Hacker News thread on a year after a 100% AI mandate
The feed snippet shows 49 points and 51 comments and does not include the post body.
Stories per day · 30 days
2 RSS
Relatedcoding-agents4opinion4claude-code2code-review2hiring2agentic-coding1agentic-rl1ai-agents1ai-disruption1anthropic1benchmark1benchmarks1career1claude1code-agents1codex1continuous-integration1developer-productivity1
27 stories, newest first · one card per story
Stories tagged #software-engineering
Blog argues unreviewed AI proxy work should be fireable
The author distinguishes guided AI use from pasting agent output with no review, disclosure, or ownership of failures.
Longtail-SWE adds 202 SWE tasks for 13 underrepresented languages
A solver-judge-repairer pipeline revises tests and descriptions; nine model setups were evaluated and artifacts released.
Thoughtworks essay: experienced developers wary of agentic coding
A no-human-code week still left a frequently red build and most human time at the whiteboard.
CodeSpecWeave builds coding tasks with interdependent specifications
The pipeline produces 4,670 tasks across 2,865 repositories and a 100-task benchmark with 1,044 specifications.
Anthropic manager drops coding interviews as Claude handles the tasks
Felix Rieseberg says prototypes and business sense now matter more than tenure or classic coding tests.
Engineer argues agent-written code still needs human ownership
A SonarQube complexity flag on Codex-generated React code led the author to inspect, then have the agent split one file into ten.
1
Engineer says coding agents raised his GitHub activity and workload
By October 27 he logged 2,102 contributions in 2026, already above every full year from 2021 to 2025.
1
Lean4AgenticRL trains LAR-9B and LAR-27B with process rewards
Authors report LAR-27B at 77.3% on TerminalBench-2.1 and a 9.89% average gain over the base model.
Fable Took My Job
An engineer says frontier models like Fable helped leadership eliminate his role without matching his judgment.
1
AI coding agents generate more code, but not more software
A study finds AI coding agents speed coding but raise pull-request review time 49% and change requests.
3
AgentSWE: Can Coding Agents Build the Agent You Actually Want?
AgentSWE shows coding agents rarely build agents that satisfy full held-out requirements.
ProgramGym: Scaling Training Environments for Long-Horizon Program Reproduction Agents
ProgramGym supplies 1,648 verifiable environments and raises Qwen3.8-27B program-reproduction accuracy from 34.87% to 42.84%.
SWE-Duel: Towards a Self-Scaling Arena for Coding Agents
SWE-Duel has coding agents author and repair hidden bugs so difficulty scales with participants.
SWE-Game: Can Coding Agents Build the Games We Want?
SWE-Game finds coding agents, led by Opus5, still score below 60 building games.
What Did the Agent Actually Do? Evidence-Grounded Oversight for Long-Horizon Agents
AgentMonBench and an evidence graph help monitors verify consequential decisions made by long-horizon agents.
The Four Horsemen of Agentic Coding
An essay argues agentic coding creates unreadable slop, deskills engineers, and weakens team collaboration.
4
Shockingly Simple Self-retrospection Improves Agentic Models Without RL
Retrospection-only fine-tuning lifts Qwen3.5-4B on SWE-bench without rewards or an external teacher.
PluginRSI: Recursive Improvement of Agent Harnesses with Reusable Plugins
PluginRSI evolves agent harnesses as reusable plugins that transfer across models and tasks.
CUA-SWE: When Computer-Use Agents Meet Visual Software Engineering
CUA-SWE tests agents that repair software using code changes and visual GUI feedback.
Compact Documentation for Coding Agents: A Benchmark, an Optimizer, and Why It Does Not Transfer
A coding-agent benchmark finds compact documentation does not help agents fix real repository issues.
Note on 24th September 2026
Willison argues coding agents raise the skill and discipline needed to engineer software well.
Agent-Editing World Model: Rethinking World Modeling for LLM Agents
AEWM edits contaminated agent state and improves six benchmarks by 3.2 to 6.7 points over the strongest baseline.
One to More, More to One: Category-Aware Iterative Expert Training for Software Engineering Agents
Category-aware expert training and distillation lift a software-engineering agent to 58–59% resolution on SWE benchmarks.