Most CEOs walk into a meeting partially informed. The dossier was supposed to get done. Then a wholesale call ran long. A tasting room issue pulled a manager out of the day. Suddenly it was 9 AM and the meeting was at 9:30. You walk in with what you can remember, ask the right questions to fake it, and hope the other side gives you enough signal to recover. Sometimes that works. Sometimes you walk out wishing you had read the one thing that would have changed the conversation.
The Researcher agent solves the part of that problem that is software-solvable. It is the first agent in the daily chain — it wakes at 6 AM, reads the industry, filters for what matters to your business, and writes a five-line digest before the day begins. By the time you sit down with coffee, the morning's reading has been done. This post is the per-agent breakdown for the Researcher in the Agent OS Webinar series.
I recommend the Researcher as the first agent to install. Week one in the 90-day install order. It pays for the stack the first morning it runs.
What the Researcher reads (Inputs)
Five categories of source material show up under every Researcher that works. The exact list shifts by operation; the shape is consistent.
- Industry news. The 8 to 15 publications and aggregators your sector reads. For a winery, that includes Wine Industry Advisor, Wine Spectator, SevenFifty Daily, Wine Business Monthly, Decanter, Practical Winery, and a handful of regional and trade-specific feeds. The Researcher reads everything new in those sources since its last run.
- Customer feedback channels. Support tickets from the last week. NPS comments. Wine club survey responses. Tasting room reservation notes. The cross-channel pile of words your customers have written, which most operations never read in one sitting.
- Competitive signals. The 5 to 10 competitors you care about. Their press releases, their LinkedIn, their wholesale price lists when public. New releases. Personnel changes. Anything visible from the outside.
- Event-triggered context. Calendar lookups for upcoming meetings, walkthroughs, or distributor calls. When a tasting room walkthrough is scheduled with the GM, the Researcher pulls the last 90 days of related notes, open punch-list items, and prior commitments.
- Your own files. The daily notes, the open commitments, the brand voice guidelines, the wholesale strategy doc. The Researcher should read your context before it reads the world's.
If you cannot list a Researcher's inputs in one breath, the agent does not yet know what it is for. The order of the categories matters too — your own files come last in the list but they get read first in practice. The Researcher needs to know what your operation cares about today before it reads the world's headlines, because the world's headlines are not chosen for your operation.
What it does with that material (Skills)
The Researcher's job is to compress. It is the agent that reads ten thousand words and gives you back five lines. The skill the Researcher needs is the same one a good newsroom assistant has: the ability to tell what matters from what does not, for a specific reader.
The Researcher has three named skills:
- Filtering. Given 60 to 120 candidate stories per morning, pick the 2 to 4 that matter for your operation today. The filter is your operation. Whatever the industry calls important is a different filter and the Researcher should ignore it.
- Clustering. Voice-of-customer weekly roll-up: read 200 to 500 customer comments and return three named themes with three quotes each. Competitive scan: ten competitor moves grouped into three named patterns.
- Briefing. When the calendar says you have a walkthrough or a buyer call, write the 90-second dossier that gets you ready for it. Last touchpoints, open items, prior commitments, what the other side cares about.
A worked example of the Briefing skill: you have a 10 AM call with a wholesale buyer you have spoken to twice this year. The Researcher pulls the two prior email threads, the order history for that account, the open commitments from your last conversation, and the published context on their business — a recent funding round, a new buyer, a press piece on their natural-wine direction. Three to six sentences sit in your daily note at 9:30 AM. You walk into the 10:00 call with the last conversation already in your head.
The Researcher does not draft your LinkedIn post. It does not write your wholesale outreach. The Speechwriter does those things downstream, reading the Researcher's output. Keeping that boundary clean is how the chain works.
The schedule (Scheduled Tasks)
A schedule is what turns a clever prompt into a system. The Researcher runs on three cadences.
- Daily at 6:00 AM. Industry news scan. Five-line digest written into the morning briefing folder, ready for the Speechwriter to read at 6:30 and the Chief of Staff to read at 7:00.
