An official survey of Go developers shows growing interest in building AI applications – but some frustration that Python dominates the space, with far more libraries available. Around a quarter of those surveyed are at organizations exploring or building AI applications, but over half of these cited “Ecosystem is Python-centric” as their biggest challenge.
These developers want to use Go for AI. “I think that the robustness, simplicity, performance and native binaries that the language offers make it a far stronger choice for AI workloads”, said one who is quoted in the survey results, based on responses submitted in January and February this year.
AI developers surveyed are most commonly building services for AI use cases such as text summarization (56 percent), text generation (55 percent), and chatbots (46 percent), often for internal use, though there may be plans for more public use if internal tests go well.
The dominance of Python is a blocker for some though. “[The] Go world is missing many AI libraries. If I have a LLM PyTorch model, I can’t even serve it (or I’m unaware how to do it). With Python it’s basically a few lines of code,” said another developer.
The biggest ask from these developers is for “Go equivalents to Python libraries”, followed by tools to migrate Python code to Go, and the ability to interop with Python.
The Go team invited submissions via the Go blog and via random prompts in the VS Code Go plug-in and in the JetBrains GoLand IDE, achieving 6,224 responses.
Go is a compiled language with static typing, garbage collection and memory safety. It was developed by Google, is well suited to concurrent programming, and performs well though typically not as fast as low-level languages such as C, C++ or Rust.
Performance, it turns out, matters a lot to Go developers. More than 70 percent regard it as at least moderately important, and for nearly 40 percent it is very or extremely important, according to the survey. However these developers do not find it straightforward, in general, to diagnose performance issues they encounter, with 30 percent finding it difficult and only 24 percent regarding it as easy.
This might be related to the fact that the biggest single challenge faced by developers is “learning how to write Go effectively”, cited by 15 percent of respondents, followed by verbosity of error handling at 13 percent.
VSCode is the most popular IDE (43 percent) for Go developers, followed by GoLand (33 percent) and Vim or Neovim (17 percent). These figures exclude those who came to the survey via a prompt in the IDE to avoid obvious bias. There is an indication that some developers migrate towards GoLand as they gain more experience and commitment to the language.
Go may have originated at Google, but for cloud deployments the respondents cited AWS as the most popular target (52 percent). GCP (Google Cloud Platform) is next (27 percent) with Microsoft Azure in third place at 13 percent. Not only is Microsoft’s cloud less popular; it is also less satisfactory for this crowd, who report 77 percent satisfaction with AWS and GCP, but only 57 percent for Azure.