For developers, Gemini 3 is expanding the promise of AI beyond simple code completion to the concept of autonomous agentic workflows.
As development teams face increasing pressure to deliver complex software faster, the bottleneck often transfers from writing syntax to managing logic and architecture. Google’s latest dual release – the Gemini 3 model and the Antigravity development platform – attempts to address this operational ceiling by reframing the developer’s role from a writer of code to an architect of agents.
The announcement marks a distinct deviation from the chat-based assistance that has defined the current generation of AI coding tools. By introducing a model specifically optimised for “long-horizon” tasks and a platform designed to manage asynchronous agents, the company is betting that the future of enterprise software development lies in delegation rather than mere acceleration.
The reasoning engine: Gemini 3
Central to this adjustment is Gemini 3, a model Google claims establishes a “new foundation of intelligence” for agentic coding. According to the technical report, Gemini 3 Pro achieved a 54.2 percent score on Terminal-Bench 2.0, a benchmark testing a model’s ability to operate a computer via a terminal, compared to 32.6 percent for its predecessor, Gemini 2.5 Pro.
Perhaps more relevant to developers assessing automated code generation is the model’s performance on SWE-Bench Verified, where it reportedly scored 76.2 percent on a single attempt. While benchmarks should always be viewed with caution until independently verified, these figures suggest a capability to handle multi-step reasoning that previous iterations struggled to maintain.
The model introduces an interesting economic proposition for businesses scaling AI adoption. It is currently available in preview at “$2/million input tokens and $12/million output tokens for prompts 200k tokens or less”. This pricing structure will likely force IT leaders to calculate the ROI of autonomous agents carefully and balance token costs against the expensive engineering hours saved on boilerplate or refactoring tasks.
The model architecture also prioritises context retention, a necessity for reviewing large codebases. Google notes that the model “handles complex, long-horizon tasks across entire codebases, maintaining context through multi-file refactors, debugging sessions, and feature implementations”. For enterprise legacy systems, where understanding the interplay between decades-old modules is vital, this long-context capability is a prerequisite for deployment.
The vehicle: Google Antigravity
While the Gemini 3 model provides the engine, the vehicle for this new workflow is Google Antigravity. This new platform, currently in public preview, suggests the traditional IDE is ill-suited for managing autonomous agents.
“The IDE of today is a far cry from the IDE of just a few years ago,” Google states. “Antigravity is evolving the IDE towards an agent-first future with browser control capabilities, asynchronous interaction patterns, and an agent-first product form factor.”
The operational implication here is heavy. In a standard IDE, the developer drives and the AI assists. In Antigravity, the paradigm flips.
Google Antigravity introduces a ‘Manager’ surface which functions “like a mission control for spawning, orchestrating, and observing multiple agents across multiple workspaces in parallel”. This allows a senior developer to theoretically manage multiple workstreams simultaneously; having one agent research a library update while another drafts a frontend component.
This aligns with the industry’s broader progression toward “agentic” workflows, but Google is explicitly targeting the “human-in-the-loop” friction that often hampers adoption. The goal is to reach a state where humans “interface with agents at higher abstractions over individual prompts and tool calls.”
A primary barrier to enterprise adoption of autonomous coding agents is the “black box” problem. If an AI model like Gemini 3 alters code without supervision, the risk of introducing subtle bugs or security vulnerabilities increases.
Google acknowledges that “most products today live in one of two extremes: either they show the user every single action and tool call the agent has made, or they only show the final code change with no context”. Neither approach builds the confidence required for production-level software.
To mitigate this, Antigravity utilises ‘Artifacts’ deliverables (e.g. task lists, implementation plans, and screenshots) that allow developers to verify the agent’s logic before the code is committed.
“Agents in Antigravity use Artifacts to communicate to the user that it understands what it is doing and that it is thoroughly verifying its work,” explains Google.
For compliance-heavy industries like finance or healthcare, this audit trail is non-negotiable. The ability to review an implementation plan before code generation could save hours of remediation during code review.
The platform also supports a feedback loop where users can comment on these artifacts. “This feedback will be automatically incorporated into the agent’s execution without requiring you to stop the agent’s process,” states Google. This helps to streamline the correction process which is often cumbersome in chat-based interfaces.
Developers aren’t locked to Google’s Gemini AI models
In a surprising decision that may appeal to enterprises wary of vendor lock-in, Antigravity is not exclusively bound to Google’s own models. The platform offers access to “Google’s Gemini 3, Anthropic’s Claude Sonnet 4.5 models, and OpenAI’s GPT-OSS within the agent, offering developers model optionality.”
This flexibility allows engineering teams to select the model best suited for a specific language or task, rather than forcing a one-size-fits-all approach. It also suggests Google is positioning Antigravity as a neutral workflow layer rather than just a delivery mechanism for Gemini.
The platform also integrates with existing developer tools. Gemini 3 Pro “can be utilised via your favorite developer tools within the broader ecosystem,” and integration is already underway with products like Cursor, GitHub, JetBrains, Manus, and Cline.
Despite the high performance on benchmarks, IT leaders should temper expectations regarding full autonomy. Google admits we are not yet ready “for days at a time without intervention” regarding agent operation. The technology is getting closer, but it remains an assisted workflow rather than a replacement for skilled engineers.
The introduction of vibe coding, where natural language is the only syntax you need, is marketed towards rapid prototyping.
According to Google, its Gemini 3 model is capable of translating a high-level idea into a fully interactive app with a single prompt. While this lowers the barrier to entry for non-technical staff to create internal tools or dashboards, it creates a new governance challenge: ensuring these rapidly generated applications adhere to security standards and data privacy regulations.
The release of Gemini 3 and Antigravity suggests that the efficiency gains from AI will increasingly come from parallel execution rather than just faster typing. By decoupling the developer from the immediate execution of code, teams could theoretically scale their output without linearly scaling headcount.
Yet, this requires a change in how engineering is managed. The skillset transfers from syntax proficiency to system design and verification.
“We want Antigravity to be the home base for software development in the era of agents,” Google says. Whether it achieves that status will depend on how well it integrates with the messy and often complex reality of enterprise legacy code, not just greenfield projects.
For now, the technology is available to test. Google Antigravity is available in public preview at no charge and Gemini 3 Pro is accessible via Google AI Studio. Enterprise leaders should encourage their R&D teams to pilot these tools, specifically to evaluate how the ‘Manager’ workflow impacts development velocity and code quality in a controlled environment.
See also: Ada: Should developers revisit the veteran programming language?
Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security Expo. Click here for more information.
Developer is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.