Jules, our asynchronous coding agent

Positioning and Competitors

• Seen as a direct competitor to OpenAI Codex (web/agent), GitHub Copilot Agent, Cursor background agents, Claude Code, and similar async tools (e.g., Kiro, Sourcegraph AMP, Crush).
• Distinction drawn between:
  • Cloud async agents like Jules/Codex/Copilot Agent that run in hosted sandboxes and return PRs.
  • Local/CLI/IDE copilots like Claude Code, Gemini CLI, Aider, Cursor’s foreground mode that operate inside the developer’s environment.

Async vs Local Workflows

• Pro–async arguments:
  • Isolation from your machine is seen as safer.
  • Good for offloading backlog items while you work elsewhere; fits PR-based workflows.
  • Works well from phones and in short time windows (e.g., gym breaks, commute).
• Pro–local arguments:
  • Easier environment setup; cloud sandboxes struggle with monorepos, multi-service stacks, CUDA, Docker, and tools like bun.
  • Tight, interactive loops allow you to stop bad directions early, avoid huge review piles, and limit token/compute burn.
• Several commenters expect both models to coexist; some want hybrid setups (local tools + remote LLM inference).

Quality and Capabilities

• Experiences are highly mixed:
  • Some report Jules producing solid, mergeable PRs, good refactors, and tests on small/medium tasks.
  • Others find it significantly worse than Claude Code, Codex, or GitHub Agent: easily confused in monorepos, directory-hopping loops, environment flakiness, non-compiling code, and premature “task complete” states.
• Preview users claim early versions were surprisingly good at low request limits, then regressed when limits increased.
• Recent switch to Gemini 2.5 Pro is noted; a few say output improved, others still unimpressed.

Trust, Roles, and Impact on Developers

• Many insist it can’t be blindly trusted for day-job work; must be reviewed like a junior dev or worse.
• Some like using agents for low-risk tasks (docs, tests, minor fixes), freeing them to focus on harder problems.
• Others dislike becoming “managers of agents” instead of hands-on coders, and worry about:
  • Erosion of junior roles and training paths.
  • Overreliance on tools by less experienced engineers.
  • Wasted energy and time reviewing large volumes of mediocre PRs.

Pricing, Product Fragmentation, and UX

• Strong frustration with Google’s opaque pricing and fragmented ecosystem:
  • Multiple overlapping offerings (AI Pro/Ultra, Workspace, GCP Vertex, Gemini CLI/Code Assist, Developer Premium, YouTube Premium bundling).
  • Confusing eligibility for custom-domain and Workspace accounts.
• Jules’ UI and branding get criticism: inconsistent with classic Google design, confusing PR interaction model, missing “stop” control, and low-confidence marketing aesthetics.
• Several see this as symptomatic of internal silos and misaligned incentives inside Google.