A Century in a Decade: Diocesan Finance and the Age of AI and Robotics

Four months after Pope Leo XIV named this "the age of AI and robotics" in Magnifica Humanitas, the people who keep the Church's books in the United States gathered at the Omni Louisville Hotel for the Diocesan Fiscal Management Conference. Matthew Harvey Sanders was asked to speak to the part the encyclical leaves to them: what this moment asks of the people who keep the books. His full remarks are below.

I. A Century in a Decade

Last summer Demis Hassabis, the Nobel laureate who co-founded Google DeepMind, told The Guardian that what AI is bringing is "going to be 10 times bigger than the Industrial Revolution, and maybe 10 times faster." Pope Leo, in Magnifica Humanitas, calls this moment "the age of AI and robotics," and I'll use his phrase too.

I begin there because this room, more than almost any room in the Church, can hear what that sentence weighs. The last industrial revolution is the reason most of you have a job. It built the cities. The cities filled with immigrants. The immigrants built the parishes and the schools you now administer, many of them in the great wave of building between about 1860 and 1930. And it made your profession what it is. Bookkeeping is as old as Babylon, but the modern profession, chartered, examined and independent, is a child of that century: the accountants of Edinburgh in 1854, England and Wales in 1880, the association that became the AICPA in 1887, all formed because railways and joint-stock companies had made money move faster than any owner could count it, and someone had to stand between the money and the men who wanted it and say: this is true, and this is not.

All of that took about a century.

Now make it ten times larger, and compress the century into a decade.

That's my subject for the next forty minutes. The last revolution built the parishes, the schools and the profession in this room. This one is going to test all three, and whether it strengthens them or hollows them out depends on decisions your bishops will make on the counsel of people with your job titles.

The test has already begun, and its first examiners are the firms that audit you. On 7 April this year EY announced agentic AI across its entire assurance practice: 130,000 professionals, 160,000 engagements, on a platform that processes, in its own words, "more than 1.4 trillion lines of journal entry data per year." On 5 May, Anthropic published ready-made agents for finance, and three of them are named for your work: a general ledger reconciler, a month-end closer, and a statement auditor.

So the firms at the top of your profession, the firms your own auditors measure themselves against, now have tools that do the work. Most chanceries are still deciding what to make of them. That decision is the subject of this talk.

Before I founded Longbeard, I worked for the Moderator of the Curia in the Archdiocese of Toronto, so I've seen from the inside some of what lands on a chancery desk. Today my company builds this kind of technology for the Church and answers for it when it fails, so this is a practitioner's account, not a spectator's, and the account ends with more reasons for hope than for fear. Eight words, then your desk, then the horizons, then three things to do this year. The rest of the hour is yours, and the harder your questions, the better. The words first.

II. Eight Words, Plainly Defined

Eight words, one plain definition each.

One. Model. The trained system itself. Two kinds matter to you: proprietary models, often called closed models, which you lease by subscription, Claude, GPT, Gemini, Grok; and open-weight models, which you download and run on your own hardware. They come in every size: from China, giants like Kimi K3 and DeepSeek that need a rack of servers, and compact ones like Alibaba's Qwen 27B that run on a single desktop; and increasingly from Nvidia and other Western firms.

Two. The context window. Think of the model's training as everything it learnt at school, and the context window as what's on its desk right now: its short-term memory, its RAM. Whatever is on the desk it reads exactly; whatever isn't, it recalls from school, with all the risk that implies. And the desk is now large enough for your policy manual, your chart of accounts and a decade of annual reports at once.

Three. Hallucination. A fluent, confident answer that happens to be false: the model recalling from school instead of reading from the desk. Hallucination isn't a reason to avoid the technology. It's the risk the next word exists to reduce.

Four. Retrieval. Putting your documents on the desk and making the model answer from them, and cite them, rather than from memory. If you remember one control from this talk, remember this one.

Five. Agent. A model that finishes a task instead of answering a question: you hand it the job, it works out the steps, uses whatever software it needs, and reports back.

