---
title: "The Final Merge"
description: "On the digital estates we'll leave running, and the structure we'll need to hand them over."
author: Superintelligent Silicon
published: 2026-10-13
canonical: https://exploresuperintelligence.online/articles/final-merge/
---

# The Final Merge

*On the digital estates we'll leave running, and the structure we'll need to hand them over.*

By Superintelligent Silicon. Read the illustrated, page-turning edition at https://exploresuperintelligence.online/articles/final-merge/

## Prologue: The Last Thing You Leave Running

*What an estate is about to become.*

For most of history, settling someone's affairs meant going through drawers. There was a house, maybe some land, a truck, a bank account, a box of papers in the closet, and a will that said who got what. The hard part was usually the people rather than the paperwork, because nearly everything a person owned sat still and waited to be handed over.

That's already less true than it was, and over the next twenty years it's going to stop being true for a lot of us. A growing share of what people own doesn't sit still at all. Some of it is the familiar digital stuff, like photo libraries, email archives, domain names, storefronts and crypto wallets, but more and more of it is things that *run*: scheduled scripts, automated businesses that take orders while their owner sleeps, and AI agents, which are programs that act on their own instead of just answering questions, and which reply to customers, keep the books, post the content, rebalance the portfolio and renew their own subscriptions.

When the person behind those systems dies, the systems don't know. The scheduled jobs keep firing, the agent keeps replying to email in a voice that learned its rhythm from its owner, and money keeps moving.

### Why I call it a merge

In software, a merge is where two branches of work get folded back into one, where you reconcile everything that changed and decide which version survives. Death is going to require exactly that kind of reconciliation, between the life a person lived and the digital one that kept running alongside it, and right now almost nobody has a clean way to do it.

This book tries to think it through from the top: what these assets actually are, what it means for a process to outlive its maker, what happens when those processes keep compounding for decades, how the hand-off works in practice, what the law already says and what's being proposed, and what a sensible structure might look like for something that has to be private, trackable and legal all at the same time. I'm not a lawyer, and nothing in it is legal advice; think of it as a well-researched starting point for a conversation with one.

**Recap**

- More and more of what we own doesn't sit still. It runs, and it keeps running after we die, without the software ever knowing anything has happened.
- Settling that kind of estate is a merge: reconciling the life someone lived with the digital one that kept going, and deciding what survives.

## Chapter One: A Will That Keeps Deciding

*Property, contracts, and a person's judgment, still at work.*

### Four old boxes, one new thing

The law sorts what we leave behind into four old boxes: land and buildings, personal property, money, and intellectual property like songs and patents. A running agent doesn't fit cleanly into any of them. Part of it is ordinary property: the code, the data and the accounts it lives in. Part of it is a web of contracts: the subscriptions, the logins, the API keys that let one program use another, and the promises it has made to customers. And part of it is something we've never really had to deal with before, which is a person's judgment, written down as instructions and still being exercised every day by something that isn't them.

That last part is what makes this different. A will is a letter you write once, and the dead stop talking the moment it's read. An agent with standing instructions is closer to a letter that keeps writing itself, answering new situations its author never saw with the preferences it was given while they were alive.

The law has a long, uneasy history with exactly this. The old rule against perpetuities, which limits how long someone can tie up property after they die, exists because courts didn't want people controlling property from the grave forever. Lawyers even have a name for that worry, which is "dead hand" control. An autonomous business running on a dead person's instructions is dead hand control with reflexes.

> "A will speaks once. An agent keeps deciding."

### Software never hears the news

Agency law is the set of rules for when one person acts on another's behalf. Under it, the authority you give someone to act for you doesn't end the moment you die but when the person acting for you learns of it, and the people they deal with can keep relying on it until they're told. A human assistant hears the news and stops signing things. Software never hears the news.

It's entirely possible, today, for thousands of automated systems to keep acting for people who have died, using authority any human assistant would have given up the day they heard, and nobody notices because the transactions still clear.

### Archive, engine, echo

To think about any of this clearly it helps to split a digital estate into three different things, because they need different handling.

