Salesforce just put a price tag on something most CIOs felt but could not quantify. The average employee, according to internal telemetry shared at their May event, loses two hours per day copying context across Slack, Salesforce, Workspace, and ServiceNow. Two hours. Per person. Every working day. That is not a productivity story. It is a tax line on the operating budget of every white-collar firm in the index.
Google Cloud reported the demand side of the same equation: 75% of its customers now use AI products, and 330 of those customers processed more than one trillion tokens last year. The agents are deployed. The capacity exists. What does not exist, in most enterprises, is the connective tissue that lets those agents read, write, and act across the systems of record without a human shuttling JSON between tabs.
ServiceNow’s response to this was telling. They did not release another LLM. They released a governed agent registry with audit trails, positioned not as a compliance afterthought but as the thing you buy. Ninety-five billion workflows per year now flow through that surface. The pitch is simple: if your agents are going to act on your record systems, they need an address book that knows who they are, what they can touch, and what they did.
That is one half of the lattice. The other half came from a quieter argument the same week.
The Case for Withholding
A piece circulating under the title The Case for Strategic Illegibility made an argument that cuts against most enterprise AI orthodoxy. Companies, the author argued, will need two versions of themselves. One that humans work inside, with all its messy tacit knowledge intact. And a digital twin that AI agents navigate, indexed and clean. The provocative claim sits in what you choose to put in the second version. The highest-value items, the ones that constitute durable advantage, you deliberately leave out.
The intuition behind this is not new. Anyone who has watched a competitor scrape a public API knows that legibility is a two-way street. Make your operations machine-readable and you make them machine-replicable. The novelty is the framing. Strategic Illegibility treats the question of what to expose to your own AI as a strategic decision, not an integration ticket.
The two arguments, registry and illegibility, look opposed. They are not. They are the two posts of the same fence.
What the Lattice Looks Like
ServiceNow’s governed registry is a buyable product feature. It answers the operational question: when an agent acts inside my company, who authorized it, what did it touch, and can I prove the chain of custody to a regulator. Salesforce’s two-hour tax is the cost of not having that registry. Every minute of human-in-the-loop context shuttling is a minute the registry would have eliminated, and the audit trail it produces is the artifact that turns agent action into something a CFO will sign off on.
Strategic Illegibility answers the inverse question: of all the surface area you could make legible to your agent fabric, how much of it produces durable advantage when exposed, and how much of it produces durable advantage by remaining opaque.
A concrete example. A pricing committee at a logistics firm meets weekly to set rates on lanes the public catalog does not cover. The deliberation is messy. It mixes customer history, regional capacity intuition, and a few unwritten rules about which competitors not to undercut. You could make this legible. You could log every decision, tag every input, and feed the corpus into an internal agent that proposes next week’s rates. You would gain speed. You would also commoditize the one thing your sales team uses to win deals from competitors with bigger fleets. Strategic Illegibility says: do not index the pricing committee. Index the parts of operations where speed compounds and judgment does not.
The lattice is the set of decisions about which side of the fence each capability sits on. ServiceNow gives you the registry to govern the legible side. The illegibility argument tells you the registry is not the whole product. The whole product is the registry plus the deliberate map of what stays out.
Why the Tax Is the Wedge
Two hours per employee per day is the wedge that forces the lattice conversation into the boardroom, whether the CIO is ready or not. At a fully loaded cost of $80 per hour for a knowledge worker, that is $160 per person per day, or roughly $40,000 per person per year, in pure context-shuttling overhead. A 5,000-person firm is paying $200 million annually for humans to be the integration layer between its own systems.
That number is large enough to fund any registry product on the market three times over. It is also large enough to fund the analytical work of deciding what not to expose. The CFO conversation is no longer “should we invest in agent governance.” It is “of the $200 million we are currently paying in integration tax, how much do we redirect to a governed registry, and how much do we redirect to the harder strategic work of mapping our illegibility surface.”
I want to be careful here. The $200 million figure is illustrative. The Salesforce two-hour number is self-reported by a vendor with a commercial interest in selling integration. Treat it as directional, not gospel. What is harder to dispute is the structural point: the cost of not having a registry is now visible enough that the budget exists. The cost of not having an illegibility strategy is invisible until a competitor reads your agent fabric like an open book.
The Predictable Failure Mode
Most firms will buy the registry and skip the illegibility work. The registry is concrete. It has a SKU, a vendor, and a deployment plan. Illegibility is abstract. It requires saying no to indexing things that look like obvious wins.
The failure mode looks like this. The registry goes live. Agents start acting across systems. Audit trails accumulate. Productivity numbers improve in measurable ways, because the two-hour tax was real and the registry collects it back. Eighteen months later, a competitor or an aggressive acquirer maps the agent surface and reverse-engineers the operational logic. The advantage that took twenty years to build leaks out in twenty weeks of API access.
This is the scenario the illegibility argument is designed to prevent. It does not prevent it by stopping registry adoption. It prevents it by forcing a question that registry adoption alone never asks: which capabilities benefit from being legible to our own agents, and which ones are advantages precisely because they remain in human hands.
We have written before about how governance becomes the operating moat once model access commoditizes, and about how orchestration capability separates winners from API wrappers. The lattice argument extends both. The registry is the apparatus of operating-AI governance. Illegibility is the strategic doctrine that decides where the registry stops.
What to Do This Quarter
Three actions, in this order.
First, calculate your own integration tax. Pull thirty days of telemetry from Slack, your CRM, your collaboration suite, and your service desk. Estimate the time per employee spent transcribing context across them. The number will not be precise. It will be large enough to justify the next two actions.
Second, evaluate the registry vendors against your top five agent use cases. Not against the demos. Against the use cases where you can name the system of record, the action being taken, and the auditor who would ask about it. ServiceNow is one option. The category will broaden quickly.
Third, and this is the work most firms will skip: write a one-page list of capabilities you will not expose to your agent fabric. Pricing committees. Vendor negotiation playbooks. The unwritten rules that govern how your team handles a regulator’s first call. Make the list small. Make it deliberate. Make sure the CEO signs it.
The lattice is not a product you buy. It is a posture you adopt. ServiceNow is selling one half of it. The other half is your job.
This analysis synthesizes The Agent Integration Tax Just Became the Biggest Market in Enterprise AI (BRNZ, May 2026) and The Case for Strategic Illegibility (TLDR Founders, May 2026).
Victorino Group helps CIOs decide what to expose to the agent fabric and what to withhold. Let’s talk.
All articles on The Thinking Wire are written with the assistance of Anthropic's Opus LLM. Each piece goes through multi-agent research to verify facts and surface contradictions, followed by human review and approval before publication. If you find any inaccurate information or wish to contact our editorial team, please reach out at editorial@victorinollc.com . About The Thinking Wire →
If this resonates, let's talk
We help companies implement AI without losing control.
Schedule a Conversation