Has the cost of building software dropped 90%?

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

The three most prevalent themes in the discussion are:

Skepticism regarding the claimed magnitude of productivity gains from AI tools: Many users doubt the assertion that AI tools have caused a drastic, quantifiable drop in software development cost (e.g., 90%), pointing to a lack of observable market changes or personal experience that contradicts such claims.
- Supporting Quote: As one user summarizes the general sentiment: "Had the cost of building custom software dropped 90%, we would be seeing a flurry of low-cost, decent-quality SaaS offering all over the marketplace, possibly undercutting some established players. From where I sit, right now, this does not seem to be the case." ("nine_k")
Focus shifting to non-coding skills for career longevity: In response to the perceived threat of automation, developers are discussing the need to pivot towards roles that emphasize business acumen, domain knowledge, and strategic thinking over mere implementation.
- Supporting Quote: A suggestion for navigating the foggy landscape is: "I think it's about looking at what you're building and proactively suggesting/prototyping what else could be useful for the business." ("martinald")
The enduring problem of maintenance/unmaintainable code, regardless of author: There is a strong, somewhat cynical consensus that while AI can rapidly generate new code (or features), that code often suffers from the same or worse maintainability issues as human-written code, meaning the bulk of development cost (maintaining old systems) remains unsolved.
- Supporting Quote: One user notes the dilemma regarding LLM output: "I've only seen huge, unmaintainable messes so far." ("JohnMakin"). This is echoed by another user's experience: "one year in, AI slop > Human-written slop" ("bdangubic").

🚀 Project Ideas

Career Path Navigation Service: The Domain Explorer

Summary

A personalized career guidance service designed for experienced developers concerned about AI-driven commoditization of general coding skills (as expressed by users like BigHatLogan and samdoesnothing).
Core value proposition: Transform developers from "spec-to-code" implementers into indispensable domain experts by mapping their existing technical skills to high-value, defensible business domains.

Details

Key	Value
Target Audience	Mid-level to senior software developers, especially frontend/full-stack, feeling career uncertainty due to LLMs.
Core Feature	Interactive module that scores the user's current business domain knowledge (e.g., finance, logistics, CRM) and suggests adjacent, future-proof "slices of the industry" (`embedding-shape`) to specialize deeply in, providing actionable learning paths to move beyond coding specs.
Tech Stack	Skill mapping engine (Graph Database like Neo4j, Python/Pandas for analysis), simple web interface (React/Next.js), integrated lightweight LLM (like Claude 3.5 Sonnet) for generating narrative suggestions based on domain analysis.
Difficulty	Medium
Monetization	Hobby

Notes

Why HN commenters would love it: Directly addresses the "fog of war" about career positioning. It operationalizes high-level advice like "get better at the business domain" (BigHatLogan) and "specialize in one slice of the industry" (embedding-shape) by providing concrete steps.
Potential for discussion or practical utility: This tool generates direct discussion on which domains are most robust against automation (e.g., moving away from general frontend work towards complex regulatory compliance or niche process optimization).

Spreadsheet-to-Application Modernization Platform

Summary

A low-code/no-code platform specifically designed to ingest, analyze, and semi-automatically convert complex, high-risk internal Excel/Sheets workflows into maintainable, version-controlled, secure internal applications.
Core value proposition: Mitigates the risks associated with critical "Shadow IT" spreadsheets—data corruption, lack of auditability, and reliance on the original creator (croes, robotresearcher)—by leveraging AI for logic extraction and scaffold generation.

Details

Key	Value
Target Audience	Developers/consultants serving SMEs, internal IT departments supporting "Shadow IT," and domain experts (e.g., finance controllers) whose critical work resides in massive spreadsheets.
Core Feature	Spreadsheet ingestion (upload CSV/XLSX), AI-driven visualization of data flow and logic derivation, automated generation of a basic web application structure (UI layout mirroring the sheet structure) using a modern framework, and suggestion/generation of unit tests for critical formulas (`robotresearcher`, `martinald`).
Tech Stack	Python backend for spreadsheet parsing (Pandas, custom logic extraction LLM prompts), Backend framework (Go or Rust for performance), Frontend (SvelteKit or Streamlit for rapid internal app prototyping).
Difficulty	High
Monetization	Hobby

Notes

Why HN commenters would love it: It directly tackles the "core sheets" problem discussed (martinald, tgv) without demanding users become full-stack developers overnight or adopting inappropriate external SaaS solutions (swatcoder, chasd00). Offering robust tooling that understands spreadsheet jank would be highly valued.
Potential for discussion or practical utility: This sparks debate on the effectiveness of current LLM logic extraction versus dedicated tools, and whether the platform is the "modern Microsoft Access" that some users wished for (ASalazarMX).

AI Codebase Context & Intelligibility Tool

Summary

A developer tool that acts as an intelligent "context management layer" over newly generated or poorly documented codebases (especially LLM output or legacy "human slop"). It prioritizes maintaining readability and structure over sheer speed of generation.
Core value proposition: Solves the "unmaintainable mess" problem (JohnMakin, pessimizer) by forcing AI-generated code to adhere to organizational standards and providing instant, deep context narration, thus ensuring long-term viability even when the original creator forgets the context (jazzyjackson, kenjackson).

Details

Key	Value
Target Audience	Developers actively using LLMs for feature generation or working in high-churn code environments where rapid iteration kills clarity.
Core Feature	Contextual Integrity Check: Integrates into IDEs (like VS Code extension) to analyze newly added LLM code blocks. It generates mandatory, easy-to-read documentation/summaries about the generated code block and its assumptions, and runs automated refactoring/style checks against organizational configs before accepting the code into the staging branch (`theshrike79`).
Tech Stack	VS Code Extension (TypeScript/WebAssembly), Local LLM invocation (potentially running optimized models like Phi or Llama for focused tasks) for summarization and analysis, hooks into Git pre-commit hooks.
Difficulty	Medium
Monetization	Hobby

Notes

Why HN commenters would love it: It validates the skepticism that LLMs produce maintainable code (bitwize, JohnMakin) by providing a solution that focuses on making the code readable and debuggable for the human who has to revisit it later. It directly supports the idea that a senior developer is needed to guide the LLM outputs (bdangubic).
Potential for discussion or practical utility: This shifts the focus from "is the code fast to write?" to "is the code fast to understand/maintain?"—a key concern for seasoned engineers. It could be positioned as the necessary governance layer for the LLM "junior dev army."