Pick the sentence that sounds like you.
Four sentences per question, scored 1, 2, 4 and 5. Pick I don’t know where you cannot answer; it counts, it never scores. The band appears after eight scored answers. Press Load our answers to see how we score ourselves, honestly, in October 2026.
Readiness scorecard · self-reported · nothing leaves your browser
Pick the sentence that sounds like your company today.
Twelve questions, two per pillar, four sentences each. Pick I don’t know when you cannot answer; a band appears after 8 scored answers.
0 of 12 answered. Answer at least 8 to see a score.
Data
Can an assistant reach your records, and can it say where an answer came from? At Pirxey: Our second brain reads twenty-five tools with each person's own logins, and the index tells it what to open first.
– of 50 of 2 answeredReason
Is there a business reason with a number on it, or a demo that impressed someone? At Pirxey: Every brief we publish starts from one question we actually had, with the count that answered it.
– of 50 of 2 answeredRules
If an answer is challenged, does somebody own the decision, and is there a trail? At Pirxey: We keep an inventory of every AI system in use and a standing group that approves a new use of a model on company data.
– of 50 of 2 answeredPeople
Do the people who will use it know what it is bad at? At Pirxey: Our monthly Pulse: 77 answers in September, 69% at Crew level or above after three editions.
– of 50 of 2 answeredPlumbing
Can you connect, watch and change a model without a heroic week? At Pirxey: One maintained setup of connections, checks on every change, and a loop with a person at five stops.
– of 50 of 2 answeredBacklog
Is there a ranked list, or a pile of wishes? At Pirxey: We rank by value against the number of systems touched, and the first working version has a six-week limit.
– of 50 of 2 answeredYour scorecard · 0 of 12 answered
Answer at least 8 of the 12 questions to see a band. 0 so far.
Self-reported, not an audit: the score says where to look, not what is true. Nothing is sent anywhere. With Remember on, the answers stay in this browser profile until you press Clear.
What each pillar asks, and what we do about it.
Six things we check before an AI project, in plain words. Under each: how it works at Pirxey as of autumn 2026, and the brief that shows it.
Data
Can an assistant reach your records, and can it say where an answer came from? Not a data lake: a list of the systems that hold what you would ask about, an owner per system, and the sensitive sets marked before anyone connects a tool. Our second brain reads twenty-five tools with each person’s own logins, and the index tells it what to open first. Tracing an answer back to its source still takes a person; that is our open item here.
Reason
Is there a business reason with a number on it, or a demo that impressed someone? The test is a business owner who will be asked about adoption, not the engineer who built it. We have two or three named problems with a measure each, and every brief we publish starts from one question we actually had. The rule for when a pilot earns more budget is the part we have not written down.
Rules
If an answer is challenged, does somebody own the decision, and is there a trail? We keep an inventory of every AI system in use, including the unofficial ones, and a standing group that approves a new use of a model on company data. What may go to a hosted model is written down, shared knowledge is checked before it is saved, and a person approves what leaves the company. Brief 017 is the long version.
People
Do the people who will use it know what it is bad at? Measured by use on real tasks, not by workshop attendance. Our leaders used the assistant weekly before anyone else was asked to, which is the honest reason the company followed. The monthly Pulse had 77 answers in September, 69% at Crew level or above after three editions, and it asks every month who is blocked by what.
Plumbing
Can you connect, watch and change a model without a heroic week? One maintained setup of connections with the person’s own logins, logs someone reads, checks on every change, and a loop with a person at five stops. The six-week build in brief 018 ran nine checks on every change; the saved set of questions that runs before a prompt change is what we add next.
Backlog
Is there a ranked list, or a pile of wishes? Value against the number of systems touched, one table, the top item in progress with a date, and a baseline measured before the start. We rank quarterly and give a first working version six weeks. The baseline before the start is the step we skip most often, which is why our own card points there first.
The same six pillars, in the order they grew for us.
We did not start ready. On 15 May 2026 the second brain ran on one laptop. The pillars grew together; this is the order in which the documented events fell, who moved them, and what the tool did. A reconstruction from our briefs, not readiness scores taken along the way. The people are the heroes of this story; the tool carried the bags.
- 15 May
People first. The founder and the directors used the assistant on their own work every day before anyone else was asked to; the questions were ordinary management work. The first Pulse, in June, put 22 of 58 who answered at Crew level or above (38%). Nobody was told to adopt anything.
- Since May
Data on the logins we had. Tools connected with each person’s own accounts, no company-wide login, no new database; twenty-five of them by autumn. An index tells the assistant what to open first, so it reads the table of contents before the book. People still decide what goes into shared knowledge and what may be sent to a model.
- August
Rules without a permission system. Two repositories per department, one private per person, an inventory of every AI system in use, and a standing group that approves a new use of a model on company data. Access stayed where it already was.
- August to September
Plumbing at speed. A consumer product from the first line of code to live in 44 days, nine checks on every change, a person at five stops of the loop. A separate delivery case, not proof that the second brain caused the pace; the same loop now runs our own work.
- September
Backlog with a limit. Ideas ranked by value against the systems they touch; six weeks for a first working version. The Pulse moved to 34 of 66 (52%), then 53 of 77 (69%); different groups each time, all AI use, not a measured effect of the second brain. Only 33 of those 77 had it running.
- 7 October
Counted, and one gap left open. 237 changes to the shared core from six people, the founder third with 47. 608 pages of written knowledge, 138 recipes. Routine by our own card, five months after one laptop; Backlog lowest at 3 of 5, because we still start counting after the work. Your next step can be better than ours: write down the baseline before you begin.
Where to look, not what is true.
A self-assessment is a conversation starter with yourself. It is answered by the people who built the current setup, about the setup they built, so it leans generous. Ours does too: we loaded our answers so you have something real to compare with, not because they are audited. A budget decision or a board sign-off wants someone to read the data and the code, not the founder’s opinion of them.
What the card can do is point, and say whether a second brain would work here yet. If Rules is your lowest pillar, the three first moves are cheap and take a month. If Data is lowest, expect the first project to spend most of its time on plumbing, and plan for that. For what to do with the result, brief 009 turns five plain scores into a ninety-day plan; this brief is the diagnosis, that one is the plan.
Our numbers, and one idea we borrowed.
Each card says what kind of source it is and whether whoever published it sells something in this category.