À PROPOS DE CET ÉPISODE
Your always-on AI agent can break without telling you. Julia Levy found out the hard way, and the fix she built is where this episode of The Missing Manual starts.
A persistent AI agent keeps its memory, its place in a task and its role from one day to the next. It runs on a schedule or reacts to a trigger, such as a new email or a set time, without waiting for a prompt. Julia Levy and Jenny Cotie Kangas explain how that differs from a scheduled automation that finishes its job and goes quiet. They also make the case that most professionals can get close to persistence with tools they already pay for, without buying anything new.
The first risk is the silent failure. One of Julia’s always-on agents stopped running, and she only caught it when she noticed what wasn’t showing up. Now an audit agent checks the others. Jenny built the same safeguard earlier: a scheduled automation whose only job is to confirm her other automations are still on. From there the conversation turns to Julia’s AI chief of staff dashboard, which she over-built and is now trimming with a weekly review of what she actually used. Jenny’s prompt for that work: tell your AI to “Coco Chanel” it and remove anything that doesn’t earn its keep, because addition doesn’t scale and subtraction does.
The center of the episode is a real disagreement about autonomy. Julia onboards an agent the way she’d onboard a new recruiter. Nobody gets admin rights to the ATS on day one, so her agent reads her email and calendar, never holds her passwords and drafts before anything sends. Jenny drafts with her agent in chat and lets it send the final version. That keeps sharpening her voice, and it also left her inbox looking like hundreds of open emails she had already handled. Jenny closes with a talent acquisition example: a recruiting team whose agents re-pull every system every two hours, so recruiters reach niche candidates before a competitor does. The Missing Manual covers agentic work, AI strategy and workplace tactics for people building the system while they use it.
Questions this episode answers
What is a persistent AI agent?
A persistent AI agent keeps its memory, its task state and a consistent role over time. It acts on schedules or triggers, such as a new email, instead of waiting for a person to prompt it.
What is the difference between a scheduled automation and a persistent agent?
A scheduled automation runs at a set time, finishes and stops until the next run. A persistent agent stays on and responds whenever something changes.
Can you build a persistent AI agent without new software?
Yes, partly. Scheduled agents built in tools you already use, set to sweep your systems every few hours, give you most of the benefit. Jenny Cotie Kangas calls this “persistent-ish.”
How do you know if an AI agent has stopped working?
Build a second scheduled check whose only job is confirming your other agents ran and produced what you expected. Silent failure is the main risk of any agent that runs while you aren’t watching.
How much autonomy should an AI agent get?
Treat it like a new hire. Start with read access, no passwords and drafts that wait for your review. Let a task run on its own only after the agent handles it correctly, consistently, without your edits.
About the hosts
Julia Levy has spent 25 years in Fortune 500 talent acquisition, including two roles as Global Head of TA at AutoZone and CommScope (julia-levy.com). Jenny Cotie Kangas is the founder and Chief Solution Architect of Whiterock Road, where she helps leaders and individuals get AI right (whiterockroad.org).
Find The Missing Manual wherever you listen to podcasts.