Associate Manager at a tech vendor with 10,001+ employees
Real User
Top 5
Jul 10, 2026
My experience with Amazon Q has been positive, but I think it would be better if it improved in some areas. Sometimes the generated code does not align with my project's existing coding standards or architecture, requiring me to manually review and refine it. While it helps with logic, sometimes the coding standards do not match, meaning I need to adjust it again. Additionally, I wish for more accurate suggestions for complex, project-specific codebases and clearer explanations for why particular solutions are recommended. If I had to select one priority, it would be better to have context awareness for large codebases and development sessions, making suggestions even more relevant and reducing the need to repeat context.
One improvement for Amazon Q is that I use it in Visual Studio, and in Visual Studio, I am not given an option to upload an image in Amazon Q. Also, this is one part of it. The second part is the context window of Amazon Q is very less compared to other GenAI tools. The moment I would be in a deep research or deep development or deep debugging mode in Amazon Q, the moment I hit the context length of the window, it would ask me to clear the complete context, and it would lose the complete context of the chat that I had previously. The two major pain points are that I have Amazon Q in Visual Studio, but I am not given an option to upload an image as a reference in Amazon Q. The second part is the context window is so limited. The moment I deep dive into some discussion in terms of development or debugging or automation, I hit a context length of the window, and the moment I hit it, it would lose the complete context. It has an option of summarizing the complete context and having it as a memory, but it would not be sufficient because I would have given a lot of details in the chat by that time. I have mentioned the two points earlier, so those are the only two points that I have in mind for improvements needed for Amazon Q.
One most important improvement would be having deeper domain-specific intelligence for Amazon Q. While it can pull enterprise data sources, adding more domain-tuned models, for example in pharma, would be beneficial so the answers are not that generic. I believe it can be more explainable and proactive. When it comes to the data itself, we can have stronger multi-data reasoning. Additionally, deeper developer workflow automation could improve the user experience.
It's very hard for me to comment because I've just been using it for three to five months, but what I could say is if they could somehow generate or collaborate with OpenAI, that would be a very big plus. They could improve more. I'm a person who believes that everybody is imperfect, so for me, nobody is a 10 on 10. The maximum they can reach is nine, so that's the thought process I go by.
Innovation Strategist at a insurance company with 5,001-10,000 employees
Real User
Top 20
Aug 7, 2025
When updates occur in Anthropic models after implementing guard rails or instruction-based flows, we must revisit everything from scratch due to hallucination issues or unexpected outputs. Additionally, Amazon Q sometimes fails to follow specific output format instructions, instead providing summarized information in a different format. The multimodal capabilities of Amazon Q are limited, lacking the ability to process or generate images. When attempting to generate enterprise-level reports requiring both text and images, it struggles to combine elements in the specified format despite clear instructions. The integration costs are considerable, making the pricing structure less feasible for scalable solutions. After successful POCs, scaling at an enterprise level becomes cost-prohibitive. Developers have reported that Amazon Q plugins in IDEs consume significant CPU resources, necessitating high-end hardware that may not be available to all employees. Data analytics and querying capabilities show limitations, particularly when handling Excel files and performing operations requiring accurate, non-hallucinated output. Advanced code generation features are lacking, including the inability to generate comprehensive unit tests or handle complex refactoring across multiple files. The platform doesn't automatically identify security or performance bugs in code. Data cleaning and ETL tools are insufficient for enterprise-level analytics or visualization. Amazon Q struggles with parsing tables and understanding complex documents containing combinations of charts, numbers, and text. There is also no session awareness, meaning context is lost when sessions are closed and reopened. Downtime has been minimal, with one notable instance last year when the LLMs were not functioning. The main ongoing challenge is managing updates from AWS or Anthropic, as there is no proactive solution to assess impact on production deployments before implementation.
Amazon Q could be improved if it could generate full functions for certain use cases. While it did not provide all the answers in some cases, it still accomplished a great deal of work for us. I was expecting it to function more as a co-pilot for AWS environment, but despite not reaching that level, it still proved highly useful.
Senior Software Engineer at a tech vendor with 1,001-5,000 employees
Real User
Top 5
Aug 5, 2025
In terms of whether Amazon Q is overall easy to use, I feel it is somewhat the same in terms of usage. The UI is friendly, but there are a few things I face issues with, such as the number of tabs you can use. In the UI version, you can only use a maximum of ten tabs at a time, which is a limitation I encounter. Also, when scanning a code base with many files, it stops and gives a message that the context is too large, which I feel is a restriction. However, there are great feature points, such as being able to select a code snippet to fix or explain directly. These edge points are things that I feel are not available in other AI tools. When it comes to issues with the stability of Amazon Q, I have not experienced bugs per se, but if I use multiple tabs, I feel it does not manage the context between them. For example, if I start with a prompt in one tab and then try to continue in another, it does not retain that context. In instances where I use only a single tab, it seems to have memory issues, so I need to stop and give the same context again in a new tab to resolve the issue. Also, when scanning a large number of files, there are limitations there.
Site reliability engineer at a tech services company with 201-500 employees
Real User
Top 20
Jul 16, 2025
I currently use the free tier, and in all my use cases, it's been able to deliver. Even though machine learning models have errors, regarding its benchmark, it's able to perform well. The paid version would likely do more compared to the free version. Sometimes feedback is needed immediately. It takes a bit of time because there is a workload. You'll get your feedback and a listening ear, but it can improve in this aspect.
