AI Strategy · Field Guide

Every Copilot Is Myopic: Your Inbox, Your Dentist, Your Enterprise

Every vendor’s AI can see only the vendor-shaped fragment of your world — and four structural locks mean it always will. The same failure repeats at three scales, which is how you know it’s architecture, not a maturity gap a v2 will fix.

Scott Farrell · LeverageAI · leverageai.com.au

Ask the AI built into your email to find the electricity invoice from June and it will find it. Ask it for “that person I had a long conversation with a few years ago about the thing we might build together — I can’t remember their name” and it is helpless. Both questions are about your inbox. The first is a lookup inside a silo. The second is a judgement across your whole life, and it happens to touch email. The vendor’s AI can do the first and will never do the second, and the reason is not that it needs another training run. The reason is structural, and once you see the structure you see it everywhere.

I run my own agents on my own email, so I’ve watched this line from both sides. Early on, a scheduled triage agent with raw inbox access was hopeless — loud about receipts, silent on the thread that mattered. The moment I gave it a compiled worldview built from a few years of my own correspondence, the same model became a genius at the job. The difference wasn’t intelligence. It was that the second agent had a model of me to reason against, sitting on my side of the boundary, and the in-app vendor AI never can. Let me make that “never” precise, because it’s a strong word and I mean it literally.

The argument, once

Here is the whole thing in two sentences, and the rest of this piece is just proof at increasing scale. Lookups inside a silo work; cross-silo judgement fails. The demo always shows you the lookup — because a demo, by construction, only asks what the silo already contains — and the job is almost always the judgement, because judgement is what you were hoping to buy.

“Cross-silo judgement” sounds abstract until you notice how much of real knowledge work is exactly that: a question whose answer is spread across systems that were never designed to talk. Why is this customer unhappy — a thread in support, a downgrade in billing, a stalled feature in the product logs. Should we run this campaign again — YouTube analytics, the CRM, last quarter’s spend. Is this email worth interrupting me for — everyone I know, everything in flight, everything I’ve already handled. None of these lives in one vendor’s box. The compiled understanding that answers them is a synthesis across boxes, and synthesis needs a place to stand that sees all of them at once.

The demo is the lookup. The job is the judgement. Vendors ship the demo and call it the job.

The four locks

Why can’t a vendor just ship the cross-silo layer? Each of them has enormous engineering budgets and every incentive to sell you more AI. The answer is that four independent forces all push the same way, and you only need one of them to hold for the boundary to be permanent. All four hold.

Hold those four in mind as we climb the scales, because the point of the escalation is that the same four locks explain the failure whether you’re one person with an inbox or a five-thousand-seat enterprise. That’s what makes it structural.

Scale 1 — your inbox

Start where everyone lives. The AI inside your mail client, and the “personal agent” the whole industry is promising, both connect to your email through a thin, deterministic bridge — search, fetch, list — the kind of interface that open connector standards were built to standardise.¹ Those connectors are genuinely useful: they ship the bottom rungs of the ladder, the reliable “find the message that says X” operations.² What they do not ship is the map above those rungs — the compiled model of who matters, what’s in flight, what you’d want to know.

So the invoice query works and the relationship query doesn’t, and it isn’t a bug. The invoice is a keyword lookup inside the silo. The relationship question is a diff against your whole worldview — it requires knowing what a “long conversation about something we might build” even means for you, which people those are, and why it mattered — and that model lives on your side of the boundary or nowhere. This is exactly why a personal agent bolted onto vendor connectors underwhelms.

A personal agent without your wiki is a genius stranger with your password. Competent, credentialed, and useless about you.

And notice lock 2 doing its quiet work: even if your mail vendor wanted to compile that worldview, it would have to build and hold a separate, permanent, cross-referenced corpus of everything you’ve ever discussed — the privacy liability alone is a reason not to, and the compute is a reason on top. The honest architecture is the one where the vendor stays a connector and the worldview is yours.

