Google AI plans could spur low-code/no-code faceoff with AWS
Google's generative AI products play catch-up with Copilot and others, but the upcoming addition of Duet AI to AppSheet might fill a market gap as AWS scraps Honeycode.
As cloud providers rush to cash in on generative AI hype, Google Duet AI's potential strengths and weaknesses lie in its close ties with other Google Cloud Platform services, including its AppSheet low-code/no-code tool, Apigee API management and cloud infrastructure.
Google's chief competitors, Microsoft and AWS, have counterparts to Google Duet AI for developers generally available in GitHub Copilot and Amazon CodeWhisperer, while Google Duet AI remains in technical preview. Google's PaLM 2 large language model (LLM) has been viewed by industry experts as less mature than those used in competitors' services, particularly OpenAI's GPT, a version of which underpins GitHub Copilot.
Google now claims that the newest release of PaLM 2, revealed at its Cloud Next conference this week, has been substantially improved, including the ability to ingest more data, along with grounding capabilities for enterprise data to ensure accuracy and a 25% quality improvement in code generation and code chat.
However, while it's still an early market overall, to some industry observers, the enterprise generative AI ship has already sailed for developers with Google's competitors.
"Most developers have already built their ecosystems around OpenAI … such as prompt engineering techniques, data loss prevention, hallucination, citation validations, scanning for vulnerable code, et cetera," said Andy Thurai, an analyst at Constellation Research. "Unless they are [in] a Google shop … I don't think Google has moved the needle for them to drop everything and use this as an alternative."
Google has also explicitly tied Duet AI to its cloud platform. In keynote demonstrations at Cloud Next, presenters said Duet was trained on Google Cloud Platform (GCP) resources and data specifically.
This could limit the potential breadth of Google Duet AI's appeal, especially compared with GitHub Copilot, which isn't tied to one cloud platform, said Rob Zazueta, a freelance technical consultant in Concord, Calif.
"Copilot is ecumenical in that regard," he said. "[Duet AI is an] interesting play to keep folks tied to the Google ecosystem."
On the other hand, there are areas where Duet AI's ties to Google's other cloud services could help it capitalize on gaps in competitors' portfolios, said Larry Carvalho, an independent analyst at Robust Cloud.
"One of the top things that enterprises are looking for is developer productivity with AI [along the lines of] CodeWhisperer or Copilot, but then when Google is talking about it, they're talking about it more in [services such as] Google Workspace," he said.
Among Google's Workspace tools is AppSheet, a low-code/no-code app development tool Google said in March will be infused with Duet AI. In the meantime, the AWS counterpart to AppSheet, Honeycode, will be discontinued by the vendor, according to a notice sent to users this month.
"It gives the opportunity to AppSheet to do better development, almost a step ahead of CodeWhisperer, where users ask natural language questions and get code," Carvalho said. "AppSheet could solve real problems very quickly."
Thurai was skeptical that this opportunity would be substantial for Google.
"Duet AI can help create a well-designed app with no code, especially one that involves forms, chat and workflows that are Google tools-related," Thurai said. "So far, it is mostly used to create basic apps, not complex apps, but hopefully, it will evolve over time. Still, I don't see [AppSheet gaining] traction from users outside of Google Cloud."
Google CEO Sundar Pichai touts the cloud giant's progress with large language models during a Google Cloud Next keynote presentation.
Google Duet AI's links to GCP present pros, cons
Among a raft of updates from Google during Cloud Next were other previewed integrations between Duet AI and established Google Cloud services, such as the BigQuery data warehouse and Looker business intelligence. Duet AI in GCP, a natural language interface targeted toward IT ops pros and site reliability engineers, was also featured.
Fears about cloud lock-in aside, there may be advantages to generative AI tools that tie together already-familiar services, said Rob Strechay, lead analyst at TheCube, an enterprise tech media company.
"[Google is] building solutions, not just a loose connection of services," he said. "[It has] a much tighter developer relations community than even AWS, and I think the AI in [Google's] products is much more tightly integrated, especially in Workspace, [though] it is yet to be seen."
At the very least, Google must achieve parity with other cloud competitors for existing customers, said Keith Townsend, principal of The CTO Advisor LLC and a TechTarget contributor. Thus, Duet AI is likely meant to check this box and won't be the strongest differentiator for Google's generative AI offerings, he said.
Duet seems to go well beyond code co-development – it's more co-DevOps.
Keith TownsendPrincipal, The CTO Advisor LLC
But Townsend said he also won't count out Duet AI in the broader competitive arena just yet.
"Duet seems to go well beyond code co-development -- it's more co-DevOps," he said. "There seems to be the potential to help with migration of on-premises applications to the public cloud. … That's differentiating [for enterprises] undergoing an app modernization process, if it delivers on that promise."
In addition to Duet AI support for code generation and code completion in developer IDEs such as Visual Studio Code and JetBrains variants, it will also be integrated with Apigee to assist in creating APIs, a pairing that intrigued Zazueta.
"[GitHub] Copilot seems optimized for code generation; Duet seems to be positioned as a friendlier interface for occasionally unfriendly software," he said. "[The Apigee integration] expands the utility of this beyond developers. API product managers are often technical enough to understand the problem space and APIs as a solution but may lack the technical depth to really help contribute to the API's design. This tool could help them rapidly prototype and iterate on designs."
But while integrations with existing tools will appeal to existing customers, in Thurai's view, those integrations would only raise the threshold for users to consider switching from other generative AI services.
"If I am a new developer, already on a different cloud, why would I want to learn how to use GCP and wait for them to catch up with others?" he said. "Right now, they haven't moved the needle on that. But never say never."
Google's reputation precedes it … for better and for worse
Google has a generally strong track record in other forms of AI that precedes the generative AI craze of the last year or so, including its DeepMind AI sister company and tools that link AI to applications, such as TensorFlow. It has also begun to offer previews under its Vertex AI product line, which organizations can use to build their own AI apps, train their own AI models and choose from a model garden of third-party and open source LLMs.
Google has emphasized a message of "bold but responsible" AI development and enterprise data privacy, as Microsoft, GitHub and OpenAI remain in litigation over data copyright issues related to Copilot. Google also expanded its partnership with AI powerhouse NVIDIA this week, in which Google Cloud will be one of the first companies in the world to have access to the NVIDIA DGX GH200 AI supercomputer. NVIDIA DGX Cloud AI supercomputing and software will also be available to Google Cloud customers from their web browsers as part of the expanded alliance.
Vertex AI may help boost the appeal of Duet AI as well, according to Strechay.
"I believe [AWS is] betting on being an open source model warehouse," with the discontinuation of Honeycode, he said. "The problem is that GCP has something like 70% of that community."
Still, Google faces challenges with its data privacy push after Google Analytics 4 ran afoul of EU regulators over the last year, who said it violated GDPR by transferring data to the US without consent. The service has since become legal again in Europe after the adoption of a new US-EU Data Privacy Framework.
It's no coincidence that Google rolled out fresh services, such as Google Distributed Cloud, with a message about data sovereignty and meeting strict privacy requirements at Cloud Next this week, Strechay said.
"They had to, after Google Analytics was outlawed in [some] EU countries," he said.
Beth Pariseau, senior news writer at TechTarget Editorial, is an award-winning veteran of IT journalism. She can be reached at [email protected] or on Twitter @PariseauTT.
