AI’s Impact on the Frontend and Developer Productivity
It’s been a year since ChatGPT entered the landscape. That’s a lot of time for technology, but not a lot of time for many organizations, especially in regulated industries.
Technology leaders from KeyBank and reinsurance company General Reinsurance Corporation (Gen Re), along with low code platform OutSystems, shared how they’re approaching AI during a media roundtable held Oct. 25.
Each has spent that year in evaluation and research mode. While they said generative AI is not ready to operate without a human in the process, they can see applications for the frontend. Dominic Cugini, chief transformation officer at KeyBank, said if generative AI could be trusted, then a chatbot would be one potential application, eventually.
“There [are] a lot of people that are like, great, put this behind a chatbot, let it start responding to our clients,” he said. “While that is fun, that also could be very dangerous, right? …Our executives have locked arms now and understand that and it’s [leadership’s] saying let’s take an intentional approach to this. Let’s have humans in the loop for the foreseeable future and […] continue to allow this to mature.”
In the long term, one KeyBank goal might be for AI to help quickly collect the information for a loan, for example, and enable the bank to process that loan in minutes rather than hours or days.
“There’s some real use cases if we can get trust with its ability,” Cugini said. “How do we get those results, those answers better that the clients looking for, whether they’re calling in for fraud or they’re calling in for new loan? It will help expedite that workflow once we get there.”
Hallucinations Be Gone
When asked if hallucinations were a concern, Cugini responded that hallucinations in LLMs “are starting to be a thing of the past.”
“Those that are out there creating the LLMs for generative AI have kind of identified the issues there,” he said. “So when we get into training and actually production of something, my expectation is hallucinations will be a thing of the past.”
If hallucinations are happening still, then that’s another reason to keep a human in the loop to ensure the output for accuracy and appropriateness, he added.
“That’s not only for hallucinations, but just for the quality of answer and making sure that it’s reflecting our brand in the right way,” he said. “The human in the loop is really that ultimate control.”
AI Reusability, Developer Productivity, and Low Code
Although as Tiago Azevedo (CIO of OutSystems) pointed out, even humans make mistakes; and for some situations, AI may outperform humans.
“Even though it comes with a lot of caution and tests and experimentation, in our view, for this type of application, generative AI actually challenged all the concepts — including the agile ones, where we are now looking at very, very quick turnarounds, experimentation within the realms of sometimes hours of a day,” he said. “There’s a lot of possibility in a controlled way to actually start moving into a customer-facing approach, at least in our industry.”
OutSystems experimented with AI in customer support. While their policy was to always have a human agent validate the AI, they used AI to support the frontline customer service workers.
“What it meant was allowing agents to have predefined answers that could accelerate the time to answer the clients faster,” he said.”
Azevedo also asked how Gen Re and KeyBank see AI impacting developers.
“It will change how we code, but also what we code,” responded Gen Re CTO Frank Schmid. “And it will go way beyond just the coding task. We are talking about application design, and not only design of individual applications, but also workflow, design, end-to-end.”
Composibility Key to AI
Composability is the ability to combine existing components to create new and more complex functionality; it will be key to realizing wins with AI as well, Schmid added.
“The thing we need for sure is composability because AI will play different roles at different points of the decision-making process,” he said.
Take for example the underwriting process: It starts with the interest of data and goes through the various nodes of decision-making until the insurer decides to price the risk and submit a bid; and that risk may be bound. That means different AI systems will play a role in the process, he explained.
“It’s important that you have a workflow tool such as low code that allows you to create a highly modular application landscape, a highly modular workflow that integrates well with other tools and is also highly reversible, that you can easily reverse certain decisions that you have made, because this is a technology that is still evolving at a rather rapid pace,” he said.
Low code doesn’t necessarily mean “citizen developer”, because a technical understanding of the software development lifecycle is key to developing with AI, Cugini pointed out. While low code has meant a faster turnaround for KeyBank projects — shaving nine months or more off of projects, Cugini said, low code still requires developer involvement.
“It’s not necessarily just taking every engineer and showing them how to use low code platform. You still need the technical understanding of software development lifecycle of critical testing, of the due diligence,” he said. “That’s where citizen development can get a little concerning to me at times because you don’t have that necessary rigor in place.”
Using AI for Complex Knowledge
Schmid said for the reinsurance company, using AI is about finding process improvement. He primarily sees AI’s use case as being decision preparation or support, not actual decision-making.
“It’s really all about workflow and inserting components that are AI-assisted,” he said. “For instance, as underwriting submissions come in an email with attachments, that we have a tool that extracts that information, then classifies that information in preparation for decision making, etc. AI will play a role in various components of this workflow, which means the workflow has to be highly modular.”
Gen Re is also considering use cases for legal, such as the classification of contracts with non-disclosure agreements and reinsurance contracts.
“Contracts are complex bodies of knowledge and we see great opportunity here to enable a conversation with this body of knowledge, the way we think about AI.”
And that may be the ultimate use case for frontend AI: Enabling a conversation with a complex body of knowledge. After all, it’s how ChatGPT wowed the world nearly 12 months ago.
