Agent-built app shows coding is cheap but verification takes effort
Overview
Ian Hobson, writing in O'Reilly Radar, says agent-generated code shortened the coding time for his Task Viewer app but not the time spent verifying it.
He reports verifying the app through UI tests, manual checks, and SQL queries against the underlying data. He argues that claims such as "the agent got me 80% of the way there" are hard to interpret, because a plausible calculation can still rest on the wrong records. He also argues that delegating tasks to an agent may cost learners the understanding they would otherwise build.
Written by AI from the articles below · updated Oct 29, 7:15 AM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- O'Reilly RadarBlogAgent-built app shows coding is cheap but verification takes effort
AIIan Hobson's essay argues that agent-generated code shortens coding time but not verification, which for his Task Viewer app meant UI tests, manual checks, and SQL queries against the data. He says claims like "the agent got me 80% of the way there" are hard to interpret, since a plausible calculation can still rest on the wrong records. He also argues that tasks delegated to an agent may cost learners the understanding they would otherwise build.
Heat Trend
Not enough continuous observations to show a trend yet.