How to Hire an Infinite Workforce, Faster

The last post argued that deploying AI in a health system is procedurally identical to hiring an employee - the vacancy, the job description, the credentialing, the total compensation, the supervision - stacked on an annual budget cycle. None of those steps is illegitimate. What takes years is running them sequentially, artisanally, for one candidate at a time, when the applicant pool just became infinite.

Hiring can be fast. Health systems fill thousands of human roles a year without a governance crisis, because a century of HR practice turned hiring into a machine: standing job families, credentialing bodies, salary bands, probationary periods, review cycles. Nobody re-invents the concept of employment for each nurse.

AI hiring needs the same machine. Here is what each party has to build.

What startups and vendors must do: arrive as a finished candidate

Most AI startups show up to a health system the way a brilliant stranger shows up to an interview with no résumé: "trust me, watch me work." In a hiring process, that candidate loses to a mediocre one with references. Speed, for a vendor, means arriving with the hiring file already complete.

Bring a real résumé. Benchmark scores are standardized test results, not work history. Only about 5% of published clinical AI studies used real patient data. Be in the 5%. Reference deployments at systems that look like the buyer - including rural and community systems, not just academic medical centers - are worth more than any leaderboard.

Write your own job description. Arrive with a one-page statement of exactly what tasks the product performs, what it must never do without human sign-off, and which existing role's work it touches. If the health system has to write your job description, you have added months.

Bring the safety file. Model documentation, bias testing, failure modes, drift monitoring, and a plan for who gets alerted when performance slips. Credentialing is fast when the packet is complete and glacial when the committee has to request each document.

Bring your own performance review. The most fundable pattern we see: products that instrument their own value reporting - costs removed, revenue captured, hours returned, quality delivered - and label each claim honestly as measured, estimated, or modeled. A tool that shows up to its annual review with audited numbers gets renewed. One that makes the health system do the math gets questioned.

Price to the value stream. If the value you create is indirect (capacity returned, risk avoided), do not price as if it were direct budget relief. Mispriced compensation is what kills renewals.

What health systems must do: build the hiring machine

Move from requisitions to a workforce plan. The unit of decision should not be "do we approve this tool?" but "what is our AI hiring plan for the year?" - a slate of roles to fill, sourced from problems operators actually raise, funded as a workforce line in the budget rather than a sequence of one-off capital requests. That single change collapses the budget-cycle penalty, because the money is waiting for the hire instead of the hire waiting for the money.

Pre-approve job families. Not every hire deserves the same interview. A low-supervision, no-PHI tool should clear review in days through a standing fast lane; the heavy process should be reserved for heavy-tier clinical hires. Tiering only creates speed if the lower tiers are genuinely fast.

Put the whole workforce on one page. You cannot manage a workforce you cannot see. Every AI tool - its total compensation including governance overhead, its adoption (the timesheet), its measured value (the review) - belongs on a single dashboard the executive team actually reads. We are building exactly this, and the exercise is clarifying: underused tools turn out to be management problems, not talent problems, and a few well-loved tools turn out to cost more than they produce.

Name the customer at hire. Every role should state on day one who it primarily serves - patient, clinician, administrator, insurer, or staff - so the inevitable design tradeoffs are decided once, deliberately, instead of re-litigated in every meeting.

Make the outsourcing answer explicit. If AI hiring is outsourcing, pair it with an insourcing commitment: a stated policy on redeploying, retraining, and growing the humans whose task mix changes. A health system that is the largest employer in its region earns the right to hire machines by being explicit about what happens to people. Silence on this question is what actually stalls adoption - in committee rooms and in communities.

The Epic question

Every hiring process has an internal candidate, and in health care the internal candidate is Epic. At HIMSS this year Epic reported that 85% of its customers are already using Epic AI, introduced named agents for clinicians, revenue cycle, and patients, and previewed Agent Factory - a platform letting health systems build their own agents inside the EHR without a vendor.

The right way to think about this is neither "Epic will do everything" nor "Epic is beatable everywhere." It is span of scope. For work inside Epic's shipped or dated roadmap, the internal candidate wins by default - integration is free, and part of the governance burden is inherited. Interviewing an external candidate for a job the internal one will do anyway is how systems waste review cycles and how startups waste eighteen months of runway.

External candidates win where the incumbent structurally cannot go: regulated and FDA-cleared AI, hardware, services layered with software, workflows spanning vendors and data outside the EHR, and depth in problems too niche for a platform roadmap. We maintain a running map of which of our operational problems fall in or out of Epic's scope, and we would encourage every health system to do the same - and every founder to have a crisp answer to one question before the first meeting: why won't Epic ship this within 18 months?

The metrics that decide who gets hired first

Given an infinite applicant pool and finite management capacity, prioritization is the whole game. Five numbers, in order:

Time-to-value. Days from approval to first measured value. This is the metric that disciplines everyone - vendors, IT, and governance alike.

Net value after total compensation. Value produced minus everything: license, implementation, infrastructure, and the governance hours nobody budgets. Many tools that look accretive on license price alone go underwater once supervision is priced in.

Adoption of the eligible population. The timesheet. A capable hire given no work produces no value, and that is a management failure, not a vendor failure - but it should still gate expansion.

Confidence of evidence. Measured beats estimated beats modeled. Weight the portfolio accordingly, and let vendors buy credibility by converting modeled claims into measured ones.

Mission alignment. Does this hire move work toward or away from the human relationships that define care? Some roles should stay human even when a machine could do them cheaper. Deciding which ones - deliberately, in advance - is how a health system deploys an infinite workforce and remains a human institution.

An infinite workforce does not mean infinite hiring. It means, for the first time, genuine choice. The systems that build the hiring machine - not just the governance committee - will compound that choice year after year. The ones that keep interviewing one candidate at a time will still be interviewing when the front door has moved again.

What we are looking to hire

We want to practice what we just preached, so here are our first three open roles.

Support for our pharmacy enterprise. Tools that help our local health system pharmacy win the prescription relationship - the case we made in The Prescription Is the Relationship: adherence, access, speed, and the clinical connection none of the mail-order alternatives can match.

Support for regulatory compliance. The rulebook - federal, state, payor - now changes faster than any human team can read it. We are looking for AI that tracks, interprets, and operationalizes compliance obligations across a multi-state system.

Support for cancer patient intake and navigation. The days between a suspicious finding and the first oncology visit are some of the most frightening in a patient's life, and some of the most operationally tangled in ours. We are looking for AI that gets newly diagnosed patients to the right care faster and guides them through it.

More roles to come. If you are building in one of these areas and can walk into the interview with the résumé, the job description, the safety file, and the performance review already written - we want to meet you.