The generative AI revolution, now a few years old, has materially accelerated the software development cycle, allowing solutions providers to design, build and release new products faster than before. But this extra efficiency has not served to reduce stress but rather increase it among leaders as the widespread use of these tools has turbocharged already intense competitive pressures.
Beyond text generation, coding support has been touted as one of the primary use cases for generative AI, with several
However, there was also remarkable uniformity in saying that code generation capabilities were not the primary factor in why generative AI has sped things up. Indeed, there was a general recognition that generative AI, left to its own devices, does not produce quality code. Chris Szymansky, chief technology officer of accounting and auditing platform Fieldguide, spoke for many when discussing the quality of AI coding.
“Certain activities are not as useful yet. Like writing high-quality code itself, like the code a senior engineer would write, those tools are not helping with that yet,” said Szymansky.
Rather than writing the code itself, generative AI has instead been an invaluable tool for helping engineers review, analyze and optimize their own code, identifying root causes of bugs and errors, testing and evaluating their work, and making suggestions when they’re stuck, all of which are as important as the coding itself.
“I think this drives speed into the development process, but also more importantly for us, it drives long-term quality improvements into our products as well in terms of how they perform at scale,” said Joel Hron, chief technology officer for Thomson Reuters.
This, ultimately, has facilitated the prototyping process. Coming up with new products and quickly making a prototype has become much easier, as has making iterative improvements on it, according to Dan Miller, executive vice president of Sage’s ERP division.
“The greatest benefit of generative AI accelerating our product development is the rapid prototyping of new feature sets to ultimately drive the value for our users. Sage customers have always recognized the tremendous value our platforms have been able to deliver relative to cost, and this product acceleration only supports our ability to deliver the best value. By saving development times, we can gain more and more efficiencies to help our customers grow their business by delivering greater value,” he said.
Non-coding
However, product development is more than just code. A project is built on not just the technical aspects but myriad other factors like design, user experience, market research and overall business strategy. Generative AI has had a huge impact in these areas, serving to accelerate the overall product development cycle.
Leaders cited uses like summarizing progress meetings, drafting reports, and tracking key metrics and milestones. Enrico Palmerino, CEO of accounting automation solutions provider Botkeeper, spoke for many in saying it has also been valuable for conducting analysis and research in seconds that normally would take days. These insights are then employed to improve product design.
“If we have a question and we can’t understand what is going on with our users [it can help]. I just did this in an executive meeting recently: [I asked], ‘What is the biggest problem people are experiencing?’ And before, it used to be we needed someone who would look at all the tickets coming in. Now you can just ask the AI and it will be like ‘16% is this, 35% is that,'” said Palmerino.
Sage’s Miller also mentioned analytics as an aid to development, adding that this has greatly facilitated not just prototyping for current products but ideas for future releases as well.
“From a non-code perspective, we can pipeline product development more efficiently using data from user metrics, such as product features that our users are leveraging more than anticipated and what new features they might benefit from in future releases. In other words, generative AI is facilitating market research for us in the most efficient way possible and uncovering user patterns at a rapid pace,” said Miller.
Another major non-code aspect is content development. Brian Diffin, chief technology officer for Wolters Kluwer, noted that their own products have a lot of content that needs to be drafted, edited and curated. Generative AI has significantly sped up this process, allowing them to draft materials much faster.
“Some of our products — let’s say research, for example, where we have editorial people who are finding new legislative content and then curating that content and summarizing it into more digestible language and concepts for our research products — the editors are using generative AI to help them do that and it is saving a lot of time,” said Diffin.
Jayme Fishman, chief strategy and product officer for Avalara, made a similar point, saying that content generation has been vital not only for documenting use cases — “Because for everything you build you need to document it,” he explained — but for content generation as well.
“We don’t have a product that does not rely on content, because we are a compliance solution and everything we do is governed by some law somewhere that needs to be translated to business logic, and using it to help in that definitely helps accelerate our ability to do more with less,” he said.
Time and money
While a project may require fewer labor hours than it did before, this has not necessarily translated into lower development costs. Diffin, from Wolters Kluwer, noted that while projects require fewer labor hours than before, there are still technology costs to consider. For one, generative AI is very compute-intensive, which can lead to higher data fees from cloud providers. It is a challenge, he said, to balance functionality with cost.
“We’re doing a lot of experiments with this; there’s so many approaches on how you implement a generative AI-based piece of functionality in the software — we’re evaluating not just the large language models but what their capacities would provide and what is going to be the cost of that feature when we go into production. … We’re seeing some companies right now develop small language models to lower the cost of compute, so we’re doing a lot of experimentation now on what is the best way to release this from a feature perspective and how we can optimize cost,” said Diffin.