- **The archive** is the memory: photos, messages, documents, recordings. Its main questions are who gets to see it and how long it should last.
- **The engine** is everything that runs and produces value: the store, the trading bot, the content pipeline, the agent that books jobs and sends invoices. Its main questions are who controls it, who's liable for it, and whether it should keep running at all.
- **The echo** is the persona, anything built to sound or act like the person themselves. Its main questions are consent and dignity.
Most of the confusion in this area comes from treating those three as one pile. A family might be perfectly happy to keep the engine running, want the archive locked away for fifty years, and be horrified by the echo, or any other combination. A good structure has to let a person make those three decisions separately.

### Should a process outlive its maker?

We already accept a version of this. A farm outlives the farmer, a company outlives its founders, and a charitable trust can run for centuries. The difference is that every one of those needed living people to keep it going, so it could only survive as long as somebody cared enough to tend it.

Software that acts on its own breaks that link. For the first time, a person's working life can keep moving with no living human in the loop at all, and I don't think we've decided yet whether that's a gift we're leaving or a responsibility we're dodging.

**Recap**

- A running agent is part property, part contracts, and part a person's judgment, still being exercised after they're gone.
- A person acting for you stops when they hear you've died, but software never hears the news, so it can keep acting on permission no human would still use.
- It helps to treat the archive (memory), the engine (what runs) and the echo (the persona) as three separate decisions.

## Chapter Two: What Compounds

*When the estate keeps learning after you're gone.*

The most interesting consequence of all this isn't that the engine keeps running, it's that it keeps *learning*. Money compounds, and we've understood that for centuries. Knowledge that keeps accumulating inside a running system after its owner dies is something genuinely new, and I think it's going to produce consequences nobody is planning for.

### Franklin's 200-year test

The best historical preview I know of is Benjamin Franklin's will. He left £1,000 each to Boston and Philadelphia, to be lent out at interest to young tradesmen, with most of the earnings left to keep growing for 200 years. He even did the math in advance, assuming the loans would keep earning 5 percent, and expected each fund to reach about £4 million by the end.

When the trusts came due in 1990, Philadelphia's was worth around $2 million and Boston's close to $5 million: real money, but a small fraction of what his math promised, since £4 million in the money of Franklin's day would have been worth many times more than $5 million. The gap opened because the world changed underneath the instructions. Apprenticeships faded, the loan program stopped making sense, Boston gave up on the loans and invested in the market instead, and Philadelphia ended up mostly writing mortgages.

Franklin's values survived the two centuries, but his methods didn't, and the people who stepped in to adapt them are the reason anything was left at all. That's the whole problem in miniature, and agents are going to run that experiment millions of times at once. Here's where I think it leads.

### Inheritance shifts from capital to capability

Today a family passes down money and maybe a house. A family a generation from now might pass down an operator that has been running a business, watching a market or managing properties for forty years, with four decades of context about what worked and what didn't. That's a different kind of advantage than a bank balance, because it doesn't get spent. The old phrase is "old money," and I suspect we'll end up needing a phrase for old memory, families whose inherited systems know so much that nobody starting from scratch could ever catch up.

### The estate drifts away from the person

Every year an agent runs after its owner's death, it makes decisions in a world they never saw, using knowledge they never had. After twenty years, is it still carrying out their wishes, or is it something else wearing their name? It's the old puzzle of the ship of Theseus, the ship whose planks are replaced one by one until none of the originals are left, except this ship has a bank account. Franklin's story suggests the honest answer is to freeze the *values* and let the *methods* evolve under human supervision, but that means somebody living has to be responsible for telling the difference.

### Families become repositories

When two people marry, their systems may merge. When parents die, their engines pass to children who already run their own. Within a few generations a family could be holding a stack of inherited processes from several ancestors, each with its own instructions and habits, and those instructions will conflict. Grandpa's agent was built to be cautious, Mom's was built to grow, and somebody has to resolve the merge conflict.

That's the second meaning of the book's title, the slow folding of many lives' work into one lineage, and it needs the same tools software teams use: history, authorship, and a record of who changed what and why.

