I have mathematical proof and code running that shows six sigma code generation accuracy (99.99966%). I define correct code with:
1.) Compiles
2.) Passes AI and/or human code reviews
3.) Passes a 100% code coverage test harness
2 & 3 are intentionally non-deterministic in the sense that the customer can chose how these code reviews should occur and the contents of the the business/domain test portion of the test harness. I provide a rudimentary test harness that covers type safety, bounds checking, etc. similar to Parasoft Jtest. This will guarantee that 100% code coverage so the customer adds whatever business tests they need with no constraints. I've focused on the programmatic prompt applications where I can dodge issues on token based pricing. Accuracy at the interactive level doesn't sell well as it will cause lower use of tokens and lower revenue. I have an approved contract as an IBM partner but don't have an IBM representative yet. As soon as I d I'll share details on how this can be accomplished.
------------------------------
John Harby
CEO
Autonomic AI, LLC
Temecula CA
9513835000
------------------------------