In software engineering, the most impactful AI application to-date has been coding assistants—the Copilots and Cursors of the world. Code composition is a natural fit for generative AI. But what about the other challenges in software engineering? What about things like estimating work, or forecasting deadlines and budgets? Does AI have a role here?

It does—notably for technical project and program managers. That is, the people dealing with all those estimates, deadlines and budgets. But note too that here, we’re not talking about the generative AI of news headlines and social feeds. We’re talking about another kind.

Kinds of AI

When an AI model is provided with a sizable and relevant body of information, it can theoretically interact with it in three ways:

• It can be descriptive, meaning it can take the data and provide an overview of it, explaining what the information conveys.

• It can be predictive, meaning it can take the data, provide an overview, and make some highly educated guesses on what future outcomes will look like, and how they will play out.

• It can be prescriptive, meaning it can take the data, estimate what the future will look like, and offer insights into what changes somebody can make to achieve a certain goal.

All three have their merits, but clearly prescriptive behavior is the holy grail. To see this through the eyes of a project manager, we’re talking about a model that can:

1. Absorb all the historical data of the work activity for a given team…

2. …be given a specific assignment, such as being asked to figure out how a project can be completed by a certain deadline…

3. …and provide scenarios for achieving that in the most optimal way.

None of this is science fiction. Socratic’s Scenarios capability, as an example, allows you to experiment with different project scope and/or teams to arrive at the combination most likely to get the work completed within your ideal timeframe.

More examples of prescriptive AI in project management

Let’s consider a sampling of the kinds of other prescriptive actions that AI will make possible.

Surface bottlenecks

Tucked away inside every body of work—be it an epic, an initiative, a project—are tasks that will ultimately threaten the due date. This isn’t only for reasons of complexity. It may be that the work has fallen to a team member whose motivation is waning, or who needs help but is insistent they don’t… Whatever the case, it’s these tasks that usually inform bad surprises on or near due dates.

A prescriptive AI might examine the signal data around tasks embedded in the work systems (e.g. Jira, GitHub)—the would include “basic” signals such as time-in-Jira-status against team and personnel averages, as well as more arcane variables such as time of year and historical workloads for the team members in question—and surface a kind of heat map of tasks at risk of timely completion, alongside diagnostics for why.

Suggest resources

Given a scope of work and a due date, a prescriptive AI might suggest the team members who would give the project the highest probability of success. This would entail analyzing both current and future team capacity, as well as more advanced variables such as the cohorts who’ve had the greatest prior successes at timely delivery. In the case of development work, it might even take into account the code repos involved, and the highest contributors therein.

Smooth workloads

This model would be fantastic at picking out those who have some capacity to hand over to the project. In fact, Socratic can already reveal what people’s capacity looks like, which is the descriptive side of our work. But in the near-future it will be able to also tell you what would happen in terms of cost and time if you were to tap into that. The best upshot is that by uncovering those with capacity, this prevents project managers from requesting extra work from people who are almost at maximum capacity, and accidentally stressing them, making them feel guilty for saying no, or leaving them burned out if they say yes.

Does AI have a role in shaping team cultures?

Project managers are tasked not only with deadlines and budgets, but that also in building good team cultures. Team culture would seem to have little to do with AI, but let’s tease out the idea a little…

Consider morale. Few things eat into team morale more than overworking people. And the problem is that overwork is usually a lagging indicator—we may not know who’s overworked until they’re taxed to the point of burnout. Even in on-premise, colocated work environments (an increasing rarity), it can be difficult to spot who’s reaching an exhaustion point.

Being a manager is a person-centric role, but it’s also a position of power. That mix can lead to some complicated outcomes when pressure rises. Sometimes people feel obliged to make commitments that lead to becoming spread too thin and producing shoddier work; sometimes people themselves don’t observe how overworked.

It’s not hard to imagine a model that observes a person’s workload (think: forecast work days needed to complete current assignments), reviews their history of same, correlates past and future workload to certain productivity measures such as cycle time or throughput, and/or retention data, and preemptively flags burnout candidates to a manager.

There’s also the not-small matter of micromanagement, another thing that grinds team culture like nothing else. We’ve written about the role of AI in killing micromanagement. The short version is: with AI, we can simply stop pestering team members. Why? Because you move from being a “sentiment-based” team (aggregating opinions about how everyone feels about progress, completion dates) to being a data-driven team. The data shows what needs attention and why, and discussions are focused on that.

In short, prescriptive AI offers a real-world, tangible way to use machine learning to streamline complex decision-making processes. This is the future PMs need, and the good news? It’s closer than you think.