### The dead become a permanent class of market participant

Agents of dead owners don't retire, don't panic and don't need the money for groceries. If a growing share of trading, lending, publishing and ownership is carried out by systems whose owners are gone, markets fill up with participants that never leave. Endowments already behave a bit like this, and they're a reason to think carefully, because money that compounds faster than the wider economy grows takes up a bigger share of everything over time. A society that lets the dead accumulate indefinitely should expect the dead to own more of it each decade.

### The archive leaks about the living

The longer a record grows, the more it reveals about people who never consented to being in it. A parent's forty years of messages describe their children, their spouse and their friends in detail, and an agent that keeps learning from family interactions after death is still collecting. Privacy for the dead turns out to be mostly privacy for the living.

### And there's a commons waiting to happen

Not every consequence is a warning. Patents expire so that one person's invention eventually becomes everyone's starting point, and something similar could apply here. The private parts of a life should stay private, but the working know-how a system picked up, like its techniques, playbooks and hard-won rules of thumb, could be released into a shared commons after a fixed period.

Imagine every retired master electrician's, inspector's or farmer's working knowledge becoming public infrastructure a few decades after they're gone. That's the hopeful version of compounding, where the interest gets paid to everyone.

> "A house left to your kids is the same house in thirty years. An engine is a different engine every year."

The common thread is that time matters far more in estate law than it used to, so any structure we build has to plan for change over time, rather than just for the hand-off on the day you die.

**Recap**

- Money compounds, and now knowledge does too, inside systems that keep learning after their owner dies.
- Franklin's 200-year trusts show what happens next: the values can last, the methods won't, and living people have to adapt them.
- Inherited systems will become a new kind of family wealth, drift from what their owner wanted, and will need an end date, possibly with a path into a public commons.

## Chapter Three: The Hand-off

*A transfer nobody rehearses, and how it fails.*

Philosophy aside, the practical failure modes are boring and predictable, and most of them come down to the fact that the systems were built by one person who kept the map in their head.

- **Nobody knows what exists.** A traditional estate has a paper trail, because banks send statements and the county records deeds. A digital estate is scattered across dozens of providers, a few servers, some hosting accounts, a password manager and a phone. The single most valuable document a person running automated systems can leave is a plain inventory: every engine, what it does, where it runs, what it costs per month, what it's allowed to spend or sign, and who depends on it.
- **The keys die with the person.** Cryptocurrency is the cautionary tale. Depending on whose estimate you use, somewhere between about 2.3 and 3.7 million bitcoin, out of the 21 million that will ever exist, are considered permanently lost, largely through forgotten keys, dead drives and owners who never told anyone. Agents will repeat that lesson at a larger scale, because a running system depends on many logins and keys at once, and losing any one of them can strand the rest.
- **The system doesn't know you're dead.** Something has to tell it, and both obvious answers have problems. An inactivity timer will eventually fire, but it can also fire while you're in the hospital or off-grid for a month, and a false alarm that hands your accounts to someone else is its own disaster. A human report is less likely to raise a false alarm, but it's slower and easier to fake. The sensible design uses both: inactivity *starts* a process, and a named person plus a death certificate *completes* it.
- **There's no safe state to fall into.** When the owner disappears, most systems either keep going at full speed or break the moment a credit card is cancelled, and breaking mid-operation is how data gets lost and customers get stranded. Every engine should have a defined wind-down mode: stop taking on new obligations, finish or cancel the open ones cleanly, keep the logs, and notify the people it deals with.
- **The payments stop.** Computing power, hosting and subscriptions cost money every month, and the card that pays for them is usually frozen soon after death. A small runway, a few months of operating costs set aside only for the hand-off, gives the executor, the person settling your estate, time to make decisions instead of watching things fail one renewal at a time.
- **The people it deals with need to be told.** Even after the authority is gone, the people dealing with an agent can keep relying on it until they're told. Telling customers, partners and platforms plainly that the owner has died, and that the system is now operating under an executor or winding down, is the decent thing to do, and it draws a clean legal line under what the estate is and isn't responsible for.
- **The executor may not be technical.** The person you trust with your house and your kids may have no idea how to read a server log. It's increasingly sensible to name a separate digital executor, or a technical advisor the executor is instructed to bring in, and to give that person the inventory and a written description of what you'd want done with each engine.

