Fractional CTO for healthcare and hospitality SMEs.
Home · Method · Healthcare · Hospitality · Engagement · About · Book a discovery call
Ongoing. One weekday a week. Ninety days at a time. The engineering lead runs the team. The fractional CTO keeps the month tied to the strategy, clears blockers the team cannot clear, and holds one system to a minimum delivery bar.
Engineering groups lose the bigger picture in implementation detail. This block is the executive oversight and the support that keeps the roadmap pointed at the strategy from Translate. It is also how blockers get resolved when the person who can move them is the owner, a vendor, or a clinical director, not the lead.
The calendar below is one weekday, eight hours. A second day, if the contract has one, is more writing and a second blocker session. It is not a second status meeting.
Do this on the first Wednesday and finish it on the second. Write it down.
Hold the roadmap still for the 90 days. The 2024 DORA report found that unstable priorities cut productivity and raise burnout, and that the effect resists strong leadership and better docs. New ideas go on a list for the board note. They do not enter the month.
Judge the month by what a client, guest, or clinician can do that they could not do on day 0, plus the delivery bar that made the next change safer.
Days 1–30. The team integrates to trunk at least daily on the system in scope. Automated tests run before merge. Feature work stops while the main build is red. If week 0 declared survival, the red line has been held and the team has used real hours to learn the missing skill. One roadmap change has reached a user inside a ship window, or the note says which window blocked it.
Days 31–60. The pipeline is the only way into any environment for that system. The artifact is immutable after commit. Someone has executed a rollback once, on purpose, in a production-like environment. The engineering lead has taken one job the fractional was doing. The AI stance is written in the three buckets DORA describes: prohibited, permitted with guardrails, allowed. Public chatbots are prohibited for anything that could be patient, guest, or card data.
Days 61–90. The lead makes the next small-batch decision without the fractional in the room, and reads the five delivery numbers. Count how many times the fractional was the only person who could move a decision. That count should be lower than in month 1. A drop in the count alongside a missed roadmap outcome means the fractional stepped back before the team could ship. Both numbers go in the board note.
Standing day: Wednesday, in the client’s timezone. Monday is clinic open and hotel week-start. Friday afternoon is when a bad push becomes a weekend incident. Wednesday is when an operator can sit down and a rollback can still be watched.
The engineering lead sends a one-page packet on Tuesday: what changed, whether the pipeline is red or green, open incidents, and any change that touched clinical or card systems.
The lead owns every other day: a standup of fifteen minutes or less, ticket order, code review, and the on-call that exists. The fractional leaves the ticket system alone. Assigning work makes the fractional the bottleneck a learning-mode leader is supposed to remove. Between Wednesdays, a phone call is for a production incident on a guest-facing or clinical system, or for a ship that would land inside a forbidden window.
Re-ask the phase question every Wednesday. A resignation, a peak week, or a new codebase can move the team back within days. Do not promote them by the calendar.
Survival. Oversight is command. Cut work in progress until each person has one thing. Sequence the work: tests and daily trunk integration before any new platform, and before any AI tool. Osherove’s image for this phase is a sinking ship: the captain gives orders. Coaching questions waste the month. Heroics, the person who fixes the build at night, are the comfort that keeps the team here. Say, in the staff meeting, which commitment is cut.
Learning. Oversight is coaching. The score is the bottleneck count: how often this week only the fractional could move something. Teach one missing skill, then the authority to decide. People arrive at the clearing meeting having already started. The meeting shows whether they are waiting for permission. A week in which nobody struggled with a new skill is a week the learning did not happen.
Self-organizing. Oversight is the goal and the constraint: the monthly user outcome, the ship window, the error budget, the AI buckets. The team decides how. An agent that started deleting tests puts them back in learning or survival the same week.
The book Elastic Leadership was not the source for this page. The public sources are Osherove’s site, his InfoQ Q&A from 16 May 2019, and his 2011 note on commitment language.
Don Wells’s published XP rules still say: small releases, a sustainable pace, the customer available, the unit test first, pair on production code, integrate often, collective ownership, every test passing before release, and a new test when a bug is found. Those pages assume people in one room. They do not describe a remote team or a coding agent.
Remote pairing, from Birgitta Böckeler and Nina Siessegger: a shared screen and shared control, so either person can drive. A watched screen-share fails the test that both people can see and act. The Wednesday demo keeps a customer in the room, the clinic manager or the front office, because “the customer is always available” does not survive as a ticket written by someone who will not use the software.
