Brigham.ai and the Rise of the AI Chief of Staff

Every operator who has run more than one business at a time knows the real bottleneck. It is not capital. It is not deal flow. It is attention. The owner-operator who can personally review a board packet, read a customer complaint, and approve a hire before lunch is a myth once a portfolio crosses two or three companies, and the usual fixes — hiring an executive assistant, promoting an analyst into a chief-of-staff seat, routing everything through a shared services function — scale linearly against a problem that compounds.

Brigham.ai belongs to a new category of software that tries to solve this at the seat where it matters: the operator, the allocator, the holding-company principal. It is not an inbox tool. It is marketed as an AI chief of staff — software that reads the same mail you read, drafts the memos you would have drafted, and chases the follow-ups you would have forgotten. The category is young and the names will shuffle, but the shape of the work the category absorbs is already visible, and the shape is worth understanding before buying anything.

What an AI Chief of Staff Actually Absorbs

The work a human chief of staff does is unglamorous and bimodal: a long tail of coordination (meeting prep, follow-up tracking, status rollups, inbox triage) and a short head of synthesis (deal memos, board packets, cross-portfolio reads). An AI chief of staff goes after the long tail first, because that is where the cost of a mistake is low and the volume is crushing. Microsoft 365 telemetry shows employees are interrupted roughly 275 times per workday during core hours. UC Irvine research puts the recovery cost at about 23 minutes per interruption, which is the actual math under every complaint about not having time to think.

Brigham.ai and its peers absorb this in four concrete places:

• Inbox triage and drafting. Reading mail against a trained voice, surfacing the five messages that actually need the principal, and drafting replies to the rest.
• Meeting prep. Pulling the attendee's background, the last three threads with them, the open commitments on both sides, and the one number that matters.
• Memo and packet production. Taking a monthly operating review from a subsidiary and rewriting it into the format the holding company reads — same numbers, consistent commentary, flags on variance.
• Cross-portfolio follow-ups. Watching for the commitment made in a Tuesday board call that never showed up in the Friday report, and surfacing it before the next meeting.

None of this is new work. It is what a competent chief of staff does on week one. The question is whether software can do enough of it, cheaply enough, to change how a holding company staffs its center.

The Cost Comparison That Actually Matters

The honest comparison is not against a $20 ChatGPT seat. It is against the fully loaded cost of the human you would otherwise hire. The BLS put the national median wage for executive secretaries and executive administrative assistants at $76,590 as of May 2025, with the 90th percentile at $109,850. Salary.com's chief-of-staff benchmark sits near $202,000 in median base pay, and the real roles that back a holding-company principal land higher once equity and bonus are counted.

Loaded for benefits, payroll taxes, equipment, and recruiting, a senior EA in a major market runs $150,000 to $200,000. A genuine chief of staff runs $250,000 to $400,000. Brigham.ai and the companies in its category are priced per seat per month, typically in the low four figures. That is a 10x to 20x gross cost advantage before accounting for the fact that the software works nights, weekends, and across all three subsidiaries at once.

The gross number is the wrong number, though. The real question is leverage. A human EA takes months to understand the business and leaves after two or three years. The software onboards in days, carries its context forward indefinitely, and does not take the institutional memory with it when it goes. That is a different asset, and it is the one institutional memory inside a holding company has always been scarcest of.

Where It Fails, and How Loudly

The failure modes are specific and worth naming before deployment, not after. The first is quiet confidence on a wrong fact. An AI chief of staff that drafts a reply citing a customer's contract terms it misread is more dangerous than one that misses the email entirely, because the mistake ships under the principal's name. Any serious deployment requires a human review step on anything leaving the building, which erodes the leverage proportionally to how much review it needs.