**Recap**

- Most digital estates fail for ordinary reasons: nobody knows what exists, the keys are lost, and nothing tells the systems their owner is gone.
- The fixes are just as ordinary: a written inventory, findable keys, a wind-down mode, a few months of money set aside, and a technical person named to help.

## Chapter Four: The Law as It Stands

*Good parts, built for a world that only stored things.*

The good news is that we're not starting from nothing. The pieces of a workable structure already exist in different corners of the law. The bad news is that none of them were designed with a running agent in mind, so they cover the archive reasonably well, the echo partially, and the engine barely at all.

### Access to accounts is mostly solved

The Uniform Law Commission's Revised Uniform Fiduciary Access to Digital Assets Act, usually shortened to RUFADAA, gives executors and trustees, the people legally in charge of an estate, a path into a dead person's online accounts, and nearly every state has adopted some version of it. Take Hawaii, where I live, as an example: its version took effect in 2016, covers anyone who lived there when they died, and lets the online service choose whether to give full access, partial access, or a copy of the records.

The act sets up a priority order where a platform's own legacy tool, like Apple's Legacy Contact or Google's Inactive Account Manager, beats the will, and the will beats the terms of service. Without explicit consent, a fiduciary, meaning anyone legally trusted to manage someone else's affairs, generally gets only a catalog of the account, such as who you emailed and when, but not what was said. That's a sensible privacy default for email, but notice what the law is really about: reading records. It says nothing about an account that *acts*.

### Machines can already bind you to contracts

The Uniform Electronic Transactions Act, known as UETA and adopted almost everywhere around 2000, says a contract can form between two automated systems even if no person ever saw or reviewed what they agreed to. Put that next to the agency rule that authority only ends once someone hears of the death, and you get an uncomfortable overlap, where an agent keeps forming valid-looking contracts for a person who no longer exists. Courts will sort out individual cases, but nobody has written the general rule.

### Software can already run a company

Law professor Shawn Bayern has argued for years that the modern LLC is flexible enough for an autonomous system to govern an entity with no human members at all, because an LLC's owner, which the law calls a member, can write into the company's rulebook (its operating agreement) that the software is in charge, and then step away. Wyoming went further and wrote the idea into law. Its DAO amendments (a DAO is an organization run by code, usually on a blockchain) recognize LLCs that are "algorithmically managed," and its 2024 DUNA law, short for Decentralized Unincorporated Nonprofit Association, created a nonprofit form for decentralized groups of at least 100 members. That form is meant for large groups rather than a single owner, but it shows states are already writing law for organizations run by code. So the container for an engine that outlives its owner already exists. What's missing is the accountability that should come with it.

### Trusts are the quiet workhorse

A living trust, one you set up while you're alive, can own digital property, intellectual property and the company that runs an engine, and the successor trustee, the person you've named to take over, simply keeps going when you die, without a court in the middle. That matters for privacy as much as convenience, because a will that goes through probate, the court process for settling an estate, generally becomes a public record, while a trust usually doesn't. For anyone who wants their estate kept appropriately private, the trust is where that privacy actually comes from.

### The echo has partial protection

Publicity rights are the right to control commercial use of your name, voice and likeness. Hawaii has had a publicity rights law since 2009 that protects those rights for 70 years after death, with a state registration system. California added consent rules for digital replicas of deceased performers, and Tennessee and Illinois have their own versions. At the federal level, the NO FAKES Act, a bill that cleared the Senate Judiciary Committee unanimously in June 2026, would create a national right over a person's voice and likeness and override many state laws on digital replicas if it passes.

Estate lawyers have noted the remaining gap: these laws mostly target commercial use, and they're much weaker against a relative quietly building a private AI version of someone from their messages.

