IdeaWell

Project ideas from Hacker News discussions.

AI coding agents generate more code, but not more software

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

1. Human code review remains valued and hard to abandon
Many commenters argue that senior engineers resist giving up line‑by‑line review, seeing it as essential for quality and knowledge transfer.

“We've had discussions about leaning more on automated review, but it hasn't translated into any practical improvements so far. I think senior engineers find it very difficult to give up on line‑by‑line review, no matter how much better the agents have gotten.” – staminade

2. AI‑accelerated coding shifts the bottleneck to review, testing, and analysis
While agents produce code faster, the process now stalls on reviewing that code, writing useful tests, or clarifying requirements.

“Code review was always a significant bottleneck and is even more so with agentic coding.” – staminade
“the baseline mean lines of code per month was 5017, the estimated increase was 1494 lines per worker‑month … the baseline mean for code review is 7.03 days” – hyperpape

3. Teams are exploring hybrid or spec‑driven processes to manage AI‑generated output
Instead of full manual review, participants suggest focusing human effort on high‑impact changes, defining clear quality bars, and codifying rules so agents can handle routine checks.

“Leave human review for the most impactful changes * Get better at defining the quality bar * Continually codify human concerns into rules or systems that can be checked against” – keybrd-intrrpt
“I think humans need to review requirements, specifications, and test scenarios.” – svachalek

🚀 Project Ideas

Generating project ideas…

SpecSync

Summary

  • Generates automated review checklists from human-written specifications (e.g., Gherkin, user stories) and validates PRs against them to reduce line‑by‑line review.
  • Core value: shifts human focus from syntax checking to architectural and requirement compliance, speeding up review while maintaining quality.

Details

Key Value
Target Audience Mid‑size to large engineering teams using PR‑based workflows (GitHub/GitLab) and practicing spec‑driven development
Core Feature Ingest spec files, derive checklists, comment on PRs with pass/fail and suggested changes
Tech Stack Node.js/TypeScript backend, GitHub Actions/GitLab CI integration, LLM API (e.g., GPT‑4o) for spec parsing, React frontend for dashboard
Difficulty Medium
Monetization Revenue-ready: SaaS subscription $15 per active user/month

Notes

  • HN commenters lament that “senior engineers find it very difficult to give up on line‑by‑line review” and want to “leave human review for the most impactful changes” (keybrd-intrrpt).
  • Provides a concrete way to codify human concerns into rules, enabling spec‑driven review and reducing tedious manual checks.

TestLens

Summary

  • Analyzes test suites to detect low‑value, duplicated, or over‑mocked tests and suggests consolidations or replacements with higher‑value scenario tests.
  • Core value: cuts CI time and maintenance burden while improving test effectiveness, addressing the complaint about agents “spew[ing] absolute mountains of code”.

Details

Key Value
Target Audience Developers and QA engineers maintaining large automated test suites (especially in AI‑generated code contexts)
Core Feature Static + dynamic analysis to score test usefulness, identify redundancy, generate optimized test variants
Tech Stack Python (pytest/ast), Rust for performance‑critical analysis, LLM for test summarization, GitHub Action, Grafana dashboard
Difficulty Medium
Monetization Revenue-ready: Usage‑based pricing $0.01 per test‑suite analysis

Notes

  • Commenters note AI “will happily duplicate some helper method… avoiding coupling” and produce “mountains of … test code” (CharlieDigital, japhyr), and that “large volume of tests waste CI cycles” (mejutoco).
  • TestLens directly tackles test bloat, giving teams a way to trust AI‑generated tests without drowning in noise.

ReviewPilot

Summary

  • IDE plugin that runs an agent‑based first‑pass review (correctness, security, style) and surfaces a concise summary, highlighting areas that need human architectural or performance judgment.
  • Core value: combines AI speed with human expertise, letting senior engineers skip line‑by‑line checks while still catching critical issues.

Details

Key Value
Target Audience Individual developers and teams using VS Code/JetBrains who rely on AI code agents and want lighter review loops
Core Feature Real‑time agent analysis, code‑summary generation, diff‑level annotation, toggle between AI and human review modes
Tech Stack TypeScript extension, Language Server Protocol, LLM API (e.g., Claude 3.5), WebAssembly sandbox for secure code execution, Telemetry opt‑out
Difficulty High
Monetization Hobby

Notes

  • HN discussants praise “agent‑assisted code review in a proper IDE… helping you to quickly construct a mental model” (skydhash, rbehrends) and criticize that “purely agentic code review remains subpar when judging architecture” (rbehrends).
  • ReviewPilot gives the best of both worlds: AI handles the tedious correctness checks, humans focus on design, addressing the desire to “get better at defining the quality bar” (keybrd-intrrpt).

Read Later