Scale 2 — your dentist

Now give it a P&L. A dental practice runs on a practice-management system, and that vendor now ships an AI feature. It answers questions about the recall list and the appointment book beautifully — because those live inside the PMS silo. Ask it the question the owner actually has at 6pm on a Tuesday — is this practice healthy, and what should I do differently this month? — and it goes blind, because that answer spans the marketing spend, the Google reviews, the payroll, the phone system, and the owner’s own inbox, none of which the PMS can see.

The right-hand column is a whole-practice operating review, and it is a cross-silo synthesis by definition. I’ve built exactly this — deterministic collectors pulling each system, layered summaries, a model at the top that reads the compressed whole and finds the cross-silo moves a single-system view can’t. The PMS vendor will never build it, and now lock 4 is the binding one: its AI will never tell the practice that a service it bills for should be dropped, or that the process the PMS was sold to run shouldn’t exist. It can optimise the recall list. It cannot question whether the recall list is the point.

Scale 3 — your enterprise

Same shape, three more zeros. The modern enterprise doesn’t have an AI; it has a portfolio of copilots — one in the CRM, one in the support desk, one in the code host, one in the productivity suite, one in the data warehouse. Each demos well inside its own silo. The pitch is that together they cover the business. They don’t, and the reason is precise: five fragments don’t compose into a picture, because composition needs a shared map that none of the five can hold.

Take the question a leader actually asks: why is churn up in this segment? Watch it decompose and die.

Here comes the serious objection, and it deserves a serious answer because a real vendor will make it: “But we ARE the cross-silo layer — our suite spans CRM, support, and productivity, so buy more of us and the fragments compose.” Answer it twice.

First, coverage. Even the broadest suite covers a fraction of a real enterprise’s estate. The billing system is someone else’s. The data warehouse is someone else’s. The phone system, the shadow-IT spreadsheets, the industry-specific tools, the email threads where the actual decisions happened — all outside the suite. A cross-silo layer that can’t see most of the silos is just a bigger silo, and the churn question still spans its boundary.

Second, and worse, lock-in. Suppose one vendor genuinely could see everything. Handing that vendor the compiled model of how your business actually works is the most complete lock-in in commercial history — not lock-in on a database you could migrate, but on the understanding of your own company. You would be renting your own worldview back from a supplier, on precisely the asset a business must own. The suite-vendor pitch, taken to its logical end, asks you to make your comprehension of yourself a line item on someone else’s invoice.

An AI strategy that is the union of vendor roadmaps is a strategy for owning nothing.

The fix is the same at every scale

Because the failure is one structure repeated, the fix is one structure repeated. Assemble the compiled cross-silo layer on your side of the boundary — a worldview you own — and demote every vendor AI to what it is genuinely excellent at: deterministic connectors into its own system, operating underneath your map. The mail vendor stays the fetch-the-message tool. The PMS stays the pull-the-recall-list tool. The five enterprise copilots become five reliable adapters. The synthesis — the judgement, the join, the “should this exist” — happens above them, in a substrate that sees across because you built it to.

This is not anti-vendor; it’s a correct division of labour. Vendor AI at the bottom rungs is fast, cheap, and reliable, and you should use it there. What you must not do is wait for it to grow the top of the ladder, because the four locks guarantee it won’t. If the boundary is structural for Google — with the best engineers and the deepest pockets in the industry — your vertical-SaaS vendor is not quietly fixing it in the v2 on their roadmap.

Run the rule on the copilots you already have. The mail assistant: understanding lives on their side — connector, fine, don’t expect judgement. The PMS feature: their side — connector, fine, but the operating review is yours to build. The enterprise suite promising to be your cross-silo brain: their side — and that’s the one to refuse, because it’s the asset you can least afford to rent.

The strategy sentence is short. Vendor copilots are features of their products; your worldview is a property of your business. Every vendor will keep shipping the impressive in-silo demo, and every demo will keep being the lookup, not the job. The work — the part that compounds, the part competitors can’t clone, the part that can say “stop” — is the compiled understanding you assemble on your own side of the line. Build that, and every copilot on the market becomes a hand you can use. Skip it, and you’ve handed your eyes to whoever sold you the last feature.