GitLab Duo plans harness growing interest in platform AI
GitLab's next release will tie its Duo AI tools to the full DevSecOps pipeline in a bid to capitalize on increased interest in AI automation among platform engineers.
GitHub Copilot remains the coding assistant of choice among developers. But GitLab's near-term roadmap appears primed for the next major wave of AI adoption among enterprise platform engineers, according to industry experts.
With the general availability of its Duo Chat tool this week, GitLab now matches many of the features, such as prompt-based code refactoring, code explanation and automated test generation, that can be found in GitHub's Copilot Enterprise and Copilot Chat for Visual Studio. A GitLab Duo Enterprise edition due out in the coming months will add more developer features that are already familiar to GitHub Copilot Enterprise customers, such as pull request summaries.
But more importantly, GitLab Duo Enterprise will include a list of AI-driven automation features for other stages of DevSecOps pipelines, such as automated root cause analysis and security vulnerability troubleshooting, that are poised for fresh growth as enterprises adopt generative AI, according to James Governor, an analyst at RedMonk.
"Things like creating, maintaining and documenting issues are a big part of how software is built today, and that stuff is actually really compelling already in terms of what GitLab is offering," Governor said. "The ability for AI to generate code for management as well as for application development is definitely going to be one of the things that is significant and important [going forward]."
Duo Chat was the last beta-stage feature left within the GitLab Duo Pro tool set for software developers, which also includes code completion and code generation tools that were made generally available in January. Duo Pro, priced at $19 per user per month for GitLab Premium and Ultimate subscribers, supports centralized access controls according to projects, groups and subgroups, giving enterprises precise power over which DevSecOps teams can access generative AI features.
This emphasis on security and enterprise control over AI has been the main theme of GitLab's AI strategy since it first added Google Vertex AI integration last year with a pledge to keep customers' data private. That strategy is likely to increase the appeal of GitLab Duo to an audience of enterprise IT organizations concerned with governance, Governor said.
"One of the reasons that GitLab [originally] took off was that they were doing direct sales into enterprises that, in many cases, wanted to do things on premises, were concerned about data residency and things like that," Governor said. "They've had to bring over some of that thinking to their AI tooling. … Copilot is a great tool, but there are some customers that want things a little bit more locked down right now."
GitHub vs GitLab: a tale of two audiences
Microsoft and GitHub, which have a user base of some 100 million application developers, have a commanding competitive lead in IDEs with Visual Studio Code and developer code assistants with Copilot. The two have the further advantage of a close partnership with ChatGPT creator OpenAI, and both already offer access to OpenAI's latest GPT-4 Turbo model, while GitLab has yet to integrate Google's latest Gemini 1.5 Pro.
"We're looking at Gemini Pro 1.5 for the broader context window. But things like code suggestions need very low latency, and [Google has] not offered a Gemini variant of the coding models yet -- things like Gecko and chat-bison," said David DeSanto, chief product officer at GitLab.
However, GitLab Duo Chat ships this week with support for Anthropic's latest Claude 3 model, which has been growing in popularity among developers, Governor said.
"A lot of developers are going gaga for Claude at the moment, because it's generating good, clean code," he said. "Microsoft and GitHub are benefiting from that OpenAI relationship, but the rest of the industry is not standing still."
GitLab Duo Chat support for refactoring code is especially significant given its enterprise focus, said Devin Dickerson, an analyst at Forrester Research.
Code refactoring using AI remains a work in progress since even the largest context window currently available -- Google's Gemini 1.5 Pro model -- can accommodate up to 30,000 lines of code, and a legacy enterprise application might easily have more than that, Dickerson said. But it represents the largest potential productivity gains for enterprise developers using AI in the long term.
"When you have a blank canvas, it's easier to take suggestions," he said. "The more complex scenarios are where you've got lots of existing code [and] lots of existing dependencies and you need suggestions that are aware of those complexities and your constraints."
Platform engineers hold the keys to AI in the enterprise
When it comes to security-conscious enterprises and platform engineers, GitLab is potentially positioned to turn the tables on GitHub and Microsoft. For example, GitHub Actions just rolled out Azure private network support for hosted runners earlier this month -- a feature that's been supported by GitLab for years.
Microsoft and GitHub are benefiting from that OpenAI relationship, but the rest of the industry is not standing still.
James GovernorAnalyst, RedMonk
These vendors are equal in analyst evaluations of DevOps platforms. Gartner ranked Microsoft and GitHub No. 1 in its 2023 Magic Quadrant for DevOps platforms, with GitLab a close second, while GitLab earned the only "leader" designation among 13 vendors included in "The Forrester Wave: Integrated Software Delivery Platforms, Q2 2023"report.
GitLab must contend with enterprise platform AI offerings from the likes of Red Hat, which offers AI-driven features in its Ansible IT automation platform and partners with Dynatrace for observability-driven AIOps. Atlassian Intelligence is another cloud-based option for enterprise platform engineers that want to incorporate AI. Google, GitLab's partner, rolled out its own Gemini Cloud Assist AI tool for application lifecycle automation this month.
Overall, these vendors are homing in on an important trend in enterprise generative AI adoption, which has yet to go fully mainstream in production, Dickerson said.
"Things like Gemini Code Assist are moving things in an interesting direction, because it really hits the pain points that I hear about from enterprise customers trying to figure out how to put the organization in a position for developers to be able to innovate," Dickerson said. "That comes down to those precursory steps of establishing platforms that can help abstract away some of the complexity that distracts developers."
Enterprise platform engineering teams might hold more purchasing power over AI tools than developer teams in a general climate of economic uncertainty following a boom in tech spending during the COVID-19 pandemic, Governor said.
"We're much less in a 'oh, let developers do what they want' sort of mode that we were in three or four years ago," he said. "So enterprises want to invest; they see AI as a possibility. But then they're like, 'You really need to show me exactly how this is going to benefit me right now.'"
That's where AI automation that injects efficiency into existing platform engineering tools will have strong appeal, Dickerson said.
"If I can push my code through an automated pipeline that has testing and provides some information assurance capabilities, it enables faster iterations out into production," he said. "Whereas if I have to tack those tools on myself or use third party services, it becomes a lot more challenging."
Beth Pariseau, senior news writer for TechTarget Editorial, is an award-winning veteran of IT journalism covering DevOps. Have a tip? Email her or reach out @PariseauTT.
