A live coding demo shows how iterative, stateless prompts fail to build robust features, highlighting the need for a persistent specification.
LLMs struggle with overly large contexts, sub-token character manipulation, and math, and they process prompts sequentially, which impacts how they interpret constraints.
Improve LLM outputs by using structured formatting like XML tags, framing tasks with constraints upfront, and using a 'say' operation to guide the model's thought process.
In traditional development, code becomes the source of truth, coupling the 'what' with the 'how' and making the system brittle as requirements, bugs, and framework changes accumulate.
A spec-centric approach makes the specification—combining requirements and tests—the long-lived source of truth from which disposable code is generated.
AI-native development creates adaptable software through context, enables autonomous maintenance for security and performance, and makes development more inclusive for non-coders.
An overview of the Tessl platform, the AI Native Dev community, and a landscape of tools that are shaping the future of specification-driven development.
47:35Laurie Voss, Rey Bango, Hannah Foxwell, Rizel Scarlett & Thomas Steiner
43:05Kenneth Auchenberg & Christian Heilmann
25:17Scott Hanselman
59:16Ryan J Salva
58:06Anna Fritsch-Weninger
32:26Krzysztof Cieślak
30:24Markus Walker
35:01Angie Jones, Kent C Dobbs, Liran Tal & Chris Heilmann
Jobs that call for the skills explored in this talk.
Speech Processing Solutions
Vienna, Austria