The second is the single-point-of-failure problem. A chief of staff who touches every subsidiary, every memo, and every follow-up is exactly the kind of concentrated dependency a long-term holder should be wary of. If the vendor changes pricing, deprecates a model, or has an outage during a board week, the holding company has a problem. The answer is not to avoid the tool. The answer is to treat the vendor the way governance without overhead treats any critical supplier: at least one backup, documented runbooks, and no workflow that only exists inside the vendor.

The third is governance that the broader market has not yet solved. Deloitte's 2026 State of AI in the Enterprise survey found that 74% of enterprises expect at least moderate use of AI agents by 2027, while only 21% have a mature agentic AI governance model in place. That gap is where unpleasant surprises come from — the agent that books a vendor the principal would not have approved, the memo that leaked a number it should not have had access to. The firms that treat deployment of agentic AI as a governance exercise, not an IT project, will have fewer of those surprises.

How a Buy-and-Hold Operator Should Deploy One

The temptation is to buy a single enterprise seat for the holding-company principal and let it radiate. That is wrong on first principles. A holding company is a federation. Each subsidiary has a CEO who is accountable for its results, and the center's job is allocation and governance, not operations. The AI chief of staff should mirror that structure.

Three deployment rules carry most of the weight:

1. One instance per operating seat, not one for the group. Each subsidiary CEO gets their own instance, tuned to their business, their vocabulary, their customers. The holding company gets its own instance, which reads what the CEOs read but reports up, not down.
2. Read-mostly at the center, write-rarely at the subsidiaries. The holding-company instance should synthesize across subsidiaries but almost never send mail on their behalf. The subsidiary instances can draft and send inside their own domain, under their CEO's name. This preserves accountability and keeps the center from creating shadow operations.
3. Measurement in hours returned, not messages processed. The number that matters is how many hours per week the CEO and the principal spent on the work only they can do. Return on attention is the right unit, and it should be baselined in month one and reviewed quarterly.

The leverage story only works if the operator uses the time back for something higher-value. Bain's research found senior executives now spend an average of 23 hours per week in meetings, up from under 10 hours in the 1960s. Automating the inbox and clawing back ten of those hours is a loss, not a win, if the ten hours go to more meetings.

The Fit With a Long-Term Owner Model

A private-equity flipper has a weak case for an AI chief of staff. The hold is three to five years, the operating team is often parachuted in, and the deal-level work is handled by associates who will leave for business school. The economics assume turnover and compress the payback window.

A long-term operator inherits a different problem. The hold is measured in decades, not quarters. The subsidiaries keep their names, their teams, and their founders where possible. The holding company's center stays small by design. In that model, an AI chief of staff is doing exactly the work that would otherwise force the center to grow — the review, the rollup, the follow-up, the memo — and preserving the structural advantage that keeps overhead low.

The economics also compound. A human chief of staff cost grows with headcount; the software cost grows with seats, which grows far more slowly than work. For a holder running six operating companies and considering a seventh under its investment criteria, the question is whether the center can absorb another subsidiary without hiring. Two years ago the answer was almost always no. The answer is now closer to maybe, and in another year it will be yes often enough to matter.

What to Buy Now, What to Wait On

Nothing in this category is finished software. Brigham.ai, its direct competitors, and the broader agentic layer are all shipping weekly, and the vendor that leads today may not lead in eighteen months. That argues for buying narrowly and under short contracts. Pilot one subsidiary. Pilot one holding-company function — most likely memo production or cross-portfolio follow-ups, because the failure modes are easiest to spot. Keep the contract annual. Keep the data portable.

Wait on the full autonomous deployment. Microsoft's 2025 Work Trend Index found only 46% of leaders say their companies are using AI agents to fully automate workflows or processes, and the ones that have done it well treat the agent as a junior employee, not a replacement. That is the right posture. An AI chief of staff is a tool for the operator, not a substitute for one. For a long-term holder, the test is not whether the software is impressive. The test is whether it compounds the operator's time, year after year, across more subsidiaries than the center could otherwise carry. That is the only version of this bet that pays.