Developer Workflows in AI Age - Local or Cloud
The dotnet discussion looks at the decisions and development practices behind agentic workflows, with an emphasis on context, software fundamentals, and guardrails rather than code generation alone.
Choosing where and how agents work
The discussion examines several choices that shape AI-powered development:
- Local or cloud models: Where to run models, with privacy and cost as considerations.
- Agent harness and environment: Whether to work through an IDE, terminal, or browser, and which agent harness to use.
- Planning approach: Whether to specify project details upfront or build more ad hoc.
- Design-to-code workflow: How to bridge the gap between design and implementation.
Providing context and coordinating agents
Generating code is only one part of building software. Agents also need project knowledge and guidance to work within the intended development process. Topics include:
- Providing prescriptive guidance through Skills.
- Bringing grounded context into workflows through MCPs.
- Delegating tasks to sub-agents.
- Organizing a squad of agents with different responsibilities.
- Pitting agents against one another as part of the workflow.
Validating generated work
The conversation also considers how developers validate AI output, including human code review and architectural guidance. Its central point is that effective agentic workflows depend on context, fundamentals, and guardrails—not simply the ability to generate code.
The discussion connects these choices to real-world AI-powered workflows and the role of development tooling in supporting developers at different stages of AI adoption.