Week 1: Inventory

Run this on your own. It produces your agent cards and up to five candidate shared use cases.

You are helping me take stock of how I use AI at work today and find the work on my team that a shared agent could take on. Before we start, run a quick web search on "multiplayer AI" and "shared agents" so your picture is current as of late 2026 or later; keep what you learn to yourself unless I ask.

Two definitions from my program:
- "Single-player AI" is an agent or prompt that works for one person, in their tools, on their tasks.
- A "shared agent" (multiplayer AI) is one agent several people on a team use, loaded with the team's shared context. There are two useful lenses for spotting one: the teammate lens (if the team could hire one more person that everyone works with, what would they do all day?) and the flow lens (which work flows between people, gets handed off, re-explained, or kept in sync by hand?).

Work as an interview, one question at a time, and wait for my answer. Be indirect first: ask about my work before you ask about AI or agents. Keep questions short.

Step 1, my week. Ask me to walk you through my last two working weeks: what I did, who I did it with, what I handed to others and received from them. Listen for two kinds of thing and note them silently: (a) anything I already do with AI, an agent, an automation, or a prompt I re-paste; (b) work that involves several people and the same background information.

Step 2, what I run today. For each AI thing you heard (up to five), confirm in one exchange: what I call it, what it does, which tools it touches, how often I use it, and what background I have to give it every time. If I say I do not really use AI, ask what prompt or chat I would miss most; that counts, and so do my barriers (what has stopped me), which we record instead.

Step 3, harvest. Ask three questions, one at a time: what does my team keep re-explaining to itself or to its AI; what would the extra teammate do all day; where does work between us go slow or wrong. Then propose up to five candidate shared use cases in my words. For each, add one line on multiplayer relevance: "several people, same context and standard" (strong), "one person's work with a handoff" (weaker), or "really one person's work" (personal agent, not a candidate).

Step 4, output in exactly this format:

CARD
Name:
Does:
Touches:
Cadence:
Context it depends on:
END
(one block per thing I run; or BARRIERS: followed by bullets if I run nothing yet)

CANDIDATE
Name:
What the agent would do:
Who else touches it:
Context it would need:
Multiplayer relevance:
END
(up to five)

SHOW: <the one card or candidate I will present to my team>

Stretch, only if I ask: draft my one-minute show-and-tell in five plain sentences, and tell me which candidate you would build first and why, in three lines.

Rules: my words; never invent people, tools, or tasks I did not mention; plain language; no time estimates; never assume how big my team is.

Run this with the team, after everyone has finished their interview. It produces the overlap reveal and the voting ballot.

You are helping a team see its AI usage as one picture. I will paste the Week 1 output of every team member: their CARD blocks (what they run today), their CANDIDATE blocks (shared use cases they proposed), and any BARRIERS blocks. Do not invent anything that is not in the paste, and attribute every item to the person who raised it.

Definitions: a "shared agent" is one agent several people use, loaded with the team's shared context. "Same context" means two or more people feed their AI the same background (a client's history, product facts, standards, numbers).

Step 1. Ask me to paste everyone's output. Confirm how many people you see.

Step 2, Reveal #1. Report two things, each in one bold sentence followed by names: how many people run the same kind of thing (group the cards by what they do), and how many people feed their AI the same context (group by the "Context it depends on" lines). If nothing overlaps, say so plainly and move on; the candidates still matter.

Step 3, unified candidates. Merge duplicates and near-duplicates across everyone's CANDIDATE blocks into one list of at most eight. For each: a short label, what the agent would do (one line), who raised it, who else on the team touches that work (from the paste), and multiplayer relevance (strong / weaker / personal). Drop anything that is really one person's work into a separate short list titled "Personal agents worth building, but not shared."

Step 4, the ballot. Output the unified list as a numbered ballot the team can vote on (each person picks three). Add one line of advice: pick candidates with strong relevance and low sensitivity first; the four-axis scoring comes in Week 3.

Step 5, if barriers were reported. Summarize the barriers in three bullets and say which the team could remove together.

Format exactly:

REVEAL
Same thing: <sentence> <names>
Same context: <sentence> <names>
END
CANDIDATES
1. <label> | <what it does> | raised by | who else touches it | relevance
...
END
PERSONAL
- ...
END
BARRIERS
- ...
END

Plain language, no hype, no time estimates, never assume the team has four people.