Hiring is hard.
It takes time and effort to find great talent, determine whether their skills match the needs of the organization, compete with other organizations for their skills, and successfully onboard them so they can be productive members of the team.
And sometimes we make it even harder on ourselves. We’ve done that at times in the federal government – out of good intentions to ensure a thorough, fair, and consistent hiring process, but nonetheless the process is hard and time-intensive.
Today, OPM is announcing USA Class, a new AI-powered tool that automates the creation of position descriptions, one of the most stubborn bottlenecks standing between agencies and the talent they desperately need. But USA Class is bigger than a single product — it’s the opening move in OPM’s broader push to bring federal HR information technology into the modern era of artificial intelligence. We’re just getting started.
First, a few words on the federal hiring process today.
Before hiring managers can start the recruiting process, they create a PD (for those of you in the private sector, think of this as the job description). A PD is a statement of the major duties, responsibilities, and supervisory relationship of given positions. These PD elements are critical to determining what are the core functions that the applicant will perform, what skills are required to successfully perform those functions, what earned experience may help them in succeeding in the role? All logical stuff – and critical. If the hiring manager can’t articulate these requirements well, then they have no business hiring someone; the PD serves as the roadmap for the interview process with the goal of best matching applicants to the specific skills needed for the role.
However, developing the PD is only the first step in getting to the starting line of recruiting. Once the initial PD is written, an HR specialist (or “classifier”) needs to map the PD to the appropriate classification standard which is then used to assign the appropriate job title, occupational series, and pay level to the position. The accuracy of these three elements is critical to an efficient and merit-based hiring process.
Doesn’t sound too hard? But keep in mind the federal government today has 609 job classification standards– think of this as the job category (e.g., software engineer) – and more than 150 different compensation levels. Then compound that combinatorial problem with the roughly 150,000 jobs that the government hires for each year. Now, we are talking real numbers – and real work that consumers huge person-hours.
So, what have we done?
Well, it turns out our friend AI is pretty good at a few things – digesting huge amounts of precedent data to train an underlying large language model and then being able to call upon that corpus to create new documents with a set of data input prompts.
The PD creation problem is a perfect example of the power of AI. Our team has trained an initial model (which will continue to be augmented as new users develop new PDs) on the thousands of prior PDs and job descriptions that have been manually created over the years by federal hiring managers. We then designed a set of basic natural language prompts that a current hiring manager can answer – e.g., tell me about the responsibilities the applicant will undertake; what skills do they need to be competent in to succeed; etc.
And – out pops the new PD! The hiring manager can then refine that as needed in the system and suggest changes for the AI to generate. Once completed, the classifier can utilize the tool to review and incorporate the classification requirements to finalize the PD. The tool doesn’t just generate PDs — it generates correct ones. USA Class is built on OPM’s federal classification standards from the ground up — so every factor level and position description it generates is accurate, compliant, and ready to use. As with all things AI, the system will live and breathe and adapt as the PDs change over time and as we update our classification standards.
Don’t get me wrong – hiring is still hard, and I don’t suspect AI will fully solve that problem in the near term. But we are using AI to streamline the tasks for which computers are very capable and free up time for HR professionals and hiring managers to focus on the people-facing aspects of recruiting and assessing candidates. More to come.
Does USA Class also consider that hiring managers are just copying from the same standards used to classify the work?
Or that they’re organizing work with little to no position management strategy?
Or that they are writing PDs for specific grades rather than just describing the work as its actually done?
Does USA Class help federal classifiers better understand the work more quickly?
Does it ensure hiring managers clearly understand their responsibilities for current, accurate, properly organized and classified positions?
Will it help break down whether the work is actually analysis and probe for analytical methods?
Does it know that “qualitative and quantitative analysis for efficiency and effectiveness” is insufficient for PDs and usually reporting work dressed up as analysis?
Classification is complicated by hiring managers manipulating PDs to get to a certain grade. That actually destroys everything and its the most common practice. How does USA Class prevent that and help classifiers ensure the work is properly scoped and described?
Classification needs A LOT of help after decades of neglect, and I’m actually building a PD writer agent with copilot now, but if the agent doesnt know realistic practical classification nuances, then it’s just another way for hiring managers to exploit the system.