IdeaWell

Project ideas from Hacker News discussions.

Plan mode is dead

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

Theme 1 – Planning mode adds useful structure and iteration
Many commenters say a dedicated planning step lets them separate design from implementation, refine ideas, and keep complex projects on track.

“I use Claude's Plan mode daily and it's great. I almost always have feedback to refine the plan and I want a clean separation of planning time before writing code.” – dbbk
“I ask it to give me a plan, as a living document, and we spend a good hour or two iterating over the plan.” – spacedcowboy

Theme 2 – Newer models make explicit plan mode unnecessary
Several users argue that modern LLMs can follow conversational instructions well enough that a separate “plan mode” is just a prompt reminder and has become redundant.

“The real reason why plan mode is dead is because you can just conversationally instruct the agent to not make changes to the repository … and it will listen.” – tcdent
“Claude code’s plan mode began its slow march towards deprecation when they hid the ‘clear context and implement’ option behind a flag.” – nojs
“Plan mode came about because earlier models were loose cannons… Today's models follow instructions better enough to not need a separate mode.” – esafak

Theme 3 – Desire for better collaboration, auditability, and flexibility
Some find generated plans hard to review or lacking an audit trail, and they prefer more flexible workflows (sub‑agents, external tools, or iterative conversation) that keep them closely involved.

“The plan can be detailed and technically correct, but still be hard to review because the important bits are buried and feel distant from my own thinking.” – aymandfire
“It also doesn't allow leaving an audit trail of plans and decisions (by default, anyway).” – throwaway27448
“I'd like a brainstorming mode, with limited side effects… I want it to go and read stuff but also invoke some commands and generate reports.” – 6thbit

🚀 Project Ideas

Generating project ideas…

PlanLens: AI‑Generated Plan Review Assistant

Summary

  • PlanLens parses Claude Code plan mode outputs and surfaces the key assumptions, trade‑offs, and decision points in a skimmable UI, addressing the reviewer fatigue described by users who find generated plans “hard to review because the important bits are buried.”
  • Core value: lets developers validate AI‑generated plans against their intent in minutes instead of hours, turning plan mode from a bottleneck into a collaborative checkpoint.

Details

Key Value
Target Audience Developers using Claude Code (or similar LLM coding assistants) who rely on plan mode for complex projects
Core Feature Automatic extraction of assumptions, risks, and open questions from plan documents, with inline highlighting and comment‑able annotations
Tech Stack Python/FastAPI backend, React frontend, LLM‑based summarization (local Llama 3 or OpenAI API)
Difficulty Medium
Monetization Revenue-ready: Subscription $12/mo per seat (team plans available)

Notes

  • HN commenters would love it because “The plan can be detailed and technically correct, but still be hard to review because the important bits are buried” (aymandfire) and they want a way to “get into flow state and carefully check them.”
  • Potential for discussion: could spark new workflows where plan reviews become explicit, auditable steps in AI‑assisted development, and could be extended to other LLM agents.

BrainstormBuddy: Safe Exploration Harness for LLMs

Summary

  • BrainstormBuddy is a lightweight harness that enables a true “brainstorming mode” for LLMs: it allows the model to read files, run safe commands, generate reports, and discuss results without making any persistent code changes unless explicitly approved.
  • Core value: satisfies the desire expressed by users who want a mode that goes and reads stuff, invokes some commands, and discusses results, but does not default to writing plans or code every turn.

Details

Key Value
Target Audience Power users of LLM coding assistants who need exploratory, low‑risk experimentation (e.g., researchers, open‑source maintainers)
Core Feature Sandboxed execution environment with configurable side‑effect whitelist, automatic logging of actions, and a simple CLI/chat interface to trigger brainstorming sessions
Tech Stack Docker for isolation, Go or Rust agent controller, TypeScript/React optional web UI, integrates via MCP or custom prompt wrappers
Difficulty High (due to sandboxing and security considerations)
Monetization Revenue-ready: Usage‑based pricing $0.005 per 1K tokens of brainstorming session + $5/mo base fee

Notes

  • HN commenters would love it because users asked for “a brainstorming mode, with limited side effects… I want it to go and read stuff but also invoke some commands and generate reports and discuss on the results” (6thbit) and felt current plan mode always goes to write a plan.
  • Potential for discussion: could become a standard tool for exploratory AI work, reducing the temptation to over‑plan and encouraging rapid, safe iteration.

PlanVault: Version‑Controlled AI Plan Repository

Summary

  • PlanVault provides a version‑controlled repository for AI‑generated plans, complete with diffing, commenting, and approval workflows, giving teams an audit trail of planning decisions that currently disappears when plan mode is turned off.
  • Core value: turns ephemeral plan artifacts into persistent, collaborative assets, addressing the lack of an audit trail highlighted by commenters who miss “leaving an audit trail of plans and decisions.”

Details

Key Value
Target Audience Teams that rely on LLMs for software design (e.g., dev shops, startups) and want traceability of AI‑assisted planning
Core Feature Git‑backed store for plan files, automatic PR‑like review requests when a new plan is generated, inline commenting, and integration with Claude Code via a hook that saves plan output to the vault
Tech Stack Node.js/NestJS backend, PostgreSQL for metadata, Git storage (bare repos), React frontend, webhooks for Claude Code
Difficulty Medium
Monetization Revenue-ready: Team plan $20/mo (unlimited repos) + $5 per active user

Notes

  • HN commenters would love it because one noted “It also doesn't allow leaving an audit trail of plans and decisions (by default, anyway)” (throwaway27448) and others wished for a way to “point the plan in the right direction, then have it iterate with a smarter reviewer.”
  • Potential for discussion: could revive the usefulness of plan mode by making plans reviewable and traceable, fostering clearer human‑AI collaboration and enabling retrospective analysis of AI design choices.

Read Later