Everything in the last few posts needed a piece of engineering that nobody asked me for. It is a daily job that pulls usage, seat, and directory data from several admin systems, joins them by person, and publishes one view with a dated snapshot behind it. It retries when a platform's server fails. It has tests and a place to publish, and it grew a feature at a time as the questions changed.
The report is only one part of it. The same job covered the layer people reach AI through: sign-in, an audit trail of every tool call, a way for the owners of each skill to hear what is not working, and a roadmap that changed when the people using it told me what they needed. We had planned to start with one assistant for one team. Talking to those teams, we reversed the order and built the governed access first, because it was the part everything else would have to trust. None of that is on the duty list either.
Where does that work sit in the federal mandate? OMB's memorandum M-25-21 lists ten duties for an agency's Chief AI Officer, and none of them is building infrastructure. The memorandum does mention it, in a different place. An agency's AI strategy has to plan to "develop AI-enabling infrastructure across the AI lifecycle". So under the mandate, building the platform is an obligation of the agency, and no named duty belongs to the person holding the title.
That gap is what you would expect. A duty list says what a role answers for, not the work the role involves. Sanchez and Levine's 2012 review of job analysis describes formal job descriptions lagging behind the work, and people reshaping their jobs as they do them.
My reading, and it is mine, not the memorandum's, is that the duties of governing and advising need instruments to govern with, and someone has to build them. A list that names the governing but not the instrument-building leaves that work to whoever notices the gap. In my case I noticed because I needed the measurement. In my experience the most valuable work sits there, between the named duties.
The objection is that the strategy already covers it. It does cover it, as a plan. A plan is not an owner. It says infrastructure should be developed, and it does not say whose judgement decides which systems to join, what counts as usage, and what to keep. Those choices shape what the organisation can see, and so what it can decide. I think that is engineering judgement. That is a hypothesis, and I would like it tested.
None of this says the memorandum is wrong to omit it. A federal mandate is not a template for a private company, and it was written for a different job. It does make a useful test for a list of duties: for each one, what would you need to build before you could do it?
In your organisation, whose job is it to build the thing that lets the AI programme see itself?