# AI Migration Agents Should Report Step-Level Progress ## Claim An unattended AI agent earns the right to be left alone with a long-running task only if it keeps surfacing what it's doing at each step, not just a terse status verb or total silence during multi-minute operations. <!-- atomicity: 1 idea ✓ --> ## Scope & conditions - Holds for any long-running, unattended, agentic task (site migrations, batch renders, large refactors) where the user is expected to walk away for minutes at a time. - Breaks down for short tasks (seconds, not minutes), where verbose per-step logging would just add noise without adding trust. - Does not by itself fix correctness — a chatty agent can still corrupt its own working state; the claim is about earned trust to be unattended, not about the task succeeding. ## Evidence Jonathan Bossenger's live-streamed test of Studio Code's `/liberate` command surfaces the same complaint on both test runs: `liberate extract` goes silent for minutes at a time with no per-step status ("pulling content," "migrating uploads," "verifying page 4 of 6"), so the only way to know it's still working is to wait. He explicitly contrasts this with wanting "Claude Code style" verbosity and logs it as the single highest-value DX fix to send to the Studio Code team — separate from, and just as consequential as, the database-corruption bug the same runs surfaced. ## Links - [[AI & the Future of Knowledge Work MOC]] — what an unattended agent owes the human who will be accountable for the result - [[3. Resources/Permanent Notes/Comprehension Debt Is the Real Cost of Unattended Autonomy]] — because extends: both claims are about what makes unattended agent autonomy trustworthy rather than merely fast; this one is about visibility during the run, that one about what's understood after it. - [[3. Resources/Permanent Notes/An Agent Is Defined by Working Without Mid-Task Steering]] — because contrasts: that note argues agent-ness requires no mid-task steering; this note argues the same unattended agent still needs to report, not be silently steerable. - [[3. Resources/MOCs/AI Agent Safety MOC]] — because applies-in: progress reporting is a concrete, checkable design requirement for any agent granted unattended write access. - [[3. Resources/Permanent Notes/Judgment at the End of an Agent Run Turns Every Late Failure Into a Restart]] — extended by: the handover needs a point to interject before the run ends (Kemple). ## Tests / falsifiers - A/B two builds of the same agent (verbose per-step logging vs. terse status) against the same test users and tasks; if self-reported trust and willingness-to-walk-away don't differ, the claim weakens. - If users abandon or interrupt verbose-logging agents at the same rate as silent ones, progress reporting isn't actually the trust lever it's claimed to be. ## Maintenance - Last tended: 2026-08-16 - Next review: 2026-11-16 ## Source - [[3. Resources/Literature Notes/Liberating sites with Studio Code (Bossenger)]] — reports the silence complaint across both live test runs of Studio Code's `/liberate` command.