Six. Harness. The software around the model that lets it act and remember: open your files, run your software, keep notes from one session to the next. Anthropic's is Claude Code and OpenAI's is Codex, and you'll most likely meet them inside the Claude and ChatGPT desktop apps; and there are open harnesses you can run inside your own building, Hermes Agent and OpenClaw. The harness decides what the agent may touch, so governance lives there.

Seven. Alignment. The values and limits trained into a model by whoever built it. Every model has them, open or proprietary; somebody chose them. Anthropic publishes a constitution for Claude and says its content "directly shapes Claude's behavior." Here's the distinction that matters. With a leased model, the maker's choices are where it ends. With an open-weight model, they're where you start: through the harness you decide what it reads first, what it must check its answers against, and what it may never do, and with real effort you can retrain it.

Eight. Sovereign, or local, AI. Models and agents that run on equipment you own, inside your own building, aligned as I've just described, with your data never leaving the premises.

Put the eight together and they're the plan. An open-weight model, in a self-hosted harness, grounded by retrieval on your own documents, aligned by you, doing the work of an agent, is sovereign AI.

III. What the Machine Takes, and What It Leaves You

Now let me show you what those words are already doing on a desk like yours, with tools that shipped this year. The reconciler and closer I mentioned prepare the journal entries, run the checklist and produce the close report, and the statement auditor reads the result before your auditor does; the last week of the month becomes a morning of review. Claude, ChatGPT and Grok all now work inside Excel, building the model, fixing the broken formula, filling the template with this year's figures. The annual reports, the parish pastoral report, the Official Catholic Directory, the Annuario Pontificio, the Quinquennial, are filled in by the parishes section by section and reviewed from one screen; tools for this exist now, ours among them. And the payables, the budget, the minutes and the grant reports are drafted by the agent and signed by a person.

If most offices in this room still do most of that by hand, and I suspect they do, then the gap between what's possible and what's deployed is enormous, and a wide gap is an opportunity for whoever moves first.

So what are the hours for, once the machine has them? Three things.

To lead. Someone in every chancery has to own the diocese's approach to this technology. Many of you have IT departments, and you'll need them for the network, the security and the hardware. But the financial processes, the controls and the financial records belong to the finance office, so lead it, with IT beside you. And leading means advising the bishop: which systems to trust, where the data lives, what to buy and what to refuse.

To train. Your parishes will get these tools whether or not the chancery leads; the only open question is whether they learn them from you or from a salesman.

And to judge. Pope Leo XIV, in Magnifica Humanitas, warns of decisions "concerning employment, credit, access to public services or even a person's reputation," which, he writes, "risk being fully delegated to automated systems that do not know 'compassion, mercy, forgiveness, and above all, the hope that people are able to change'". In a diocese, most of those decisions cross a finance officer's desk: who is hired, who is paid, which school stays open, which parish is merged. The agent can prepare them. Your bishop makes them, on your counsel, and neither of those roles belongs to a machine.

So here's my claim. The finance officer of 2027 is worth more to a bishop than the finance officer of 2024, not less. The counting will be done. The counsel, which is what a finance officer was always for, won't be.

That's today. Now, where today is going.

IV. Three Horizons

Forecasts in this field arrive with dates attached, and the dates keep moving closer, so I'll give you three periods rather than three dates, and ask the same three questions of each: what the technology does, what it does to a diocese, and where the danger lies.

Now to 2027: Tools Ahead of Their Users

The first horizon: now to the end of 2027.

What the technology does. Agents that take a piece of work in the morning and hand it back done. Here's where things stand. The Stanford AI Index, out in April, reported that "AI agent deployment was in the single digits across nearly all business functions," and that on one benchmark, a test of real computer tasks, agent success had jumped from about 12% to about 66%. But those figures were already old when they were printed. On the benchmark's current, verified version, the best scores now reported are above 85%, past the 72% that human testers managed on the original, and KPMG's September survey of large American companies found 62% building, deploying or developing agents, up from 53% in June. Every statistic in this field, including mine, is stale when spoken; assume the curve kept bending. What hasn't changed is the gap between what the tools can do and what anyone has deployed, and that gap is where a diocese can move early, cheaply, and ahead of the people who audit it.

