During the Apollo 11 descent, a few minutes from the lunar surface, the guidance computer threw a 1202 program alarm. Nobody in Mission Control was expecting it. The GO or NO-GO call fell to Steve Bales, the 26-year-old Guidance Officer, and he called GO. The landing continued.
The part that matters happened weeks earlier. In a training simulation, the Simulation Supervisor had thrown a program alarm at that same team, and they aborted a landing they should have saved. So they went back and wrote it down: which alarms were survivable, which weren't, what to check before calling GO. When the real alarm fired, Bales was carrying out a decision the team had already made, on a quiet day, with time to think.
If you work Medicare, your own descent starts October 15. The Annual Enrollment Period runs from then through December 7, and the decisions you make in those eight weeks will be made tired, between appointments, with a client on hold. The days before it are the last quiet stretch you get, and the most useful thing to do with them is decide.
The build window has mostly closed, and that's fine
I'll be straight about the timing, because a post that pretends otherwise would be selling you something. Back in August I wrote that AEP prep with AI is an August job, because a working AI assistant takes several rounds to get right and those rounds belong on quiet weeks. That's still true. If you started then, you're in dress rehearsal now, and you should keep rehearsing.
If you didn't start, don't start a new build now. A first version always teaches you what the second version should be, and in October the second version gets built on your clients. Nobody needs that, least of all the clients.
What you can still do, in an afternoon, is the part most agencies skip even when they did build: write down what your AI is allowed to do this season, what it's never allowed to do, and who answers for it. That applies whether you have a full Digital Crew running or one producer with ChatGPT open in a tab. Arguably it matters more in the second case.
What a crew can carry through the season
The work AI handles well in AEP is the work that follows the appointment and precedes the next one. A Scribe drafts the recap letter and the to-do list from your notes. A Scout reads plan documents and summarizes what changed, so you walk into the ANOC conversation with the comparison already laid out. A Coach hands you a morning briefing on who you're seeing and what you promised them last year. All of that is drafting and reading, and all of it leaves the deciding with you.
That line, between the drafting and the deciding, is the whole job before the season opens. Under pressure it moves. In week one you review every recap letter. By week five you're skimming, and somewhere in week six a Scout summary goes into a client conversation unchecked because it's 7:40 at night and it's been right forty times in a row. Nobody decides to let that happen. It happens because nobody decided it couldn't.
Write the rules your crew never breaks
In a Digital Crew, a Flight Rule has a specific shape. It names a decision the AI never makes, it's written down before the AI runs, and it carries its reason with it, so the person tempted to bend it at 7:40 on a Tuesday can see why it's there. It's the same move the Apollo team made after that failed simulation, scaled down to an agency.
I'll admit most of the rules my own crew runs on were written the same way, right after something broke. One of them says no outreach email ever offers continuing education credit, because I can't offer CE today, and a draft once did. Nobody imagined that mistake in advance. It happened once, and now there's a written rule standing between it and a second time.
The format is one sentence and a because: the crew never does this, because of this, and if it's asked to, it does this instead. A few that fit most Medicare agencies, to edit rather than adopt:
- The crew never recommends a plan. The recommendation is the reason the client came to you, and it stays yours. If a draft starts leaning toward a plan, it flags that for you and stops.
- Nothing reaches a client without a named person reading it first. Not the recap letter, not the plan-change summary, not the reminder text. Week six is the reason this one exists.
- Client information never goes into a personal AI account. Only the tools the agency has approved, set up the way your compliance guidance says. This is where Shadow AI turns from a habit into an exposure, and AEP is when it happens most, because everyone's in a hurry.
- When a plan detail and a crew summary disagree, the plan document wins. The summary gets corrected, and the correction gets logged so the same mistake doesn't come back next week.
Treat those as starting points to edit. Your carriers and your FMO will have their own requirements for marketing and client communication, and they outrank anything here. Check the rules you write against theirs before October 15, while you still have the time to read them properly.
Every seat needs a name next to it
A rule nobody owns is a suggestion. So the second half of the afternoon is putting a person's name beside each thing the AI does. Somebody reads the Scribe's letters when the producer who ran the appointment is in back-to-back meetings until six, and somebody decides whether the Scout's summaries are still good enough to use after a bad one gets through. If the honest answer to either is "whoever notices," that's the gap, and it's much cheaper to close now than in November.
A quick test for each thing AI does in your office this season: does it have a named role, a defined job, a human who owns its output, and at least one decision it never makes? Four yeses and it's a crew member. Anything less is a subscription with ambitions, and subscriptions with ambitions are exactly what slips during a busy season.
Decide now what you'll do when it breaks
Something will. A summary will be wrong, a tool will be slow on the busiest morning of the season, or a prompt that worked in September will start producing letters that sound nothing like you. The worst time to work out what to do about that is while it's happening.
So write one more rule: during AEP, a broken workflow falls back to the manual way, and gets fixed in December. You don't debug in the middle of the season. You write down what went wrong, in one line, the day it goes wrong, and you keep working. By December 8 you'll have a list of everything the season taught you, written while it was fresh, which is worth more than any fix you'd have rushed in November. The December chapter of the AEP prompt book has prompts for turning that list into next year's plan.
The last quiet afternoon
Bales got to call GO in seconds because the hard thinking had been done weeks before, by people who had time to do it well. You have that time until October 15.
Spend one afternoon of it on the rules and the names. Then go have the season you've been preparing for, with the relationships in your hands and the paperwork in someone else's.
If you want the prompts for the season itself, the AEP Prompt Book walks through it chapter by chapter, from the ANOC weeks and appointment days to the back office and the daily debrief. Pick the one stretch that buried you last year and set those prompts up while there's still time to edit them in your own voice.