Amazon Q provides context-aware responses and integrates seamlessly with AWS, supporting efficient cloud task management, multi-language frameworks, and documentation capabilities. It's an asset for diverse development needs with auto-logging, intuitive interfaces, and fast deployment.Amazon Q offers advanced natural language interpretation, enriching productivity with robust features like Git-related insights for tracking code changes and built-in redundancy. It supports multi-language...
My experience with Amazon Q has been positive, but I think it would be better if it improved in some areas. Sometimes the generated code does not align with my project's existing coding standards or architecture, requiring me to manually review and refine it. While it helps with logic, sometimes the coding standards do not match, meaning I need to adjust it again. Additionally, I wish for more accurate suggestions for complex, project-specific codebases and clearer explanations for why particular solutions are recommended. If I had to select one priority, it would be better to have context awareness for large codebases and development sessions, making suggestions even more relevant and reducing the need to repeat context.
One improvement for Amazon Q is that I use it in Visual Studio, and in Visual Studio, I am not given an option to upload an image in Amazon Q. Also, this is one part of it. The second part is the context window of Amazon Q is very less compared to other GenAI tools. The moment I would be in a deep research or deep development or deep debugging mode in Amazon Q, the moment I hit the context length of the window, it would ask me to clear the complete context, and it would lose the complete context of the chat that I had previously. The two major pain points are that I have Amazon Q in Visual Studio, but I am not given an option to upload an image as a reference in Amazon Q. The second part is the context window is so limited. The moment I deep dive into some discussion in terms of development or debugging or automation, I hit a context length of the window, and the moment I hit it, it would lose the complete context. It has an option of summarizing the complete context and having it as a memory, but it would not be sufficient because I would have given a lot of details in the chat by that time. I have mentioned the two points earlier, so those are the only two points that I have in mind for improvements needed for Amazon Q.
While using Amazon Q, I faced some challenges, such as navigating the interface initially.
One most important improvement would be having deeper domain-specific intelligence for Amazon Q. While it can pull enterprise data sources, adding more domain-tuned models, for example in pharma, would be beneficial so the answers are not that generic. I believe it can be more explainable and proactive. When it comes to the data itself, we can have stronger multi-data reasoning. Additionally, deeper developer workflow automation could improve the user experience.
It's very hard for me to comment because I've just been using it for three to five months, but what I could say is if they could somehow generate or collaborate with OpenAI, that would be a very big plus. They could improve more. I'm a person who believes that everybody is imperfect, so for me, nobody is a 10 on 10. The maximum they can reach is nine, so that's the thought process I go by.
When updates occur in Anthropic models after implementing guard rails or instruction-based flows, we must revisit everything from scratch due to hallucination issues or unexpected outputs. Additionally, Amazon Q sometimes fails to follow specific output format instructions, instead providing summarized information in a different format. The multimodal capabilities of Amazon Q are limited, lacking the ability to process or generate images. When attempting to generate enterprise-level reports requiring both text and images, it struggles to combine elements in the specified format despite clear instructions. The integration costs are considerable, making the pricing structure less feasible for scalable solutions. After successful POCs, scaling at an enterprise level becomes cost-prohibitive. Developers have reported that Amazon Q plugins in IDEs consume significant CPU resources, necessitating high-end hardware that may not be available to all employees. Data analytics and querying capabilities show limitations, particularly when handling Excel files and performing operations requiring accurate, non-hallucinated output. Advanced code generation features are lacking, including the inability to generate comprehensive unit tests or handle complex refactoring across multiple files. The platform doesn't automatically identify security or performance bugs in code. Data cleaning and ETL tools are insufficient for enterprise-level analytics or visualization. Amazon Q struggles with parsing tables and understanding complex documents containing combinations of charts, numbers, and text. There is also no session awareness, meaning context is lost when sessions are closed and reopened. Downtime has been minimal, with one notable instance last year when the LLMs were not functioning. The main ongoing challenge is managing updates from AWS or Anthropic, as there is no proactive solution to assess impact on production deployments before implementation.
Amazon Q could be improved if it could generate full functions for certain use cases. While it did not provide all the answers in some cases, it still accomplished a great deal of work for us. I was expecting it to function more as a co-pilot for AWS environment, but despite not reaching that level, it still proved highly useful.
In terms of whether Amazon Q is overall easy to use, I feel it is somewhat the same in terms of usage. The UI is friendly, but there are a few things I face issues with, such as the number of tabs you can use. In the UI version, you can only use a maximum of ten tabs at a time, which is a limitation I encounter. Also, when scanning a code base with many files, it stops and gives a message that the context is too large, which I feel is a restriction. However, there are great feature points, such as being able to select a code snippet to fix or explain directly. These edge points are things that I feel are not available in other AI tools. When it comes to issues with the stability of Amazon Q, I have not experienced bugs per se, but if I use multiple tabs, I feel it does not manage the context between them. For example, if I start with a prompt in one tab and then try to continue in another, it does not retain that context. In instances where I use only a single tab, it seems to have memory issues, so I need to stop and give the same context again in a new tab to resolve the issue. Also, when scanning a large number of files, there are limitations there.
I currently use the free tier, and in all my use cases, it's been able to deliver. Even though machine learning models have errors, regarding its benchmark, it's able to perform well. The paid version would likely do more compared to the free version. Sometimes feedback is needed immediately. It takes a bit of time because there is a workload. You'll get your feedback and a listening ear, but it can improve in this aspect.