Agents need good developer experience too (modal.com)

Modal argues that as coding agents take on more of the work, developer experience (DX) must evolve—but not be reinvented. Their core claim: the same principles that make humans productive—tight feedback loops, clear errors, runnable examples, unified infra and logic, and consistent naming—also make agents faster, less error-prone, and easier to hand off work to. This reframes the current inflection point (agents writing substantial code) as a stimulus to sharpen DX so both humans and agents can iterate quickly and reliably. Practically, Modal recommends technical changes: expose functionality programmatically (CLI → SDK/API) so agents can access logs, billing, and runtime details via text-based interfaces; produce actionable, high-signal error messages that agents can consume within limited context windows; place explanatory examples next to code to enable pattern-based learning; and keep infrastructure and application logic together (e.g., Python-based infra-as-code instead of scattered YAML) to enable static analysis, autocomplete, token efficiency, and fewer hallucinations. Consistent, intuitive naming (they changed concurrency_limit to max_containers) reduces mental load and token waste. The takeaway: designing for humans inherently prepares systems for agent collaborators, accelerating iteration and reducing stalls in agent-led development.