption of the systems that had intensified the crisis. Individual firms had to decide how to trade the following morning. They bought machinery that had allowed competitors to respond in time, even while regulators were examining what happened when everyone responded at once. Reducing the shared risk fell to market operators and public institutions that were themselves struggling to supervise the new volume and speed of decisions.

Ashland & Pike folded during the Correction. Its technology did not save the surrounding partnership quickly enough. Other firms survived by changing their ownership, pricing, or scope. The survivors entered a market in which advice arriving after a transaction had completed was useful mainly for disputing the consequences.

V.The Rebuilding

Between 2037 and 2039, emergency arrangements became the basis of ordinary administration. Lenders continued to demand inspectable records after the immediate need for rescue financing had passed. Large companies retained the compatible systems they had required suppliers to adopt. Public agencies, faced with the same interconnected records, rebuilt around them.

The I.R.S. was dismantled in its old administrative form and rebuilt around continuous reconciliation of income, payments, and obligations. Priya Venkatesan, who led the reconstruction, became famous as the woman who had automated taxes. Congress still set the rates and the broad entitlements. The administrative system had acquired much more of the consequential work between the statute and the amount a taxpayer paid: identifying discrepancies, interpreting recurring fact patterns within delegated rules, and determining which claims needed investigation.

Its advocates emphasized the narrowing gap between what the law required and what different groups paid, while its critics asked whose interpretations had been made routine and how a taxpayer could challenge a classification used across millions of accounts. Both were describing real features of the same arrangement. Greater consistency could reduce favoritism and make an administrative choice harder to escape. The Senate hearings over the new agency were arguments about the location of discretion after the people who had exercised it case by case were gone.

Securities disclosure changed for related reasons. Records written in a common, inspectable form could be connected to transactions while those transactions were still underway. Regulators gained a view of exposures that documents had previously concealed. The firms operating the infrastructure gained it too. The practical importance of access rights increased as a commercial record became something on which other systems could act.

Europe entered the crisis with more of the necessary arrangements in place. Years of standardization, portability requirements, and common commercial infrastructure had made it easier to move records and change providers, and a series of smaller shocks had already prompted revisions to its safeguards. The advantage was uneven across countries and sectors, but substantial enough that American accounts of the crisis tended to understate it. The different outcomes suggested that the model itself had been only one component of preparedness, and that legal permissions, shared formats, and the ability to coordinate a response mattered at least as much once the model was available to everyone.

The reconstruction also increased concentration. A customer might possess the right to export its data and still struggle to leave a provider whose system coordinated its suppliers, payments, and dispute process. Portability had weakened one form of dependence without guaranteeing an easy exit from the next. Halloran’s clauses remained valuable. They could not, by themselves, reproduce the network around the files.

Inside companies, authority was redistributed along with the work. At Halloran’s former employer, Devin Cho now holds the title Chief Intelligence Officer. He is thirty-four, works in a part of the building that once housed the chief legal officer, and is responsible for setting limits, assigning authority, and identifying the decisions that require someone outside the ordinary process to become involved.

During our conversation, a notification arrived about a supplier changing a settlement arrangement. Cho inspected it and typed four characters. “It means I saw it,” he said. The change was already effective. He could revise the limit governing the next transaction or refer the relationship for review. His acknowledgment established who had received notice.

The notice belonged to a larger system of supervision. Whether that amounted to control depended on his access to the underlying record, the time he had to examine it, and the authority available before the next transaction completed. A company could satisfy the requirement for a named supervisor while withholding one or more of those things.

The remaining legal work often sat at these points. Companies needed people who could set defensible limits and inspect failures. Courts needed people who could reconstruct a chain of automated decisions. Regulators needed to distinguish an explanation generated after the event from a record establishing what had governed it. These occupations employed fewer people than the work they displaced and asked for different combinations of experience.