- Weekly on Sunday evening. Voice-of-customer roll-up. Competitive scan. Both go to a Research Briefs folder for review during the planning cluster.
- Event-triggered, 30 to 60 minutes before a meeting. Pre-event dossier. Pulled into your daily note so it sits next to your calendar entry for the meeting.
The 6:00 AM time is not arbitrary. The Researcher has to finish before the Speechwriter wakes up at 6:30 to consume its output. The Speechwriter has to finish before the Chief of Staff assembles the briefing at 7:00. The chain is timed. Slip one piece and the downstream agents work with stale or missing inputs.
What lands on your desk (Outputs)
Three kinds of output, one folder for each.
- The morning digest. Five lines. Today's three industry stories that matter for your operation, with one sentence of "so what" attached to each. Two short sentences left over for context, surprises, or open questions.
- The weekly briefs. Voice-of-customer roll-up and competitive scan, each one page. The same shape every week so you can scan them in 90 seconds and dig into the one section that matters.
- The event dossier. Three to six sentences listing the last four touchpoints with the people you are about to meet, the open punch list, and the prior commitments. Sits in your daily note 30 minutes before the meeting.
The Researcher never sends email or auto-replies. It is read-only on outbound, writing to your files and nobody else's.
The goal: walking in informed
The Researcher exists for one outcome. You walk into every conversation already informed. The meeting that used to take 20 minutes of last-minute prep takes 90 seconds of reading the dossier the Researcher already wrote.
This is the agent that earns its place fastest. Most operations have a list of things they would read if they had time. The Researcher creates the time. Five lines of reading in the morning replaces 30 minutes of throat-clearing across the day.
Once you have a Researcher running, the rest of the team can tell. Your wholesale buyer notices that you opened the call already knowing about their funding round. Your GM notices that the punch list never gets surprised by something you forgot to track. Your team starts to trust that the meetings you walk into are the ones you walked into prepared. That is the part of the system that compounds beyond the digest itself.
How the Researcher demonstrates Interview
The Researcher is the cleanest example in the shadow team of the Interview teaching move. Most winery leaders keep up with the industry by reading more. The Researcher inverts that. The agent reads. It asks questions you would not have asked. It puts the answer where you will see it before the meeting.
Here is what this looks like at Not Really Wines, our fictional demo winery at NunnCurtis Labs. NRW makes premium non-alcoholic wine, vacuum-distilled from Sonoma fruit. Maren Holloway, Co-Founder and CEO, gets a digest every morning that looks something like this:
Three stories worth your attention today. (1) A major retailer reports non-alcoholic wine sell-through up sharply week-over-week. So what: the segment narrative is shifting from niche to mainstream faster than the press is tracking. Worth one piece of content this week. (2) A competitor dropped its 6-pack price. So what: pricing pressure on the bundle is real. Consider adjusting your Starter Pack messaging. (3) A wine publication ran a long piece on adult non-drinkers. So what: useful raw material for The Society's Friday note.
Maren did not ask any of those questions before 6 AM. The Researcher did. By 6:30 the Speechwriter is reading the digest and deciding what to write. By 7:00 the Chief of Staff has the digest in the morning briefing. By 7:30 Maren is sitting down with coffee with all of it.
Why this is the first agent in the install order
I recommend the Researcher as the first agent for a single reason: it pays for itself in week one without needing any of the others. The other eight all become more useful when they are part of the full chain. The Researcher is useful from the first morning it runs, on its own, even if you never install another agent.
Three more reasons it goes first:
- The Researcher's outputs are the simplest to evaluate. Read the digest. Was it useful? You know within five mornings whether the agent is earning its place.
- It teaches you the five-field anatomy in the simplest possible case. Inputs are easy to list. The skill is one job. The schedule is one trigger. The output is a five-line file. The goal is one sentence.
- It produces the data the next several agents need. The Speechwriter, the Chief of Staff, and the Connector all read its output. Skipping the Researcher and installing those agents first means installing them blind.
Build this one this week. The webinar recording, the slide deck, and the rest of the Agent OS series live at nunncurtislabs.com/webinars.