AI coding assistants support development workflows by automating tedious context work. They complement context engineering, and they are not always the right tool: static refactoring can be more reliable and cheaper for changes that static analysis can infer. section 3.5 section 2.8.1
Package renaming
An AI prompt to remove “jboss” from package names and references missed some changes, leaving the application unable to compile. IDE refactoring completed the rename without breaking compilation. section 2.8.1
Table 9.2 Multi-tool strategy by workflow
|
Workflow
|
Tool
|
Why
|
Model
|
Cost
|
|
Interactive coding (in-IDE)
|
Cursor Pro
|
Best inline edit and agent mode UX; fast completions
|
Haiku (completions); Sonnet (agent mode)
|
$20/seat/month + modest overage
|
|
Autonomous refactoring
|
Claude Code (terminal)
|
Best at multi-file autonomous work; no IDE coupling
|
Sonnet (via API)
|
Pure API pricing; no subscription overhead
|
|
Tab completions only
|
GitHub Copilot (for the 3 devs who prefer it)
|
Fastest completions; flat-rate pricing; no overage risk
|
Copilot's internal model
|
$19/seat/month; unlimited
|
|
CI pipeline (code review, tests)
|
Direct Anthropic API
|
Full control over model, prompt, and batching
|
Haiku (first pass); Sonnet (deep review)
|
Pure per-token; batch API discount for async work
|
|
Local inference (offline/privacy)
|
Continue + Ollama
|
Zero API cost; works offline; sensitive codebases
|
Qwen 2.5 Coder
7B (Q4_K_M)
|
$0/month (hardware already owned)
|
Compare tools by workflow rather than treating AI tooling as a single system: interactive coding, autonomous refactoring, completions, CI, and offline use have different needs. section 9.2.7

























