Rob Walling and Einar Vollset gave a MicroConf talk called "Your Features Won't Save You: The Eight Moats That Still Work in the Age of AI". It's aimed at software founders. I watched it and kept thinking about what AI in fund administration means for the firms that do the work.
Their model is simple. AI attacks a business from two sides: more alternatives to you, and less need for you. Then they name eight moats that still hold, each with a question you can score yourself against. Their worksheet is free.
Here's how I'd apply it to fund administrators, and to the fund-ops teams at GPs who buy from them.
Five attacks that land in fund admin
FundAdmin AI · Field notes
Five attacks that land in fund admin
AI attacks from two sides: more alternatives to you, and less need for you.
More alternatives to you
- The GP builds itA fund-ops team rebuilds the two workflows it uses every week.
- Open source gets good enoughFree toolkits draft a source-linked first pass on fund documents.
- The model absorbs the first readA model drafts the summary, with section references.
Less need for you
- Headcount-based delivery shrinksIf AI halves the effort per fund, clients ask about the fee.
- An agent becomes the userPortals need scoped permissions and an audit log for agents.
The GP builds the workflows it actually uses. A fund-ops team only uses a few of your workflows every week, such as the capital call check and the side-letter tracker. A small team with an AI assistant can now build those two for itself.
Open source gets good enough. I'm part of this one. FundAdmin Skills is free and MIT-licensed: 60 skills that let an AI assistant draft a first-pass review of fund documents such as an LPA or a capital call, with every finding linked to its source. Anyone can install it today. If part of your value was "we read the document first", that part now has a free alternative.
The model absorbs the first read. The first pass through a long LPA used to be a junior analyst's job. A model now drafts it, with section references, before anyone has opened the PDF. The person who checks the draft still matters. The person who typed the summary matters less.
Delivery that scales with headcount shrinks. Walling's warning about per-seat SaaS applies to anyone whose fee tracks the people it takes to do the work. If AI halves the effort to service a fund, expect clients to ask what happened to the fee.
An agent becomes the user. Investor portals were built for people who click. Soon a GP's agent will log in overnight to pull the capital account statement and reconcile it. The portal then needs to give that agent scoped permissions and log what it did.
Scoring an administrator on the eight moats
Walling scores moats from 0 to 100. One or two strong ones are enough to be hard to replace.
FundAdmin AI · Field notes
Score your firm on the eight moats
Rate each from 0 to 100. One or two strong moats are enough to be hard to replace.
- Buyers know and trust youWho has built their practice around recommending you?
- You're the system of recordIf a client leaves, what can't they export?
- You're allowed to actDo you take the action, or recommend it?
- Clients trust your outputHow long does it take a client to verify your work?
- Someone official already said yesWhich licence or assurance report would a newcomer wait for?
- Everyone's already in itWho has to say yes before a client can replace you?
- Every client makes you smarterDoes one client's near miss protect the next client?
- It's hard to keep runningWhat do you maintain that you resent maintaining?
1. Buyers know and trust you. GPs choose administrators on reputation, and reputation travels through auditors and fund counsel. A faster newcomer starts that from zero. Check: who has built their practice around recommending you?
2. You're the system of record. The investor register and the history of every transfer and approval live with you. A client can export a spreadsheet in an afternoon. Rebuilding years of side-letter elections and sign-offs is the expensive part. Check: if a client leaves, what can't they export, and what breaks downstream?
3. You're allowed to act. Walling draws a hard line between a tool that recommends and a business that's authorised to do the thing. Issuing a notice and releasing a payment sit on the acting side. FundAdmin Skills stays on the advising side on purpose: it drafts and flags, and a person decides. It never moves money. Check: do you take the action, or recommend it?
4. Clients trust your output. For an administrator this is years of statements that tie out. For AI output I'd aim for fast checking: every finding linked to the clause or line it came from. In the synthetic capital call example, the per-LP schedule adds up to $115M against a stated $200M call, and the bank details change inside the notice. The review cites both, and its status is blocked: payment not approved. Checking that takes a reviewer minutes. Check: how long does it take a client to verify your work?
5. Someone official already said yes. Walling's distinction is sharp here: serving regulated clients is different from holding the approval yourself. Code can be copied in a weekend. Authorisations and independent controls reports take months or years. Check: which licence or assurance report do you hold that a newcomer would have to wait for?
6. Everyone's already in it. An administrator sits between the GP, every LP, the auditor and the banks. Leaving means moving all of them at once, and getting all of them to agree. Check: who has to say yes before a client can replace you?
7. Every client makes you smarter. Walling's learning loop runs situation, action, result, correction, and it only counts if each client's outcome improves the service for the next. Fund admin adds a hard edge: client data stays confidential, so the loop learns patterns and never shares documents. Take a wire-change email. The wire-change worked example has 8 planted warning signs, and the review found 6. The page says so. A loop that learns from every near miss is how that gap closes. Check: does one client's near miss protect the next client?
8. It's hard to keep running. KYC refresh cycles, sanctions list updates, investor tax forms that expire, filings whose templates change. Few people enjoy maintaining any of it, and that's the moat. A GP that builds its own tools inherits the upkeep. Check: what do you maintain that you resent maintaining?
Going on offence
Walling's biggest move is to become the worker: sell the outcome, and let your own agents do the clicking.
FundAdmin AI · Field notes
Where the AI stops and people decide
An agentic back office keeps a named reviewer between the draft and the action.
- 1 · InputA document arrivesA capital call notice, an LPA, a side letter or a wire-change email.
- 2 · AI draftsFirst passDrafts and flags, with every finding linked to its source. FundAdmin Skills stops here.
- 3 · People decideNamed reviewerChecks the evidence, then signs off or sends it back.
- 4 · ActionNotice and paymentThe administrator issues the notice and releases the payment.
In the synthetic capital call example, the first pass finds a $115M schedule against a stated $200M call, and the review status is blocked: payment not approved.
In fund admin that's the agentic back office. Agents draft the capital call, reconcile the capital accounts, check the wire instructions and prepare the KYC refresh. A named reviewer signs off before anything leaves the building. The administrator keeps the record and the authority to act. The agents take the typing.
His other move is to welcome the agents. GPs' agents will come to your portal whether you plan for it or not, so give them a proper door, with permissions and an audit trail. His test is a good one: ask an AI agent to solve your client's problem without naming you, and see whether it picks you.
And his pricing question translates directly. Take what a client pays you today for one fund. What would you have to deliver for them to happily pay ten times that?
Questions fund administrators ask
Will AI replace fund administrators?
AI is taking over the first read and the typing. The record and the authority to act stay with the administrator, and those are the moats that matter.
What is a moat for a fund administrator?
A reason a client can't easily replace you. Rob Walling and Einar Vollset list eight that still hold against AI, from being the system of record to holding approvals a newcomer would have to wait for.
Which fund admin tasks can AI draft today?
First-pass reviews of LPAs, capital calls, side letters and KYC trackers, with findings linked to the source. A qualified person still checks the draft and decides.
Does FundAdmin Skills move money?
It drafts and flags findings for human review, and it never approves or releases a payment.
Where FundAdmin Skills fits
I built FundAdmin Skills as the free first pass: source-linked drafts for a qualified person to check. It sits on the advising side of moat 3 by design, and it's one of the attacks in this list. The moats belong to the people who sign off and move the money.
If you want to see what a first pass looks like, the six worked examples, with their source documents and full reports, are on the FundAdmin Skills homepage.
Framework and moat names: Rob Walling and Einar Vollset, "Your Features Won't Save You: The Eight Moats That Still Work in the Age of AI", MicroConf. Worksheet: microconf.com/moats.