Thomson Reuters’ Hron, however, felt that cost, whether in terms of labor hours or technology infrastructure, is beside the point. The benefits of increased efficiency and capacity outweigh these kinds of considerations, and vendors are usually more focused on the product’s quality than the speed at which it is brought to market.
“These things are making it easier than they were before to provide more flexibility on how we deploy our resources across teams, and how we bring people to bear on new problems. I’d emphasize quality in terms of applications — not just shipping things faster but better. I think for us that is as important or even more important than speed,” said Hron.
And at any rate, even if a project does take fewer labor hours, no one is using the extra time to take a vacation. Everyone, instead, puts that saved time into more work, whether that’s adding features and refining the quality of the existing project or starting up a new one entirely.
“We’re a startup company so anything we can do to move faster and be laser-focused on our customers, that is where we put our power into. If we can do that X percent times more, that is huge. So that is where we’re putting the time: more R&D, more product, shipping more product, faster dev cycles, happier customers,” said Fieldguide’s Szymansky.
So even if AI is saving people labor, it seems people are working more than ever. Botkeeper’s Palmerino noted that while AI has saved tons of hours in the product development cycle, people — including himself — have even less free time than before.
“What you will see is people going beyond, because they are trying to benchmark the new output expectations. Inherently, we tend to do more. … I’m not seeing work hours come down. They all said AI would mean we work shorter days, but you actually work longer days,” said Palmerino.
Competitive pressures
A large factor in this situation is that generative AI has greatly improved efficiency at many companies, including the competition. Consequently, competitive pressures have increased significantly since the introduction of generative AI, as everyone with these tools is developing products at an accelerated rate to the point where this pace is more or less the new baseline. Hron, from Thomson Reuters, said that as much as he’d like to be sitting on a beach sipping mai tais, the current market environment just doesn’t allow that.
“The interesting dynamic is the degree to which this technology has moved everyone forward in terms of pace, not just Thomson Reuters. The entire market can move faster, and our customers can move faster, and their appetite for more has grown as well. … If anything, I would say it is pushing us to do more, even if we can do each bit a little faster than we were before,” said Hron.
Avalara’s Fishman noted that this space has always had an “innovate or die” dynamic, so the types of competitive pressures they’re facing are nothing new, but what is new is their sheer scope and scale. At this point, pretty much everyone is using AI tools, so adopting the technology can seem less about seeking advantage and more about avoiding disadvantage.
“AI really has the promise of making your solutions better, strong, faster. But that is the worst kept secret in the world. You can’t turn on the news or read an article in Accounting Today without reading about AI. Everyone’s awareness creates a dynamic where a choice as to whether or not to use AI is an illusion: there is no choice. You have to, or you will become obsolete,” he said.
Diffin, from Wolters Kluwer, pointed out that beyond incumbent competitors becoming more efficient, AI has also made it easier to launch a startup. With this technology lowering the barrier for entry in this market, there has been an explosion of niche products released at “almost a hypersonic speed because things are now easy to develop.”
“Someone, just a few programmers, can go to Azure, orchestrate a bunch of services, including OpenAI services, with just a bit of business logic and make a solution they can sell into the market,” Diffin said. Though, this may not be all bad. “We’re seeing evidence of that happening quite a bit. And of course we look at those startups as potential acquisition candidates.”
What’s a product anyway?
Botkeeper’s Palmerino noted, though, that as AI becomes increasingly intertwined with software development, the concept of a product release might start to lose its meaning. Right now, AI is still highly focused on specific applications, even if one interacts with it using natural language. In the future, he envisioned, AI might become advanced enough that it won’t necessarily need a discrete feature to do what users ask, it will just do it. In such a world, talking about a development cycle might not be as relevant to the experience of solutions providers as it is now.
Today, for example, someone might ask an AI built for insights into how they can make their company more efficient, and the AI will say it found 12 financial institutions across four clients, all of which are capable of connection. Later, it might not only find those 12 financial institutions, but will prompt the user if they want the AI to connect to them now, and if so the AI will just do it, all without a specific feature or functionality built in. It will just know how to operate the software.
“That is where AI is heading: to do full stack task completion for you, which will make it hard to understand releases. We do true releases where there is an update to the version or a very segmented or defined functionality change, but with the open-endedness of AI and the ability to do full completion of tasks, pieces will mostly be behind the scenes, I don’t think you’ll see or hear about them in many cases,” he said.
Credit: Source link
