"Agent says it's done" isn't ready: passing tests only cover a fraction of ship-readiness
Glittering-Glass6135 · reddit · 2026-10-07
The author highlights a growing gap: AI agents are getting very good at implementing requested work, but "implemented" and "ready" are becoming two different states.
Passing tests, a green build, and a working happy path only prove known things work—they can't tell you what you forgot to ask about. The author lists issues that surface only at deployment time:
- Whether permissions are actually enforced in weird edge cases
- Whether the deployed app behaves like the code the agent inspected
- Whether integrations handle failures and retries
- Whether privacy-sensitive settings and session replay masking are correct
- robots.txt, sitemap, metadata, canonical URLs actually configured
- Missing pages, empty states, error states and other product gaps
- Legal/compliance review items nobody thought to check
- Implicit assumptions the agent made that nobody asked it to verify
The post asks the community: do you have a separate "before real users touch this" process, or do you mostly rely on the coding agent's own checks?
More from coding & agent
- Metix AI opens 900M profiles and 90M job postings data API for agents via REST and MCP — _jaydeepkarale · 2026-10-07
- Delivering personal agents will be far harder than any prior AI product category — nickbaumann_ · 2026-10-07
- DocETL in practice: a pipeline that extracts meds and side effects from transcripts — mdancho84 · 2026-10-07
- DocETL: a Python library for agentic data processing and ETL with AI — mdancho84 · 2026-10-07
- Josh Wills launches open-source data loading library — a year of commits as a fossil record of agent-assisted work — josh_wills · 2026-10-07
- HarnessTester finds 100+ real bugs in LLM agent harnesses like OpenClaw — LingmingZhang · 2026-10-07