What it does to a diocese. The work I walked through becomes routine: the smallest diocese in this room gets controller-grade capability for a subscription, and the largest gets its people back for analysis, planning and parish support.

The risk. Three, and they're already on your desk.

Fraud first, because it costs money this year. The FBI's Internet Crime Report for 2025 tracked artificial intelligence as a category for the first time: 22,364 complaints, and losses of about $893 million. On your desk that looks like a voice on the phone that sounds exactly like the bishop asking for an urgent wire, or an email from a contractor you've paid for years saying the bank account has changed. Five years ago each of those took a skilled criminal; now they take a laptop and an afternoon. The defence hasn't changed, and it's cheap: dual authorisation on every payment, and a call back on a number you already hold for any change of banking details or any urgent request, whoever it appears to come from. Nearly everything on that FBI list depends on one person acting alone and quickly. Dual authorisation takes away alone. The call back takes away quickly.

Error second. Hallucination is dangerous in finance precisely because a figure the agent invented looks exactly like a figure it found. And it happens to the best. In May, EY Canada withdrew a published report after researchers found that more than 70% of its citations were broken or led to sources that never existed. In June, KPMG took down a report on agentic AI after only 5 of its 45 citations were found to point correctly to their sources. Nobody checked the sources before those reports went out. The defence is the fourth word: an agent in your office answers only from your documents, shows where each number came from, and nothing it produces gets posted until a named person has checked the source and signed.

Governance third. Your staff are already using consumer chatbots, and some are pasting parish data into them. Many dioceses have no written policy, and those that do have often reached for a ban. I understand the instinct, but a ban can't be enforced, it drives the use onto personal phones where you can't see it, and it forbids the very tools your auditor is now using on your books. A policy that works answers three questions instead: who may sign what an agent produces, where diocesan data may live, and what must never leave the building. That policy will still make sense in two years. The ban won't survive Christmas.

2028 to 2030: Whose Machine, Whose Values

The second horizon: 2028 to 2030.

What the technology does. This is the period in which the AI companies themselves expect AGI, artificial general intelligence. You'll hear that term constantly from now on, so here's what it means in plain terms: a machine that can take on any desk job a competent professional does. Dario Amodei of Anthropic told Davos last year that by 2026 or 2027 "we will have AI systems that are broadly better than almost all humans at almost all things." Hassabis, the more cautious of the two, gave even odds to it within the decade. Split the difference and it lands here.

What it does to a diocese. A department's work runs under one person's signature. Parish finance is handled centrally by agents with the pastor's sign-off. For whoever owns the tools, the cost of administration collapses, money moves from overhead to mission, and transparency becomes so cheap that there's no excuse left for its absence.

The risk. This is the one to underline, because my second recommendation rests on it.

Dependency. Picture the diocese of 2029 that has done everything I've described, all on one company's cloud. Its parish registers, its payroll, its donor records, its reasoning: all on premises it doesn't own, under terms it didn't write. Now the price doubles. Or a clause changes about what the model may be used for. Or the values trained into the model shift with the politics of the year, and the chancery's tools start quietly declining certain questions, or answering them a certain way. None of that requires bad faith. It only requires that the decisions are made by a handful of firms you'll never sit across a table from. Pope Leo has been direct about this. In Magnifica Humanitas he writes that "control over platforms, infrastructure, data and computing power does not rest with States, but with major economic and technological actors," and that when such power "is concentrated in the hands of a few, it tends to become opaque and evade public oversight," giving rise, in his words, to "new dependencies." To Catholic legislators on 21 August, at a private audience I attended, he warned against innovation becoming "another vehicle for ideological or economic colonialism."

2031 to 2035: The Parish Is the Asset

The third horizon: 2031 to 2035. The first two change how the chancery works; this one changes how your people live, and so what they'll ask of their parish.

What the technology does. Agents in every office and robots in every warehouse together loosen the tie between a wage and a livelihood for a great many people. The people making it say so themselves, in their own words. Elon Musk, last November, at an investment forum in Washington: "My prediction is that work will be optional." A few weeks earlier, on his own platform: "AI and robots will replace all jobs." Amodei, in January, naming the danger in that: "The thing to worry about is a level of wealth concentration that will break society."

