So glad I found you finally, someone writing about the future. What happened? Are we standing up for this process and an enforcing sovereign local systems? Don't give your code bases away to a cloud providers - please. We have got to actively fight for sovereign systems with AI native infrastructures where we're calling design and code from a repository that we own. We cannot integrate with cloud systems and maintain regulatory oversight and control.
What happened is we're actively shipping this to customers. We have an open source version available here: https://www.inxm.ai/oss/inxm-local, and our proprietary platform for enterprises available for on prem use or private cloud use.
Very cool. Seems like the clouds are trying to capture these companies production process. Good to see they are building internally with smaller models. Thanks.
If your entire ai agent workflow can be compiled in to code then it's wrong use of AI agent, workflow could have been generated as usual software project using ai e.g claude code, instead of AI Agent. This point you already clarified in the end ""Compiled AI does not mean “no LLMs at runtime ever”" . Thanks for the article.
It depends. For a developer that might be true, but for non-developers this unlocks superpowers, being able to run deterministic code to automate tasks.
Different domains require different levels of non-determinism. The author's approach of bounding LLM invocations is interesting. I would only remark that Code Factory and BFCL sit at opposite ends of the paper's input spectrum and accordingly require different implementation techniques. My recent article on Medium describes the pure BFCL case for GCP landing zones: [https://medium.com/google-cloud/compile-time-ai-for-gcp-landing-zones-2555560fbd2f]. Cloud foundations belong to the structured end — multiple invocations with the same parameters have to produce identical results, and any change in the result has to have an unambiguous explanation.
So glad I found you finally, someone writing about the future. What happened? Are we standing up for this process and an enforcing sovereign local systems? Don't give your code bases away to a cloud providers - please. We have got to actively fight for sovereign systems with AI native infrastructures where we're calling design and code from a repository that we own. We cannot integrate with cloud systems and maintain regulatory oversight and control.
What happened is we're actively shipping this to customers. We have an open source version available here: https://www.inxm.ai/oss/inxm-local, and our proprietary platform for enterprises available for on prem use or private cloud use.
Very cool. Seems like the clouds are trying to capture these companies production process. Good to see they are building internally with smaller models. Thanks.
If your entire ai agent workflow can be compiled in to code then it's wrong use of AI agent, workflow could have been generated as usual software project using ai e.g claude code, instead of AI Agent. This point you already clarified in the end ""Compiled AI does not mean “no LLMs at runtime ever”" . Thanks for the article.
It depends. For a developer that might be true, but for non-developers this unlocks superpowers, being able to run deterministic code to automate tasks.
Different domains require different levels of non-determinism. The author's approach of bounding LLM invocations is interesting. I would only remark that Code Factory and BFCL sit at opposite ends of the paper's input spectrum and accordingly require different implementation techniques. My recent article on Medium describes the pure BFCL case for GCP landing zones: [https://medium.com/google-cloud/compile-time-ai-for-gcp-landing-zones-2555560fbd2f]. Cloud foundations belong to the structured end — multiple invocations with the same parameters have to produce identical results, and any change in the result has to have an unambiguous explanation.