Ambient code is a paradigm where code is infinitely available, and generated on-demand. Planning for the age of code-abundance, several new and a few existing capabilities combine + stack to provide a unique and modern approach to the software development lifecycle.
Left to right: ground truth –> software factory –> better, cheaper code
What’s unique here:
- Business, test and applications specifications as inputs, including SLOs.
- Test driven development coupled with LLMs, means we reap the benefits and control of TDD in a structured, repeatable, enforceable way.
- Pattern learning, where specially crafted small language models identify, observe, collect, recommend, implement, progressively deliver and rollback improvements as necessary, safely, within business constraints and without violating SLOs.
- Artifact caching research, where LLMs proactively search catalogs (signed, hashed fingerprints) of blessed artifacts (functions, libraries) during the process of code generation. This leads to lower development and maintenance costs. LLMs could be taught to be CDNs for software factories.
- Orchestrator, the policy engine that governs the factory, including pattern-learning LLMs and risk modulation.
- Code that “can’t get worse”. Sets of TDD and SDD agents that proactively deploy SLMs like “the mRNA of software” to solve problems.
What are the headwinds for ambient code…
- Trust – we don’t really trust this stuff yet. It’s hard to control and quantify. Trust but verify.
- Usage of TDD or SDD “in anger” is rare. Many codegen platforms are starting to realize these are the keys to economically viable virtual development teams.
- How can these concepts benefit and accelerate brownfield codebases?
- How will this environment be supported?
You must be logged in to post a comment.