Google is making the case that the rise of AI-generated code is changing the criteria developers use to assess programming languages, with readability, verification, and long-term maintenance becoming more important as coding agents take on more development work.
The company argues that Go is suited to this model because its language design and development tools were built around consistency across teams. Rather than focusing mainly on how quickly developers can write code, Google says Go was designed to make software easier to read, test, maintain, and update over time.
From code generation to code review
AI coding assistants can produce large amounts of syntactically valid code in a short period. Google argues that this reduces the importance of human coding speed while increasing the amount of generated code that developers need to review and verify.
Human developers also remain responsible for application architecture, boundaries between services, security controls, and the reliability of production systems.
Research into agent-generated software suggests that human involvement continues after the initial code is produced. A 2026 study accepted at the International Conference on Evaluation and Assessment in Software Engineering examined more than 1,000 AI-generated files and about 3,200 subsequent changes across 100 popular open-source repositories.
The researchers found that human developers performed most of the subsequent maintenance on AI-generated files. Feature extensions were the most common changes to agent-generated code, while bug fixes accounted for a larger share of changes to human-written files.
Another 2026 study analysed 278,790 code-review conversations across 300 open-source GitHub projects. It found that human reviewers went through 11.8% more review rounds when assessing AI-generated code than human-written contributions and provided more feedback on testing, code understanding, and knowledge transfer than AI reviewers did.
Google argues that this changes which characteristics become more important in programming languages and developer tooling. Languages and tooling must support not only code creation, but also testing, validation, maintenance, dependency management, and collaboration.
Developers also remain cautious about relying on AI output without checking it. Stack Overflow’s 2025 Developer Survey found that 46% of respondents somewhat or highly distrusted the accuracy of AI tools, compared with about 33% who expressed some level of trust. Only 3.1% said they highly trusted AI-generated output.
Security remains part of the review process. A 2026 qualitative study of 15 professional developers found that participants did not initially include security requirements in their prompts during observed coding sessions, even when they had relevant security knowledge. The researchers said AI-assisted development shifted some security attention from code creation to later review.
Go’s focus on predictability
Go was developed at Google by Robert Griesemer, Rob Pike, and Ken Thompson, with its design shaped around software engineering rather than programming alone. Its creators sought to provide consistent ways for teams to structure and maintain software instead of adding multiple mechanisms for expressing similar logic.
Google argues that languages offering multiple ways to express the same logic can introduce more variation into AI-generated code, while Go’s conventions limit some of that variation. It points to standard formatting, a relatively small language specification, and widely used project conventions as ways to reduce differences in structure and style.
Formatting is one example. Go includes gofmt, which automatically applies a standard format to source code rather than leaving formatting conventions to individual developers or projects.
The reasoning behind gofmt predates generative AI. The Go team has previously said that automatic formatting removes stylistic arguments and makes source code easier to read because developers do not need to adjust to different conventions between projects.
Standard formatting can also make large code changes easier to review by keeping formatting differences from obscuring the underlying source changes. Google is now applying that principle to AI-generated code.
Google’s argument extends beyond syntax. Go ships with a common toolchain covering formatting, testing, package management, vulnerability checking, profiling, and code analysis.
That tooling can also be used by coding agents during development. An agent can generate code, compile it, run tests, identify failures, and revise the implementation before handing the result to a developer.
Go’s static type system provides another layer of validation. The compiler can reject code containing issues such as incorrect types or references to methods that do not exist, giving an AI agent concrete errors it can use during subsequent attempts.
The connection between static typing and AI-assisted development is not limited to Google’s argument for Go. GitHub has also pointed to type information as useful when coding models generate software because types constrain which operations are valid for variables, functions, and interfaces.
GitHub has cautioned that passing type checks does not establish that an application’s business logic is correct. Testing and human review are still required to determine whether the software behaves as intended.
Google also points to Go’s compilation speed as useful because agents can repeatedly run compile-and-correct cycles. The company argues that external validation is important because repeated AI-generated changes without testing can introduce additional errors.
Third-party packages are another area Google identifies as relevant to AI-generated software. Coding models can recommend dependencies based on information in their training data, including packages that have become outdated or are no longer maintained.
Go’s standard library covers areas including networking, cryptography, HTTP, text processing, testing, and file operations, reducing the need for external packages for some common tasks. Google argues that standard components can reduce the number of external dependencies that developers and coding agents need to manage.
The Go module system provides additional controls when external packages are required. Its checksum database records module hashes so downloaded dependencies can be checked against previously recorded versions, while the Go module mirror stores copies of modules.
Go also provides govulncheck, which can identify whether an application calls functions associated with known vulnerabilities. Its native testing framework and fuzzing support add further checks during development.
Maintaining code written by agents
Research on coding agents is beginning to distinguish between whether generated code works when first produced and whether developers or other agents can successfully modify it later.
A June 2026 study introduced a framework called CodeThread to test what happens when coding agents continue working on code previously written by another agent. Across four frontier coding agents, the researchers found lower task-resolution rates when agents extended agent-written code rather than human-written code, with declines reaching as much as 13.1% in some comparisons.
Conventional measures such as code complexity and verbosity did not fully explain the difference. The researchers instead identified differences in areas including input validation and error handling that affected how easily subsequent agents could build on the earlier implementation.
Google argues that compatibility and automated maintenance become more important as coding agents contribute additional changes over time. Go’s compatibility policy is intended to allow programs written for earlier Go 1 releases to continue working with newer versions of the toolchain.
The guarantee is not absolute. Go’s official compatibility documentation lists exceptions including security fixes, reliance on unspecified behaviour, changes in operating-system behaviour, and some uses of the unsafe package.
The platform also includes tools for updating existing code. The gopls language server supports code navigation and automated changes, while go fix includes modernisers that can replace older code patterns with newer Go conventions.
Recent changes to Go’s tooling make the connection with coding agents more direct. The gopls language server includes an MCP server that can expose compiler and analysis diagnostics to LLM-based coding agents.
The rebuilt go fix tooling also includes modernisers that replace older code patterns with newer language or standard-library constructs using predefined analysis and transformation rules. These tools give developers and coding agents a deterministic way to update parts of a codebase without relying entirely on generative rewrites.
Coding agents are also moving beyond autocomplete and isolated code suggestions. Research published in 2026 analysed 24,014 merged pull requests produced by coding agents, covering more than 440,000 commits, and compared them with 5,081 merged human-authored pull requests.
The researchers found differences between agent and human contributions in areas including commit counts, the number of files modified, and deleted lines.
Google Research has also examined broader expectations for software-engineering agents. Research based on developer-defined agent rules and interviews with experienced software engineers identified adherence to engineering standards, code quality, reliability, problem solving, and collaboration alongside task completion.
Google’s argument is that Go’s standard formatting, static checks, common tooling, and compatibility model already provide some of those controls when coding agents contribute to software projects.
See also: Google Cloud details full-stack AI architecture for developers
Want to dive deeper into the tools and frameworks shaping modern development? Check out the AI & Big Data Expo, taking place in Amsterdam, California, and London. Explore cutting-edge sessions on machine learning, data pipelines, and next-gen AI applications. The event is part of TechEx and co-located with other leading technology events. Click here for more information.
Developer Tech News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.