The new training arrangements lagged behind. Firms had paid junior lawyers to do repetitive work through which they learned to recognize what could go wrong, and much of that work had since disappeared. Simulations could expose a trainee to more examples, but institutions still had to decide what experience justified giving someone authority over a live process and who would pay for that preparation. The profession had been better at pricing work out of existence than financing the expertise needed to check what replaced it.

By the end of the reconstruction, moving a legal record was easier and leaving the surrounding service could be harder. The systems had acquired suppliers, payment relationships, and recognized procedures for resolving disputes, and their influence now extended to the terms on which someone could challenge a result. That was where the next institutional dispute developed.

VI.The Cases That Came Back

As ordinary commercial claims moved out of court, a different caseload grew. Cheap enforcement made small debts worth pursuing. Connected agreements allowed one adverse decision to affect several relationships. People who had received an answer began seeking something the ordinary process was not authorized to give: time, an exception, or relief from the consequences of an otherwise valid claim.

The courts hearing criminal cases and catastrophic injuries had retained human judges throughout the transition. Their authority rested on a public decision about who could exercise those powers, and better performance on a sentencing benchmark did not, by itself, authorize a model to imprison anyone. A model could recommend leniency, and a human judge could apply a rule mechanically. Neither observation settled who should have the power to decide.

Employment brought the dispute into ordinary commercial life. Automated systems could identify patterns of discriminatory conduct and terminate access without a manager deciding that a successful salesman deserved another chance. That could be an improvement for the people whose complaints a human institution had ignored, and it could just as easily leave a dismissed employee struggling to contest the record on which the system had acted. An account that described only the employee’s humiliation missed the people the automation might have protected. An account that described only the reduction in bias missed the possibility of a consequential error.

As the systems expanded, the country needed procedures for deciding which of these disputes could remain with the automated decision and which demanded a hearing. The route itself became a source of power, since a default classification could determine whether a person received an immediate decision or entered a queue, whether collection continued while an application was considered, and whether any assistance was available. Decisions once negotiated between professionals were embedded in the rules governing admission to the process.

Lawyers came to call the resulting arrangement the two windows. The first applied rules to a record at a price low enough to make a decision almost universal. The second could hear an argument about what should follow, including whether an established entitlement should be enforced in full. Both concerned money and justice. Reaching the second required admission to another procedure.

Judge Carol Lindqvist’s hardship docket was one of the places those rules became visible. A man I will call Earl came before her with a medical debt of sixty-one thousand dollars. The system had already removed two duplicate charges. He did not dispute the remainder. He wanted the court to change how it could be collected.

Earl was seventy-three and had brought a manila envelope containing a pharmacy receipt and a photograph of a boat. The photograph was relevant because the hospital thought the boat could be sold. He still used it with his grandson. The hearing concerned the boat’s value, his expenses, and the difference between the monthly payment on his application and what he could afford after buying his prescriptions. Lindqvist set collection at eleven dollars a month, subject to review if his circumstances changed. The debt remained owed. At that rate it would never be paid.

The order redistributed a loss. The hospital would receive less, and the effects would fall somewhere in its budget, its financing, or its prices. A system could calculate several possible arrangements, but the judge had the authority to choose one of them and make the parties accept it. The claim that mercy could not be computed was an inadequate explanation of her role. The harder question was who could impose the cost of granting it.

Getting to that point had already cost Earl money. His daughter helped with the filing fee. The hearing was postponed once. Collection continued during the wait because an application for hardship did not itself produce a stay. People who failed to reach a judge could appear in the platform’s records as ordinary completed collections, and their absence from the docket was easy to mistake for the absence of a problem.

The collection platform came from a vendor Halloran’s former company had been among the first to adopt for its own receivables. As one of its largest early customers, the company had a say in the standard options, and she had helped approve the one under which payments continued while a hardship application was pending. The concern had been that filing would become an inexpensive way to delay collection. Hospitals could buy additional review; many did not. Changing the default required revised terms, staff, and an agreement about who would pay. The software could already pause the withdrawals.

