For me transitioning from the Air Force is a bittersweet moment, being blessed with an amazing education and stable income (and a 2.25% Covid-era mortgage) has made life easier for me and my wife than most, and for that I’m truly grateful. But ultimately I decided it was best for me and my young family to transition to the private-sector, and with that comes a certain freedom to post your opinion on the internet without the fear that your next assignment will be at Minot AFB.
I wanted to share my experience as a junior military officer (JMO) in a technical career field and discuss my frustrations and successes and give my very unqualified opinion on how things should change.
I’ll be breaking this down into a series with key-themes:
- Access to high-quality software and infrastructure
- Becoming a middle-level manager (The good and the bad)
- Compelling work and leadership that understands technology
- Final thoughts and what’s next for me
The first thing that immediately frustrated me when coming on to my first (and only) active-duty assignment, was the lack of quality software, infrastructure, and data. Procuring basic freeware (R, Python, PowerBI, etc) became an enormous burden of scanning through every U.S. gov’t agencies’ software approval site for a letter that an important person signed that allowed you to do your job. Not to mention access to the myriad of different sites requires four-different signatures, and setting a 27 digit username and password that expires every 5 days. Anyone who has worked on getting basic software installed on a gov’t computer knows this pain. Not to mention the non-sense way the AF makes you track the number of “licenses” you can have for free-ware. Worst of all, the people with the admin-rights to install software aren’t even military or civilian employees, but random contractors that have no interest ensuring you have the software you need to do your job, but instead actively resist at all cost to protect against any perceived risk.
We have to do better, and regularly review and approve new software so our people can be the most effective and productive scientists, and not be stuck with versions of 5-7 year old software. Additionally, we need to ensure our people have access to recent packages and libraries that make life easier, which is often it’s own problem. Dependency management is an absolute nightmare in and the “It works on my computer” problem is persistent across anyone programming or using software in the Air Force.
Software, however, is just one piece of the picture. We need the digital infrastructure to move products off of personal laptops and into cloud environments where they can be easily run and shared. Having built a variety of simple apps, dashboards, and the like it’s always been a nightmare to share and run apps. Between find a hosting platform and getting the required licenses it’s incredibly frustrating and prevents insights from getting into the hands of the decision makers that need them. Though the Air Force has made some improvements in this regard, the systems are less than user-friendly and suffer from the same software issues as above (e.g. building app with R 4.1 but the cloud only runs 3.6). The current analytic cloud platform of choice for the AF is a Palantir Foundry instance, however it encourages heavy use of their own built-in apps with point and click GUI interfaces. Almost everything is abstracted away behind settings, drop-downs, and menus. As a result, we teach our people how to make simple bar graphs in a proprietary software, not true analytic skills. Not to mention the mess of access control that prevents our own active-duty people from accessing the data they need. This is not inspiring work for our analysts, whom are often way over-qualified to be building dashboards with simple summary statistics.
The Air Force often broadcasts the need to develop advanced AI technology, but fundamentally lacks the software and infrastructure to do so. We have incredibly talented and smart people among our ranks but we don’t give them to tools to be successful or to create meaningful products and analysis. There’s no place to put a docker container for a model API, or a means to train a deep-learning model with GPUs. We are stuck with boring reports, slide-decks, or a glorified bar-graph dashboard. We can and need to do better, our people are capable of much more. Stay tuned for my next topic - becoming a middle-level manager.
(BTW I had originally intended to include data in this topic, but in short it’s a complete mess of stovepipes and silos).