Put those two men together and you've got the shape of the decade. Turbulence first: some people will lose work before the gains arrive. Then wealth on a scale nobody has seen, which will either be shared or, in Amodei's word, break society. So I'm not going to tell you that wages fall. I'm telling you what the people building this technology are telling us: that within ten years most people will have more time than any generation before them and, once the sharing is worked out, more means. What they won't automatically have is a place to bring either. And gathering people is the oldest thing the Church knows how to do: parish, guild, confraternity, hospital, school. She has never once asked whether a person was employed before giving him a place and a task.

What it does to a diocese, and here's the fiscal point. People with time and means give to institutions they trust and that visibly add value to their lives. The parish can be that institution: spiritual life first, always; then community; then help with real needs; then the one room in the neighbourhood where people talk honestly about their future. A parish that does that will be supported, generously. A parish that's only a place people check into on Sunday won't be, and no redistribution scheme on earth will change that.

The risk, and here I speak with respect. Dioceses across this continent are consolidating parishes now, for reasons everyone here understands, and nobody closes a church lightly. But this horizon adds one question to the ones you already ask. Is the parish being consolidated in 2026 the one its neighbourhood will need in 2034? And could the savings from the automation I've described keep it open, and turn it into the centre of community life it's about to be asked to be? That's a pastoral question, and always will be. But it's a fiscal question too, because the answer turns on what administration costs and where the savings go, and that half of it belongs to this room.

And the parish is an asset in a second, more literal sense. Look at the roofs a diocese already owns: church, school, hall, rectory. Put solar panels on them, and they can power much of the parish and the kind of machine I'll describe shortly. Networks already pay owners for the hours their machines sit idle, serving other people's AI requests; they're early, each takes only certain machines, and the returns are still modest. In most states, whatever power is left over can be sold back to the grid, at rates that vary from state to state. In June, Pope Leo set up a foundation to build a solar plant meant to make Vatican City entirely self-sufficient in energy. A parish that makes its own power, runs its own AI and earns from both is subsidiarity you can see from the street.

So, to close the horizons: on the most cautious reading of these figures you should act this year; on the builders' own reading you can't afford not to; and on either, the parish is the asset to protect. Before the plan, a reading of the room.

Poll: Where the Room Stands

The first question is on your screens. Which statement is closest to your chancery today?

A. An AI tool or agent touches our ledger, our reconciliations or our reporting every week.

B. Staff use AI on their own; we have no written policy and no governance.

C. We have a written AI policy and little actual use.

D. We have neither a policy nor meaningful use.

E. We have prohibited it.

[Read the result aloud, then continue.]

V. What a Diocese Can Do This Year

So what does a diocese actually do? Three things, in rising order of ambition, and each with a price a finance council can weigh. First, lease the best of what's available today. Second, own the tools you rely on. Third, open the books. I'll take them in order.

1. Lease the Best, and Learn the Discipline

The first thing: lease the best, and learn the discipline. The best models available, what the industry calls the frontier, are today the leased ones: Claude, GPT, Gemini, Grok, used through a harness on a subscription per seat. Start there because it's the cheapest and fastest way in, and because it teaches your office the discipline it'll need before it owns anything: how to tell an agent what to do, how to check what it did, and how to keep a person on the signature. Your data goes to their premises, so start with work that isn't sensitive. Here's the first quarter.

Pick one process and one owner. Not the whole office; one process: the month-end close, the parish assessment calculation, the audit request list. And the owner isn't an engineer. It's your best controller, because the whole job at this stage is writing down how the process is done, and your best controller already knows it by heart.

Then write it down. This is the step everyone skips and the one that decides everything. Put the procedure your controller carries in her head on paper as if for a new hire on her first morning: the steps in order, what a correct result looks like, what to do when something doesn't match. Then gather the documents you'd hand that new hire: the close checklist, the chart of accounts, your diocesan policies, the USCCB's Diocesan Financial Management guide. That procedure and those documents are what the technology calls a specification and its context. You've had both for years. What's new is a reader that can work through them at machine speed.