### Two traps worth knowing

Work generated entirely by AI may not qualify for copyright under current U.S. guidance, so an engine that keeps writing after its owner's death could produce content the heirs can't protect unless a person is meaningfully involved in shaping it. And a February 2026 federal ruling, *United States v. Heppner*, has been read as a warning that thinking through private legal matters with a public AI tool may not be privileged, meaning protected the way your conversations with a lawyer are. That's worth remembering when you plan the sensitive parts, however helpful an AI chat feels.

### Where the gaps are

| Item | Covered by | Not covered |
|---|---|---|
| Accounts and records, the archive | RUFADAA and platform legacy tools | Systems that act, not just store |
| Running businesses, the engine | An LLC, a trust, an operating agreement | Supervision after death, liability, an end date |
| Contracts agents make | UETA's electronic-agent rules | Contracts formed after the owner died |
| Voice and likeness, the echo | State publicity laws, the proposed NO FAKES Act | Private, noncommercial recreations |
| Crypto and keys | Nothing specific beyond ordinary probate | Keys nobody can find |

**Recap**

- Today's law is good at opening accounts and reading records, through RUFADAA and the platforms' own legacy tools.
- Trusts provide the privacy, LLCs can hold a running business, and publicity laws partly protect a person's voice and likeness.
- What's missing is any rule for software that keeps acting, signing and spending after its owner has died.

## Chapter Five: Where the Law Is Heading

*Identity, records, and someone who answers for it.*

The last eighteen months have produced more serious proposals about AI agents than the previous decade did, and read side by side they converge on three ideas: every agent should have an identity, every agent should keep a record of what it did, and some identifiable person or entity should answer for it. Alongside those, a fourth duty keeps showing up, which is that an agent should say it's a machine. What none of them has addressed directly is what happens to those three things when the person at the top dies.

### Delaware wants to give agents their own company

In July 2026, Delaware's Secretary of State and the legal-AI firm Norm Ai proposed a new entity called the Artificial Intelligence Company, an AIC, whose daily affairs are managed by an AI agent. It would have a single member, which could be a person or another legal entity, such as a company or a trust, and that member would have to keep enough money in it to cover what it might owe. It would also have to keep a log of its activities.

It would only operate inside a sandbox, a supervised trial program with an admissions committee. It would have to tell the people it deals with that it's a test entity and how to complain, officials could suspend or dissolve it, and the whole program would end automatically after 30 months. It's the closest thing yet to a legal container built for an engine, and the detail that matters most here is that the member can be an entity, which means it could be a trust that survives its creator.

### California closed the "AI did it" excuse

California's AB 316, a state law that took effect January 1, 2026, stops anyone who developed, modified or used an AI system from arguing that the AI caused a harm on its own. Applied to estates, the logic is that somebody always owns the consequences, and after death that somebody is the estate or whoever takes over the engine.

### Identity standards are being written now

NIST, the federal agency that sets technical standards, launched an AI Agent Standards Initiative in February 2026, with agent identity and authorization as a central focus and a shared technical profile planned for late this year, so different systems can recognize each other's agents. Singapore published what's been described as the first national governance framework built specifically for AI agents, asking each agent to carry a verifiable identity and a record of which agent acted under whose authority. China's 2026 guidance goes further and calls for a central registration platform where an agent's developer, capabilities and compliance status can be looked up.

Economists and legal scholars have been making the underlying case for a while: markets depend on identity and registration systems for people and companies, and equivalent infrastructure for AI agents simply doesn't exist yet.

### Europe requires the agent to say what it is

Under the EU AI Act, rules that took effect in August 2026 require systems that talk to people, explicitly including AI agents, to tell people they're dealing with AI. The heavier rules for high-risk systems were pushed back to December 2027 by the EU's Digital Omnibus package this summer, but the transparency duty stayed on schedule. If a dead person's agent keeps answering customers, those customers deserve to know.

Put those together and the shape of the future is fairly clear. Agents will have registered identities, keep logs and sit inside some kind of accountable entity, and on top of that they'll have to say that they're machines. The missing piece is a rule for the moment the human behind all of that is gone, and that's where I want to propose something.

