The purpose of the new process is to ensure that any technology projects incorporating GenAI pass a series of tests emphasizing security, viability and effectiveness, and that they be done in more of an iterative fashion, rather than the waterfall methodology. The PDL process changes how tech projects are vetted — information that’s crucial for those selling to the state. PDL began a soft launch Feb. 25, and it shifts to a hard launch July 1.
Wednesday’s virtual Vendor Forum included remarks from Jared Johnson, deputy state chief information officer and CDT's chief deputy director.
“The stakes are high,” Johnson said. “Technology powers nearly every state service, from health care and social services to transportation, licensing and public safety. That's why we developed Project Delivery Lifecycle — to provide state entities with an adaptable approach that provides innovation, delivers business value sooner, and partners with the vendor community to develop proofs of concepts (POC) and minimum viable products (MVP) to confirm the solution we are seeking meets our needs.”
Sahana Ayer, CDT’s chief counsel, explained to attendees that when a vendor offers a solution, new criteria kick in under the PDL framework, among them:
- They must disclose any “meaningful” use of GenAI; that excludes low-level uses like having ChatGPT correct faulty grammar. Under the new framework, use of GenAI must be disclosed in cases where it would impact risk, operations, data, security, integrity or reputational risk to the agency or state.
- Contractors must receive approval for use of GenAI. The state is also not allowing the use of state data, especially confidential data or sensitive personally identifiable information. Contractors are not permitted to use that type of data for training large language models.
As with so many emerging technologies, GenAI drove these changes, said Tim Issertell, deputy director of CDT’s Project Management Office.
“When we started talking about PDL, we started with GenAI in mind,” he said. “That's where this whole concept came from.”
Issertell added: “The Project Delivery Lifecycle prioritizes an iterative, value-driven approach to GenAI project planning, leveraging through some concepts and minimal viable products. It's an iterative phasing to reduce risk and confirm business value before scaling into production.”
He also noted that the new protocol is more aligned with phased agile project methodology rather than the one-and-done waterfall approach.
“When we think about PAL, that was very project-focused,” Issertell said. “As we went through that, we developed things to get us ready for the project. And we took a waterfall approach through our project life cycles. I know through the years, we've kind of adapted and tried to do some iterative things through that. But PAL was designed in that waterfall approach. PDL is designed in a product mindset approach.”
CDT suggested that vendors and contractors visit its overview of PDL and the GenAI for California website, which offers additional guidance and resources. The slide deck, along with responses to attendees' questions, will be published by CDT this week.
“Today is just the beginning,” Johnson said. “We are planning additional workshops with the agencies and departments, and we look forward to other engagements with all of you.”