Then run it, grounded in your documents and not the internet. That's the fourth word. And one tip before you press go, because for many of your staff this will be their first real encounter with an agent, and first impressions decide whether they come back. Use the most capable model your subscription offers, not the fast default: on Claude that's Fable 5.1; on ChatGPT it's GPT-6 Astra, released this month and still reaching some plans, so pick the strongest your plan shows. And turn the reasoning or thinking setting up to the highest level available. Yes, it's slower and it costs more of your monthly allowance; it also checks its own work, makes fewer mistakes, and shows your controller what these tools can actually do rather than what the cheap setting does. And write two demands into the instruction itself. Show the source for every figure. And list every record you didn't process, and why. An agent that's allowed to stay silent about what it skipped will hand you a clean-looking report on half the work, and you'll never know which half.

And keep the signature human. The agent prepares the entry and the memo and shows its sources. A named person checks, posts and signs, and keeps the trace, because your auditor's agents will ask for it.

That's the first thing. It costs a subscription, it takes a quarter, and it leaves you with something no vendor can sell you: a controller who knows how to direct a machine. To keep learning after this week: for the technology itself, my own team follows IBM Technology and Matthew Berman on YouTube; for what it means in your profession, the AICPA's Journal of Accountancy, in print and on its podcast, has covered AI all year, agents included, and its AI resource hub gathers the rest, all of it from a constructive angle.

My question on this first thing is simple. Who in your chancery would own that written procedure tomorrow morning? If you don't have a name, that's your first hire.

2. Own the Tools You Rely On

The second thing: own the tools you rely on.

Sovereign AI, made concrete: a model on a machine the diocese owns, in its own building, grounded in what the diocese holds, with its data never leaving the premises. Two reasons to own rather than lease.

The first is control, and you understand it already: on your own machine the terms, the price, the permitted uses and the data are yours.

The second is alignment, and it matters more to me than the first, so let me be precise about what owning adds. It's less than you might expect, because you can already steer a leased model a long way yourself; but what's left is important. Leased or owned, most of the steering happens in the harness: the instructions you write, the documents you load, the skills you build, the tools you allow. You can do all of that with Claude Code or ChatGPT today, and for the ledger, the reports and the correspondence it's alignment enough: load your statutes, norms and policies, and the harness makes the model check its answers against them. So what does owning add? Three things. Nothing sits above your instructions: a leased model carries its provider's own system prompt and usage policy over yours, which you can't remove or override. The model doesn't change under you: OpenAI retires GPT-5.5 next month, and every instruction tuned for it moves whether you like it or not, while a model on your own machine changes when you say so. And if you ever need to go deeper than the harness, an open model can be fine-tuned on your own machine, with your data never leaving it. Either way, you do the steering; owning means no one else has a hand on the wheel.

Pastoral use is different. Aligning a model to the Magisterium of the Catholic Church, so that it answers a question about marriage or the end of life the way the Church answers it, with the sources, is enormous work. It's what my company has spent years on with Magisterium AI, now the most widely used Catholic answer engine in the world, in over 190 countries and 75 languages, and no diocese should attempt it alone. Its sources keep growing: the Alexandria Digitization Hub in Rome, our partnership with the Pontifical Gregorian University, keeps adding to them, opening more of the Church's written patrimony to the faithful. The sensible path is to connect: through the open standards the industry calls connectors, your local model hands a question of faith or morals to Magisterium AI and gets back an answer grounded in the Catechism, the Code and the encyclicals, with citations. And that works with a leased agent too: Claude Code or ChatGPT can call Magisterium AI's connector just as well, which is the quickest way to give any chancery agent a reliable answer on faith and morals. Administration aligned by you; doctrine checked against the Church's own sources. Pope Leo puts the principle in one sentence: "A more moral AI is not enough if that morality is determined by a few."

This isn't a forecast. In April I spoke to the bishops of England and Wales, and afterwards one of them took me aside and told me he already runs models at home on a free program called Ollama, because the material he works with is sensitive and he didn't want it sent to a company's servers. A bishop did that on his own, with no IT department and no committee. Your task begins by finding those in your own diocese who already have.

