AI Excels at Project Management, Says Moovila. GenAI Sucks.
Which is why Moovila is using pre-LLM AI developed in house to power its project management software rather than LLMs.
Allow me to begin this piece with an irony: I just asked ChatGPT to teach me the role of vertices and edges in discrete math. You’ll understand both why I asked that question and what makes the place I asked it ironic in a moment.
First, some back story. A couple of weeks ago, I mentioned in passing that mega MSP Logically has doubled the output of its project managers using MSP-specific project management software from a vendor named Moovila. It had been a while since I’d spoken with anyone at Moovila, so the comment got me wondering if advances we’ve seen in generative AI since that last conversation were responsible for Logically’s results.
Answer: not really. Yes, today’s LLMs are way better at a lot of things than the models we all marveled at two years ago, says Mike Psenka (pictured), Moovila’s CEO, but “they sort of suck at graph data and the error rate is terrible.”
Which brings us to vertices and edges. Vertices, I now know, are points on a data graph. Edges connect them. Unlike the mostly linear tasks that genAI excels at (like answering questions about vertices and graphs—see the irony here now?), project management is heaped with dependencies that can impact multiple variables in complicated ways. Vertices, edges, and graphs are the best way to model those dependencies, according to Psenka, and LLMs (still) aren’t nearly accurate enough in that environment to be useful.
“Anything other than 99.999 is going to contaminate the rest of the graph,” Psenka says, leading to waste, delays, and unhappy clients. As a result, Moovila relies on pre-genAI technology designed in house that utilizes discrete math principles and directed acyclic graph structures to monitor dependencies, assess resource requirements, anticipate timing conflicts, and propose remedies.
Not that Moovila isn’t using generative AI to enhance its platform and partner experience. LLMs will help power an interactive learning management system and best practice hub the company has coming soon. Moovila’s using agentic AI on a limited basis as well to identify project issues. But it’s intentionally opted not to make that technology fully autonomous.
“We could have flipped the switch to say, ‘we’re going to deterministically update these project dates and reschedule people,’” Psenka says. “People don’t want to do that yet culturally.” Moovila, therefore, will continue using “supervised AI” that mixes agents with human input going forward.
“The future around larger, more complex processes is a hybrid model,” Psenka says.
(One last note: I intentionally opted not to explain what “discrete math principles and directed acyclic graph structures” are a few moments ago. If you’re curious to learn, I suggest you ask ChatGPT.)
Meanwhile, over at the podcast …
I’ve got a post about changing trends in partner programs coming soon. Nothing in it, I promise you, will be more insightful than what Ryan Morris shared on the latest episode of MSP Chat, the podcast I co-host. Check it out here.
