TL;DR
Responses in the Ask HN thread range from roughly 2× to a subjective 10–20× change in individual output, depending on workflow, domain knowledge and tooling. Some engineers report reorganizing work around GitHub and Copilot agents, while others warn about poor prompts, refactoring and risks in unfamiliar stacks.
What happened
On Hacker News, contributors debated how much AI has changed their productivity as software engineers. One commenter described a radical workflow redesign: treating GitHub as the primary workspace, wiring Copilot and agents into repository folders that contain context and instructions, and using those repos to hold KPIs and operational workflows. That user said AI tightened the loop from idea to iteration, turning tasks that once took weeks into days or hours and even replacing regular stand-ups with AI‑mediated summaries. Another commenter gave a more conservative estimate, saying overall productivity is about 2× higher. They added nuance: when they understand the domain and stack, the same engineer can be roughly 10× faster; when not, AI can produce misleading or low‑quality code that requires lengthy refactoring. Small environment tweaks (dotfiles, shell configs) account for a modest share of gains, they said.
Why it matters
- Engineers are experimenting with AI beyond code completion, embedding agents and rules directly in repositories to coordinate work.
- AI can compress development cycles from idea to shipping, potentially changing how teams structure processes and communications.
- Benefits vary greatly by domain knowledge and familiarity with the tech stack; poor prompts can create more downstream work.
- If teams adopt AI‑mediated tracking and summaries, traditional rituals like stand‑ups could be rethought or reduced.
Key facts
- One commenter reported a subjective 10–20× increase in output after restructuring workflows around AI agents in GitHub.
- That user said they rarely use standalone ChatGPT/Gemini/Claude; most activity happens inside GitHub with Copilot and agents.
- They described creating 'master' repos with folders that contain context and instructions for AI agents, covering hiring, reviews and operations.
- AI was credited with shortening the full cycle: idea → spec → implementation → iteration, turning multi‑week tasks into days or hours.
- Another commenter estimated about 2× overall productivity gain compared with the pre‑LLM era.
- When the domain and stack are well understood, that commenter said they can be about 10× faster; in unfamiliar domains the work can require extensive refactoring.
- They attributed roughly 10–15% of their improvement to quick dev‑environment tweaks (e.g., modifying .zshrc or .vimrc via AI).
- Both accounts emphasize subjective impressions and note that precise measurement of productivity change is difficult.
What to watch next
- Whether more teams adopt repository‑centric agent workflows and formalize repo‑based rules and KPIs (not confirmed in the source).
- How organizations measure and standardize productivity gains from AI tools across different domains and stacks (not confirmed in the source).
- If integrated agents inside developer platforms (e.g., Copilot agents wired to repos) become more common than standalone LLM chat tools (not confirmed in the source).
Quick glossary
- LLM: Large language model — a type of AI trained on vast text data to generate or transform language, often used for code and writing assistance.
- Copilot: An AI code assistant integrated into development environments that provides completions and suggestions based on context.
- Repo: Short for repository — a storage location for project files and history, commonly hosted on platforms like GitHub.
- Prompt: Input given to an AI model or agent to elicit responses or actions; quality of prompts affects outputs.
- OSS: Open source software — software with source code that anyone can inspect, modify, and distribute under its license.
Reader FAQ
How much has AI improved engineer output overall?
Estimates in the thread vary: one contributor described a subjective 10–20× change after workflow changes, another said roughly 2× overall.
Are engineers replacing chat tools with integrated agents?
One commenter reported preferring GitHub with Copilot and agents over standalone chat models; broader adoption is not confirmed in the source.
Do AI tools eliminate the need for code review or refactoring?
No. The comments note AI can speed delivery but may produce low‑quality code when prompts or domain knowledge are lacking, leading to refactoring.
How much improvement comes from environment tweaks?
One commenter estimated about 10–15% of their productivity improvement came from quick dev‑environment changes like updating dotfiles.
For me it feels like roughly a 10–20x change, but mostly because I restructured how I work rather than just adding an “AI helper” on top. In the last year…
Sources
- Ask HN: By what percentage has AI changed your output as a software engineer?
- Building a HackerNews "For You" Feed | Shaped Blog
- Will AI replace Software Engineers?
Related posts
- Manus Reaches $100M ARR and $125M Revenue Run-Rate Eight Months After Launch
- Kapwing Report: Up to a Third of New YouTube Shorts Feed May Be AI Slop
- C -> Java isn’t the same as Java -> LLM: why source code remains central