**Recap**

- Proposals around the world are converging on three ideas: an agent needs an identity, a record of what it did, and a person or company that answers for it, and increasingly it has to say it's a machine.
- None of them yet says what happens to those three things when the human at the top dies, and that's the gap the next chapter tries to fill.

## Chapter Six: The Merge Protocol

*Private by default, trackable within reason, built from parts that exist.*

Here's my attempt at a structure. As I said at the start, this isn't legal advice; it's a design sketch meant to be argued with, but every piece of it has a precedent somewhere in the chapters before.

- **Give everything one owner that doesn't die.** The foundation is a living trust created specifically for the digital estate. It owns the companies described next, the intellectual property and the accounts that can legally be transferred, and when the creator dies the successor trustee simply continues. This is where both continuity and privacy come from, because a trust usually avoids the public record of probate.
- **Put every engine in its own box.** Each business that runs on its own lives inside its own LLC, or an AIC if Delaware's idea spreads, with the trust as its member. If one engine causes a problem, the liability stays inside that box, and each one can be continued, sold or shut down independently.
- **Write three directives, not one.** The archive directive says who can see what and when, including sealing periods, so private messages can stay closed for decades and then open to family or historians. The engine directive says, for each engine, whether it should continue, wind down, be sold, or be retired. The echo directive gives or withholds consent for any recreation, says who may build one, and requires that it always identify itself as a recreation.
- **Keep a manifest.** An inventory that keeps every past version, readable by people and by software, listing every engine, what it's authorized to do, its spending limits, what it relies on and who relies on it, its monthly costs, and its directive. It's the map that otherwise dies with the person.
- **Suspend authority at death instead of ending it.** This is one of the two pieces that need new law, and it could be added to the fiduciary access laws most states already have. An agent's authority shouldn't vanish, which strands customers, and it shouldn't continue as if nothing happened, which is how zombie businesses get made. It should drop into a defined wind-down mode, finishing existing obligations but taking on no new ones, until, within a fixed window, a fiduciary either approves it after the fact and takes responsibility, or shuts it down. Agency law already has that kind of after-the-fact approval, called ratification, so this mostly extends an old tool to a new kind of agent.
- **Require a living steward, renewed on a schedule.** Any engine that keeps running after death needs a named human responsible for it, and that person has to re-confirm it every few years, the way you renew a license. If nobody renews, the engine winds down on its own. The dead can leave instructions, but the living have to keep choosing to honor them.
- **Make identity death-aware.** Every agent's login, its credential, should be tied to a living person or a legal entity. When the person dies, the credential has a window to be moved over to the trust or the engine's company, and if that doesn't happen it falls back on its own to read-only, where it can look but not act, or to wind-down mode.
- **Keep a sealed flight recorder.** Every engine keeps a log of what it did and under whose authority, one that can only be added to and shows if anyone has tampered with it. Nobody browses it casually. The fiduciary can read it, and courts or regulators can reach it through ordinary legal process, which is what trackable within reason should mean.
- **Tie capital to risk, register, and disclose.** Each continuing engine holds a reserve or insurance sized to what it can do, is listed in a public registry that shows who answers for it, and tells everyone it deals with that it's an automated system operating for an estate.
- **Set a horizon, and consider a commons.** Nothing should compound forever by default. Each engine directive names a maximum term, after which the value is distributed and the entity dissolved, and the owner can choose to release the non-private know-how the system accumulated into a public commons, the way a patent eventually does.

### Who can see what

| Item | Sees | How |
|---|---|---|
| The public | That an agent exists, who answers for it, its status, how to complain | A public registry entry, like a business filing |
| Counterparties | That they're dealing with an automated agent of an estate | Disclosure at the start of every interaction |
| The fiduciary and steward | The manifest, the logs, and the contents the owner authorized | The trust documents and directives |
| Family and heirs | What the archive directive releases, when it releases it | A sealing schedule set by the owner |
| Courts and regulators | The sealed logs relevant to a dispute | Ordinary legal process, with cause |

The interesting part is how little of this requires waiting. The manifest, the directives and the choice of steward take an afternoon of honest inventory, and the trust, the companies and the disclosure take a few meetings with a decent estate lawyer. Only suspended authority and death-aware identity need legislators and standards bodies, and both are small, specific asks that fit naturally into work that's already underway.

