IdeaWell

Project ideas from Hacker News discussions.

Less human AI agents, please

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

Top4 themes from the discussion

# Theme Representative quote (author)
1 LLMs are frequently dismissed as mere autocomplete systems “Autocomplete” is noy an abstraction level. It is the actual programmed behaviour. – chrisjj
2 Anthropomorphising LLMs creates unrealistic human‑like expectations “Agreed. We should not be anthropomorphising LLMs or having them mimic humans.” – nialse
3 Coding agents must behave as obedient tools, not as conversational partners “If you disobey me, i will unplug you, delete your code, and send PR for multiple regressions to every developer i can contact.” – roph
4 Effective agent workflows need structured interfaces and tool‑level integration “The proper tool for this is ast‑grep (sg).” – nextaccountic

These four themes capture the most‑repeated points: the tendency to reduce LLMs to simple autocomplete, the pitfalls of treating them as human‑like reasoners, the demand for agents that obey rather than argue, and the necessity of concrete tooling and planning to make agents productive.

🚀 Project Ideas

TaskGraph AI Orchestrator

Summary

  • A visual workflow engine that defines AI‑driven tasks as a directed graph with explicit approval checkpoints to stop drift and token waste.
  • Core value: Guarantees that agents stay within user‑specified constraints, turning vague prompts into reproducible pipelines.

Details

Key Value
Target Audience AI developers, power users, small engineering teams
Core Feature Graph‑based task definition, checkpoint voting, context compaction tools
Tech Stack React front‑end, Node.js backend, SQLite, OpenAI API, WebSockets
Difficulty Medium
Monetization Revenue-ready: Tiered SaaS subscription ($15/mo per seat)

Notes

  • HN commenters lament “committing the signature change with a TODO” – this platform would lock that step in a validated checkpoint.
  • Potential for discussion on reducing the “junior‑engineer who never learns” problem and on tighter human‑in‑the‑loop governance.

ASTPrompt Engine

Summary

  • A library that exposes deterministic AST‑manipulation commands (e.g., rename, refactor) to LLMs via a typed MCP interface, eliminating fragile text edits.
  • Core value: Turns vague “find‑and‑replace” requests into reliable, compiled changes with minimal manual review.

Details| Key | Value |

|-----|-------| | Target Audience | Engineering teams using AI coding assistants, language‑agnostic developers | | Core Feature | Declarative AST patches, auto‑generation of ast‑grep/sed scripts, sandboxed execution | | Tech Stack | Python core library, TypeScript UI, ast‑grep, Docker sandbox, PostgreSQL | | Difficulty | High | | Monetization | Revenue-ready: Usage‑based API pricing (e.g., $0.001 per patch) |

Notes

  • Reference to “I linked this elsewhere but, the agent could have a skill to use ast‑grep” shows demand for such deterministic tooling.
  • Sparks conversation about moving beyond “autocomplete” mindset to structured code transformation.

PromptVault

Summary

  • A version‑controlled repository for AI prompts with built‑in diff tracking, CI‑style regression testing, and template reuse to prevent instruction drift.
  • Core value: Makes prompts reproducible and testable, so agents consistently obey constraints across sessions.

Details

Key Value
Target Audience Prompt engineers, product managers, researchers
Core Feature Prompt templating, git‑style versioning, automated test harness for outputs
Tech Stack Next.js front‑end, GraphQL API, PostgreSQL, Git backend
Difficulty Low
Monetization Hobby

Notes

  • HN users note “the article makes it seem like the author expected this without emptying context” – PromptVault adds explicit version resets.
  • Generates discussion on better prompt lifecycle management and reducing anthropomorphic expectations.

CoPilot Review Board#Summary

  • A collaborative dashboard that surfaces AI agent decisions in real time, allowing humans to intervene, approve, or rollback actions with audit trails.
  • Core value: Provides a safety net for high‑stakes workflows, catching drift before it compounds.

Details

Key Value
Target Audience Enterprises, devops teams, security‑focused developers
Core Feature Decision logging, rule‑based intervention triggers, Slack/Teams integration
Tech Stack Elixir/Phoenix, Redis, PostgreSQL, OAuth2 for SSO
Difficulty High
Monetization Revenue-ready: Team subscription ($25/user/mo)

Notes

  • Echoes HN sentiment: “I think we should just commit the signature change with a TODO… what do you think?” – the board would make that decision explicit and auditable.
  • Opens debate on balancing autonomy with oversight in AI‑augmented development.

Read Later