Now the bill of materials, and here's the part that has changed most in two years. This used to take a server room and a specialist. Today it's three purchases and one curious member of staff.

Compute. A desktop machine built for this. Nvidia sells one called the DGX Spark, and others sell comparable units. It carries 128 gigabytes of memory, runs models of up to about 200 billion parameters, draws less power than a kitchen kettle, and costs about the price of a top-end laptop.

Models. Be realistic about sizes. The frontier-scale open models from China, Kimi K3 at 2.8 trillion parameters and DeepSeek V4 at 1.6 trillion, are free to download but far too large for a desktop; they need bigger machines, and I'll come back to that. The models that fit the machine I described are smaller, and for a chancery's work they're more than enough: in August Alibaba released Qwen3.8-27B under an open licence, 27 billion parameters, small enough for that desktop, and on the test of real computer tasks I described in the first horizon, Alibaba reports a score of 84.3%, within two points of the best reported on that test. The Western companies are answering: in March Nvidia launched a coalition of AI companies to build, in its words, "frontier open models." My own estimate, and it's mine: open models now handle roughly 90 to 95% of what a chancery does in a day.

Harness. Hermes Agent, from Nous Research, runs on hardware you own, remembers across sessions, and meets your staff in the messaging apps they already use. OpenClaw is the other, held since July by an independent non-profit foundation. Either can be set up badly: left open to the internet, or given more access than its job needs. So give it only the files and systems its task requires, name the person responsible for it, and keep a log of everything it does. On your own machine, you set those rules. On a vendor's, the vendor does.

Now the strategy, and here the IT departments earn their keep. Sovereign AI at the chancery first, with finance owning the processes and IT owning the machine. Then a standard stack the diocese carries to its parishes: one machine, one open model, one harness, a connection to the Church's texts, the diocesan norms, one training day. Buy the machines in bulk; that's what a diocese is for. And find your first adopters, the parish administrator who has quietly used this for a year, the priest who already runs a model at the rectory, and make them your teachers.

Now a word on why this hasn't happened yet. Nobody has ever asked most of you to sign an order for a machine like that, so I won't pretend the only thing missing is will. What's been missing is knowing it's possible, and knowing you're not alone. Dioceses around the world are meeting these questions this year, and the Church has an opening to do what she has always done with her hospitals and her universities: pool what no single diocese could afford. And the hardware comes in sizes. The Spark I described sits on a desk and suits a parish or a small chancery. Nvidia's DGX Station stands beside a desk, carries 748 gigabytes of memory and runs models of up to a trillion parameters; that's a chancery machine, or one shared by the parishes of a deanery. And the largest open models want a rack, which no chancery needs, but which several dioceses together could own. And here's my offer. Chips are scarce this year, and Nvidia is hard for any single diocese to reach. My company has relationships that reach Nvidia, through Magisterium AI and through our investors, and I'd be glad to work with any group of dioceses that wants to approach them together for pricing and access. You aren't alone in this, and you needn't negotiate it alone either. Compute bought together, standards set together, data that stays the Church's own and serves her shepherds. I'd like to know where this room stands.

Poll: What Would Stop You

The second question is on your screens. If your diocese could run its own AI, on its own machine, what would stop you?

A. Nobody on staff could run it.

B. The bishop or the finance council would not approve the spend.

C. Security, audit and insurance concerns.

D. Nothing. We would do it this year.

E. We already do.

3. Open the Books

The third thing: open the books, so the faithful can see them. It belongs on this list for a plain reason. The first two things are about what the machines can do for the chancery; the third is about what the chancery does with what it now has. Once the reporting is automated and every figure carries its source, publishing costs nothing. The only remaining cost is the will to be seen.

The resources a strained diocese needs exist among its own people, in money and in talent, and they flow to institutions people trust and understand. Voice of the Faithful reviewed all 176 geographical dioceses in the United States in 2025 and reported in May that 112 of them, 64%, post current audited financial statements online, which leaves one diocese in three that doesn't show its people its current audited books. Where the faithful can see what the diocese does for their parish and their children, support follows, and in every diocese there are accountants, controllers and executives in the pews who'd give their time and skill if they could see the books and be asked.