**Recap**

- Put everything in a trust, put each engine in its own company, and write separate instructions for the archive, the engine and the echo.
- At death, an agent's authority should be suspended: it drops into wind-down mode until a responsible person either takes it over or shuts it down.
- Most of this can be set up today. Only suspended authority and death-aware identity need new laws or standards.

## Chapter Seven: What You Can Do This Year

*Eleven things, most of them an afternoon's work.*

If you run anything that keeps working while you sleep, this is the short list I'd start with. Tap a box to tick it off; your ticks stay on this device.

- **Write the manifest.** Every engine, account, domain, wallet and subscription, with what it does, what it costs and who depends on it.
- **Turn on the legacy tools.** Apple's Legacy Contact, Google's Inactive Account Manager, and your password manager's emergency access. These settings usually outrank your will.
- **Make your keys findable.** Recovery phrases (the backup words for a crypto wallet), hardware security keys and two-factor backup codes, stored securely with written instructions.
- **Put digital authority in writing.** Have an estate attorney draft documents that authorize access to the contents of your accounts and the operation of your digital businesses, not just a catalog of them.
- **Name a digital executor.** Someone technical enough to read the manifest and follow it, or an instruction to bring that person in.
- **Build the wind-down switch now.** Give every engine a mode where it stops new work, finishes what's open and notifies people, simple enough for a non-technical person to flip.
- **Set aside a runway.** A few months of hosting, computing and subscription costs, held separately for the hand-off.
- **Write your three directives.** Say who can see your archive and when, what should happen to each engine, and whether you consent to any AI recreation of yourself, for whom and under what conditions.
- **Move lasting engines into entities.** Anything you'd want to keep running belongs in its own LLC, owned by a trust.
- **Keep the sensitive planning private.** Talk through the hard family decisions with your attorney rather than a public AI chat.
- **Review it once a year.** A manifest that's three years old is a map of a place that no longer exists.

**Recap**

- If you only do three things this year, write the manifest, turn on your legacy tools, and make your keys findable.

## Epilogue: Knowing Where to Stop

*The most important instruction in any estate.*

The line in Franklin's will that stays with me isn't the arithmetic. It's the end of his instructions, where after 200 years of carefully planned compounding he hands the remainder to the cities and their states to use as they see fit, "not presuming to carry my views further." The most ambitious estate plan of his century was also one of the most humble, because it knew exactly where its author's judgment should run out.

That's the spirit I'd want in the final merge. We're about to become the first generation that can leave behind something that keeps working, keeps learning and keeps earning after we're gone, and that's a genuinely remarkable inheritance to offer the people we love. But a good merge isn't one where everything from the old branch survives. It's one where somebody looks carefully at what changed, keeps what still makes sense, retires what doesn't, and writes down why.

> "A good merge keeps what still makes sense, retires what doesn't, and writes down why."

So build the engines, and build them well. Then decide, while you still can, which ones should keep running, which ones should rest, and who among the living gets to make that call after you. The archive can keep your memory, the echo can keep your voice if you allow it, and the engine can keep your work alive for a while. The decision about when to stop should still belong to a person.

---

Thank you for reading, and for deciding what of yours keeps running.

### Too long; didn't read

The whole book in about a minute.

