AI agents and scheduled automations are not set it and forget it. The requirements you start with are an educated guess, and you often discover what you need only after the agent does something you don’t need. In this episode of The Missing Manual, Julia Levy and Jenny Cotie Kangas explain how to tune an agent once it is running, and why telling the AI where it falls short is the step most people skip.

Jenny starts with a question from a recent workshop. An attendee built a Gmail agent to triage her inbox, and the agent kept stopping to ask for permission. The fix was in the architecture. An agent connected to Gmail on the back end should not need permission every time, but one working through the browser will. Jenny then describes her own signal brief, which drafted replies in her voice across email, WhatsApp, LinkedIn, and text. If she left one email unanswered for five days, she came back to five drafts of the same reply. She rebuilt the agent so it shows her where her attention needs to go, and then they work through each response together.

Julia adds that AI models change, so anything built on a model can drift over time. She recommends putting regular reviews of your automations on the calendar, the same way you would review a website. She has Claude interview her at the start of a build and ask what she isn’t thinking of. Then she runs the same conversation through Gemini and ChatGPT to catch what she missed. Jenny suggests going one step further with a scheduled automation that reviews your other automations, checking which model each one uses, its outputs, and its token use.

The closing advice from both hosts is direct. If an agent frustrates you, say so to the AI, because it cannot match your idea of good until you define what good is and what it isn’t. Then have the AI ask you questions and challenge you rather than agree with everything. The Missing Manual is also opening up to guests, so if you are building something interesting with AI, Julia and Jenny want to hear from you.

Find The Missing Manual wherever you listen to podcasts.

AI in recruiting fails most often because companies apply it to the wrong problem. In this episode of The Missing Manual, Julia Levy and Jenny Cotie Kangas bring back what they heard at RecFest Nashville. Their main observation is that many organizations are using AI to get to the wrong answers faster.

The clearest example of AI used well came from a team that pointed AI agents at its recruiting funnel conversion data instead of adding more applicants at the top. The analysis showed that Indeed was not producing quality candidates for those roles. The team turned it off and ended up with stronger candidates. The counterexample was a large company that fixed its internal process but never brought its MSP along. Candidates got a worse experience, and that experience reflected on the employer's brand, not on the MSP.

Julia Levy explains why AI screening breaks down when job postings are never fixed. Candidates use AI to match their resumes to the posting, and screening tools grade them against that same posting. If the requirements are wrong, both sides are optimizing toward the wrong target. She also shows how a weak needs analysis and a rushed intake conversation lead to scope creep three months into a search. Jenny Cotie Kangas calls this the same problem wearing different clothes: misalignment that never gets captured in the process. Her fix is to treat AI change as a program rather than a project, and to use AI to build a runbook that helps a first-time hiring manager pre-mortem the search before the role opens.

The Missing Manual is a podcast for working professionals about the unwritten rules of work, careers, leadership, and AI strategy, with every episode under 20 minutes. Find The Missing Manual wherever you listen to podcasts.

Talent acquisition teams do not need more convincing to use AI. At RecFest 2026 in Nashville, most of the leaders in the room already had leadership buy-in and access to tools like Claude and ChatGPT. What they wanted was the actionable how: which use case to start with and how to build it. In this episode of The Missing Manual, Julia Levy and Jenny Cotie Kangas recap two days of hosting, speaking and listening at one of the biggest TA and HR conferences in the US.

Jenny Cotie Kangas spoke on the applied AI stage. When she asked the standing-room-only audience who had leadership buy-in and who had AI tools at work, almost every hand went up both times. The notes came out when leaders described agents they had actually built, such as a job posting checker and a signal brief. Julia Levy spoke on the "Your ATS Sucks" panel, where teams of every size described the same problems and different ways to work around them.

The hosts also explain why RecFest feels like an unconference, and why the relationships built there matter. On the Unplugged Stage, they gave every audience two minutes to meet a stranger, and attendees kept telling them about those connections for the rest of the event. Julia connects this to her relationship bank account idea and to the peers she has called for advice over the years, including people at competing companies, because the problems in TA rarely change from one company to the next.

Julia is also honest about her plans that never happened: the QR code business cards, the badge photos and the notebook that stayed blank. The takeaway for your next conference is to go in with a plan, meet people outside your own team and follow up afterward. Find The Missing Manual wherever you listen to podcasts.