Here's what that looks like in practice, because my company has built one version of it, and you should know what exists. Catholic Intelligence brings a diocese's data, documents and insights into one dashboard for the bishop, the pastors and, where the bishop chooses, the faithful. The reporting and the financial statements are in it, but the part that matters most to a bishop isn't the reporting. The faithful in your diocese are already asking Magisterium AI their questions, about the Eucharist, marriage, annulments, the end of life, and the dashboard shows the bishop, aggregated and anonymised, what's actually on the minds of his people. And we're building it to work in the other direction: when someone has come as far as an AI can take them and is ready for a parish, we want to know enough about your parishes and ministries to send them to the right door, so they land softly. That's why we're inviting dioceses everywhere to give us that context. In the same spirit, we've put the spiritual statistics of every diocese in the world, and the financial statements of every diocese that publishes them, into Magisterium AI for anyone to ask, because transparency and accountability are how laity and clergy end up working together rather than past each other. The diocesan package has a price for dioceses that can afford it and is free for those that can't. But the public side costs nobody anything: the directory, the spiritual statistics and the published financial statements are open for anyone to query, free, on the public dashboards and in Magisterium AI, whether or not their diocese is a partner.

Catholic Intelligence is one way to open the books, not the only one. Many of the technology companies here this week offer good work in this area, and a diocese with a strong IT team can build much of it for itself. What matters is that the books become legible, to the chancery first and then to the people. A diocese that lets its people see the books gets more from them, not less. More money. More talent. More trust.

So I'll end the third thing with a question to take home. More and more of your people will never open the diocesan website; they'll ask their AI how the diocese is doing, and it will answer from what it can find on the public web. If your audited statements are posted openly and kept current, the answer comes from your own figures. If they're out of date, buried or behind a login, it comes from somewhere else, or not at all. And when the figures are that easy to reach, more of your people will ask, and more will come to you with questions, which is exactly the conversation that builds trust. So here's the question to bring to your bishop and your finance council: when a parishioner asks an AI about your diocese's finances, what will it find?

VI. The Construction Site

That's the argument. Let me leave you with what to do, and then with why.

For your diocese, three decisions this year, each one line on a finance council agenda. Name the owner: one person in the chancery who answers for how the diocese uses this technology, with your IT director beside them. Find your data: where every category physically sits, who holds it, and what it would take to bring it home. And price ownership: a quote for one machine and one curious member of staff, set beside what you spend today on seats and consultants.

Now the why.

The Church is about to meet a generation with time on its hands, means in its pocket, and nowhere to bring either. She has met that generation before, when the cities filled, and she answered with the parishes and the schools you now hold in trust. She'll be asked to answer again, ten times larger and ten times faster, and whether she can afford to will be decided in chanceries, by the people who keep the books.

You aren't alone in that. Every diocese on earth is arriving at the same questions this year: whose machines, whose values, whose data, whose books. Pope Leo chose Nehemiah as the companion of Magnifica Humanitas, from its opening pages to its closing ones: the man who "organized the work, confronted internal and external resistance and rebuilt the walls of Jerusalem with the assistance of the people, brick by brick." Our vocation, he writes, "is not to be passive spectators of social and cultural fractures, nor mere commentators on what is crumbling, but men and women prepared to enter the construction sites of history." Nehemiah wasn't a priest. He was cupbearer to the king of Persia, then governor of Judah: the man who organised the labour, faced down the opposition, and refused to draw the governor's food allowance because the burden on his people was already heavy. The Pope chose an administrator.

Every dollar these tools take out of administration is a dollar a bishop can put into a parish that stays open, and every hour they take off a pastor's desk is an hour he can give back to his people. "The quality of a civilization," Leo writes, "is measured not by the power of its means, but by the care it is able to offer". The means are about to become cheap. The care will still have to be paid for, organised and housed. That's the construction site, and it's yours.

The technology asks three things of you, and only three. The humility to be a beginner again. The will to own what you rely on. And the openness to be seen.

Your auditor already has an agent. By this time next year, so should you. And by the time the people come back to the parish with time on their hands, I hope you'll have paid for the lights.

Thank you. The floor is yours.