Other cases reached Lindqvist through a different route. A contractor I will call Pia lost work after an incorrect delivery record triggered defaults across nine connected agreements. Each system had followed its instructions. Correcting the first record did not restore every commercial relationship the subsequent decisions had ended. Some customers had already hired someone else.

Each process in Pia’s case offered an appeal. None could, by itself, repair the sequence. Correcting the delivery record did not compel a former customer to cancel its replacement contract and hire her again. A successful challenge needed to reach every system that had relied on the error, and even then some of the commercial damage would remain.

The expanding human docket became a problem for the budgets that were supposed to support it. Routine cases had become cheaper while the remaining ones required hearings, assistance, and reconstruction of events across several services. The savings and the costs appeared in different institutions, so that a hospital could reduce its collection expenses while a county found that it needed more clerks. No automatic transfer moved the first institution’s savings into the second one’s budget.

VII.Every Third Tuesday

Merritt County District Court, in Tolliver, Montana, still hears ordinary commercial disputes in person every third Tuesday. It is one of a small number of courts where parties can choose that route without first establishing that they qualify for an exception. Montana advertises the arrangement. Some visitors attend for the spectacle of watching people do work that, elsewhere, happens before anyone has time to observe it.

Judge Eleanor Brandt is seventy-one. On the day I visited, she heard a disagreement between a rancher and a contractor about a barn. The outstanding invoice was not seriously disputed, but the rancher objected to the removal of an old cupola his father had built, having agreed to a replacement roof without understanding that the cupola would come off with it.

The eventual arrangement required the rancher to pay the invoice and the additional materials needed to restore the cupola. The contractor agreed to supply the extra labor. They had spent much of the day arriving at it. In the hallway, the contractor told me he would rather have used the machine, since he had another job to finish and would now have to arrange a return visit. The rancher had proposed the court, and he had agreed, partly because the man was a neighbor of a customer.

The result was an accommodation neither had written into the original agreement. It had a cost, some of which the contractor would bear. Brandt thought the arrangement might leave the two men better able to live with each other, though she did not pretend to know that yet. When I asked whether it had justified the day, she suggested asking both men after the work was finished.

The court’s significance was broader than the quality of that particular compromise. It maintained a way to have one. The county paid a judge, employed a clerk, and kept a procedure available to parties who wanted someone to consider more than the terms they had agreed to. That capacity could be lost through a budget decision as readily as through a declaration that human judgment had become obsolete.

Rosa Delgado had no interest in giving up the service on her phone. When I described Pia’s case, she wanted a correction process that could repair the damage, and then she asked what sort of contracting work Pia did and whether she needed customers. The possibility of failure did not make Delgado want the old courts back. It made her want the new remedy to work when the consequences were larger than a disputed delivery charge.

The legal services people had acquired were useful enough that most would be reluctant to surrender them, and the remaining disputes concerned their terms: which decisions could be delegated, who carried responsibility, and what would be available when someone needed a process to stop. Preserving a human hearing meant paying for the people who made it possible to reach one.

By the end of my visit, Brandt had turned the conversation from the future of law to the clerk’s position. The county was considering leaving it vacant after a retirement, and while the court would remain open, people would have to complete the forms themselves or find help elsewhere. Brandt wanted that distinction understood. The hearing I had watched would still have been possible. She was less sure the two men would have found their way into it.

Afterword: Eight Conditions

This imagined account looks backward from 2040. From 2026, the arrangements it describes remain contingent. Better models alone would not produce them. The story depends on changes in business incentives, institutional authority, infrastructure, and public expectations, and on an unusually compressed period of crisis and adoption.

The eight conditions below are the assumptions I would watch. They are questions about whether this future could emerge and about the form it would take if it did.

These conditions make the account contestable. They also locate the practical decisions within it. The future of commercial law would depend on what systems can do, what institutions allow them to do, and what remains available to a person who needs the process to stop.