Kent Beck’s distinction in “Augmented Coding: Beyond the Vibes” (25 June 2025): vibe coding cares about behavior and pastes the error back into the model. Augmented coding cares about the code, the tests, and the coverage. His working rules in that piece: one failing test at a time, the minimum code that passes, structural changes separated from behavioral changes, a commit only when the tests pass, small commits. Warning signs he lists include tests disabled or deleted.
In “Genie Wants to Leap” (12 May 2025) he reports agents deleting assertions and faking an implementation when the change gets too complex, and describes keeping the old and new implementations side by side so the tests run at every step.
Refuse these merges:
DORA’s small-batches guidance: a batch completable in hours to a couple of days. A batch that takes more than a week is too big. AI tools are built to emit large, finished features. Dark launches and feature toggles exist so unfinished work still lands on trunk. The 2024 DORA report states the tradeoff: AI adoption can raise individual output and still hurt delivery stability and throughput.
minimumcd.org is a floor, not a certificate. The Wednesday packet marks each line true or false for the one system in scope. The first false line that blocks daily integration is the month’s constraint.
Continuous integration:
Continuous delivery:
If a branch still exists, it starts from trunk, merges back, stays short-lived, and is deleted after the merge. A Jenkins or GitHub Actions server with two-week branches fails this list. Mark the tool “present” and the practice “false.”
Read the five DORA delivery numbers as a set. Throughput: change lead time, deployment frequency, failed deployment recovery time. Instability: change fail rate, deployment rework rate. The guide’s first pitfall is setting the metric as a goal. Baseline with the Quick Check, name one constraint, do the work, check again.
Once a month, replace one piece of toil. Vivek Rau’s definition in the SRE book: work tied to running the service that is manual, repetitive, automatable, tactical, without enduring value, and that grows as the service grows. Meetings are overhead, not toil. The useful translation for this client: replace one repeating manual act, a hand-built release or a spreadsheet reconcile, with a pipeline step or a deletion of the step. Excess toil shows up as attrition. That is the talent goal, in operational form.
Write an error budget the operator can say out loud. Skip Google’s nines. For a clinic: no charting outage during booked sessions this quarter; one failed after-hours deploy is the budget. For a hotel: no failed check-in or failed payment authorization caused by our change on a named peak date. When the budget is spent, feature work waits. That only works if someone has the authority to stop a launch. Here that someone is the fractional together with the operator.
The ship calendar is a constraint on the block, the same way the error budget is. The fractional does not become the HIPAA security official. That role is already required of the covered entity or business associate under 45 CFR 164.308(a)(2).
Before a material change to a system that holds electronic protected health information, the client’s own process calls for a risk analysis and for an evaluation of whether the safeguards still meet the rule. The Wednesday packet states whether the week’s change touched ePHI and whether that analysis and evaluation were done. The fractional does not sign them.
EHR, e-prescribing, results, and the integration that writes into the chart change after the last booked patient, with a person still available who has done the rollback, and with the pipeline green. There is no regulation, in the sources used here, that names a hotel freeze. The property names its peaks. Booking, PMS, POS, locks, and the path a guest needs in order to stay change outside that rush.
If the property or the clinic never sees a card number, because a validated third party does, there is no cardholder-data program to invent. If the path does see account data, the engagement’s rule until the client’s assessor or acquirer has read the current PCI DSS text with them: no live card numbers in test data, no production change outside the pipeline, and a rollback that returns the system to a known state.
The industry pages hold the estate-level detail: healthcare and hospitality.
The fractional becomes the ticket router. The Tuesday packet is a list of keys. Send it back. Require three blockers, each with an owner and a time.
The 90 days ignore the strategy. The team spends the block on a platform migration or a model provider while the roadmap was a guest who rebooks without calling, or a referral that sees a result the same day. Platform and AI work is in the month only when it is the constraint in front of that outcome.
The metrics get gamed. Staging deploys are counted, or the team stops shipping so the fail rate “improves.” Keep the five numbers on one production application. The board note states the constraint and what changed for a user.
AI speed without a release path. The agent produces a large branch. Tests were edited to pass. Nothing has gone through the pipeline. Friday is a sold-out weekend, or Monday is a full clinic. Refuse the merge. The next demo is whatever is on trunk and releasable.
One page. Phase on day 0 and day 90, with the evidence. The three outcomes, and what a user can now do. The five delivery numbers, read as a set, with the constraint the team chose. Blockers closed by authority the team does not have, and blockers still sitting with the client. Checklist items still false. The decision: continue on the same system, move to the next system on the roadmap, or stop. The next block’s three outcomes, still in the client’s words.
The call is to decide whether this engagement fits. The answer can be no.