Ado Kukic is a software engineer and developer educator at Anthropic who helps programmers work effectively with Claude and Claude Code. His career charts the evolution of coding assistants from autocomplete tools into systems capable of investigating repositories, editing multiple files, executing commands, and checking their own work under human supervision.
Kukic developed his approach to technical education across Auth0, MongoDB, DigitalOcean, and Sourcegraph. His writing at Auth0 focused on authentication, security, and practical web development; he subsequently held developer-advocacy roles at MongoDB and DigitalOcean before becoming Sourcegraph’s director of developer relations.
At Sourcegraph, he helped developers understand Cody, the company’s code-aware assistant. His six-level framework for coding automation distinguishes predictive completions, repository-aware assistance, requirements-driven feature development, proactive maintenance, and increasingly autonomous software delivery. In 2025, he described Cody’s agentic context gathering across repositories, terminals, and connected tools, alongside context-aware editing that follows changes across files and responds to editor diagnostics.
At Anthropic, Kukic applies these ideas to Claude Code. His Advent of Claude guide collects 31 practical lessons on project instructions, skills, subagents, plugins, and browser integration; his Bash guide for AI engineers teaches developers to recognize destructive commands and inappropriate privilege escalation.
- Code AI maturity model: Coding tools advance through six levels as repository awareness increases, software-development tasks become more autonomous, and responsibility shifts from writing code toward reviewing results.
- Codebase-aware AI assistance: Useful suggestions must reflect a project’s actual dependencies, conventions, and architecture; plausible generic code can still be wrong for the repository.
- Agentic context retrieval: Effective assistants inspect files, search code, consult documentation, and gather relevant context before proposing changes, with terminal actions remaining subject to approval.
- Plan, act, and verify: Capable agents investigate a problem, make changes, run tests, and revise failures; developers retain responsibility for permissions, project-specific instructions, and final outcomes.
