Agentic Development Environment Setup

June 29, 2026 ยท View on GitHub

Recommendations for setting up your agentic engineering dev environment

1 Foundational Engineering Setup

For any target technology.

Model

The best models are the latest Opus and Codex, with effort High and biggest context available (validated through Opus 4.7 1M and Codex 5.5 High).

Opus is best for:

  • UI Design. Less mess, looks better.
  • Functional design; "getting" your core intent from functional prompts and creating plans for that

Codex is best for:

  • Logic - both writing and codebase analysis
  • Edge cases - misses a lot less that Opus, up to the point of overengineering

Other models & compute

The frontier models are barely good enough for sustainable agentic engineering, and they still need a lot of tools and human expertise and assist to deliver that. This is hard enough to do well as it is; don't bother wasting time with lesser models - yet.

At some point the frontier model labs are going to stop heavily subsidizing tokens, and depending on your budget that is the moment to evaluate the state of OSS models and running those on your own hardware - both are improving rapidly, but still lag behind frontier model labs.

Harness

The harness choice has just as much impact on the quality of the output as the model choice. For best results, use the CLI harness of the model maker (Claude Code CLI / Codex CLI). The brand owned CLI remains the 1st class harness for each company; it receives earliest updates and fixes, has the most features, and is tested most heavily (because agents can use and test CLI's much better than apps).

Cost is also a factor that favors the brand-owned harnesses: they benefit from the heaviest token-subsidizing subscriptions. API tokens used by 3rd party harnesses are much more expensive.

See:

Claude Code Get Started

Codex CLI Setup

Enable automatic approvals

Enabling automatic approvals is probably the single most important tip to increase your productivity.

  • Use the /permissions command in Codex CLI to configure Approve for me
  • In Claude Code CLI, turn auto mode on (shift+tab to cycle)

OS

Mac, linux or WSL on Windows works best - the harnesses 1st class shell is bash, as is the bulk of the model's training data. Even if a harness understands another shell well, such as Claude understanding PowerShell, it makes a lot more mistakes using that shell on Windows due to OS-specific differences.

IDE

You mostly need a UI that has good UX with git on local changes, commits and branches. Git worktree support is useful for parallel agent workstreams.

IDE functionality for reading and navigating code quickly and easily is useful to understand / validate the implementation architecture.

Actual manual editing and debugging is more exception than rule in agentic engineering, but you do need it when the agents don't cut it.

VS Code and it's plugin ecosystem works well for all of the above, for many technologies including markdown, web, dotnet, C#, PowerShell and bash. VS Code keeps pace with the latest models and agentic engineering practices (weekly release cycle).

GitHub

Install the gh CLI - agentic-check requires it for it's skill command.

2 .NET Engineering Setup

Prerequisites:

Recommended:

3 Cross-platform UI with Uno Platform Engineering Setup

Prerequisites:

  • Foundation Setup You need both Opus and Codex for UI with Uno platform - neither alone is capable enough yet, no matter how much tools and guidance you provide (validated through Opus 4.7 1M and Codex 5.5 High).
  • .NET Setup

Required:

Which model for what

  • For anything involving UI markup: use latest Claude Code CLI with Opus 1M High effort; no matter what you tell & give Codex, it often makes a visual mess when creating UX. Validated on Codex 5.5 High effort and older.
  • If Claude gets stuck in complex UI logic issues (e.g. how to use a complex UI library like LiveCharts2), latest Codex CLI High can get you unstuck.