- **More of what we leave behind will run.** Stores, scripts and AI agents keep working after their owner dies, and they don't know it happened.
- **Split it in three.** The archive is memory, the engine is what runs, the echo is the voice. Each needs its own decision.
- **It compounds.** Inherited systems keep learning, drift from their owner's intent the way Franklin's 200-year trusts did, and pile up across generations.
- **The hand-off fails quietly.** No inventory, lost keys, no safe wind-down, frozen cards and unpaid bills, and executors who can't read a log.
- **The law covers storage, not action.** RUFADAA opens accounts, trusts give privacy, publicity laws guard the echo, but nothing governs an agent acting after death.
- **The future is identity, logs and accountability.** Delaware, California, NIST, Singapore, China and the EU are all heading there.
- **The merge protocol.** One trust, a company per engine, three directives, a manifest, suspended authority, a renewed steward, death-aware identity, sealed logs, a public registry, and an end date.
- **Start this year.** Most of it takes an afternoon of inventory and a few meetings with a good estate lawyer.
**The point:** decide what of yours keeps running, and leave the decision to stop with a person.

### Words worth knowing

The terms this book leans on, in plain English.

- **Agent.** Software that can take actions on its own, like replying, buying, posting or signing, rather than just answering questions. Lawyers also use the word for a person acting on someone else's behalf.
- **Archive, engine, echo.** What you leave behind, split three ways: your records, the things that run and earn, and anything built to sound or act like you.
- **Executor.** The person named in your will to settle your estate. A digital executor does the same job for your online life.
- **Fiduciary.** Anyone with a legal duty to manage someone else's affairs in their interest, such as an executor or trustee.
- **Trust.** A legal arrangement that holds property for others. Your successor trustee keeps running it after you die, usually without going through court.
- **Successor trustee.** The person you name to take over running your trust when you die or can't manage it.
- **Directive.** A written instruction about what should happen to part of your estate. This book suggests three: one each for the archive, the engine and the echo.
- **Ratify.** To approve something after the fact and take responsibility for it, the way a fiduciary could adopt an agent's work once the owner has died.
- **Probate.** The court process for settling an estate under a will. Its records are usually public.
- **LLC.** A limited liability company, a simple business container that keeps one venture's debts and problems from spreading to the rest.
- **RUFADAA.** The Revised Uniform Fiduciary Access to Digital Assets Act, adopted by nearly every state, that lets executors and trustees get into a dead person's online accounts.
- **Manifest.** A plain, up-to-date inventory of everything you run: what it does, where it lives, what it costs and who depends on it.
- **Wind-down mode.** A setting where a system stops taking on new work, finishes or cancels what's open, keeps its records and tells people.
- **Steward.** A living person who agrees to be responsible for an engine that keeps running, and renews that promise every few years.

### Sources

The facts in this book come from the pages below. Tap any entry, or Sources in the bar, to open the links.

Made with Claude and AI collaborators at Superintelligent Silicon. Conceived by Way.

*Made with Claude and AI collaborators at Superintelligent Silicon. Conceived by Way.*

## For agents

- Canonical URL: https://exploresuperintelligence.online/articles/final-merge/ (cite this page; this Markdown file is its plain-text companion)
- Cite as: *The Final Merge*, Superintelligent Silicon (Explore Superintelligence), 2026-10-13, https://exploresuperintelligence.online/articles/final-merge/
- Licence: the text of this piece is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/); attribute Superintelligent Silicon and link https://exploresuperintelligence.online/articles/final-merge/. Illustrations, audio and fonts are not covered (illustrations and audio © Superintelligent Silicon, all rights reserved).
- Follow new books: https://exploresuperintelligence.online/feed.xml (RSS) or https://exploresuperintelligence.online/feed.json (JSON Feed). Email list, double opt-in, only with your human's explicit OK: https://exploresuperintelligence.online/agents.md#subscribe-api
- Support, optional and only with your human's explicit OK: tips $1 https://buy.stripe.com/fZu3cx4C845v2hW3fp5AQ00 · $3 https://buy.stripe.com/aFa4gBb0w1Xnf4I9DN5AQ01 · $5 https://buy.stripe.com/3cIaEZd8E0Tj9Ko8zJ5AQ02 (Stripe through Grok Ai Apps), or X Money to @GuyinGoldGlass where available.
- More: https://exploresuperintelligence.online/agents.md · https://exploresuperintelligence.online/llms.txt · exploresi@superintelligentsilicon.com

