agiletoai · companion sheet

Technical drawing of a canvas tool roll laid open, with a caliper, a micrometer, a dial indicator, a feeler gauge and a scriber each in its own pocket.

Nothing asked.Nothing collected.

Everything from the session, in the order you met it — and the nine-principle scorecard in the middle is live: score it here, print it as a one-page sheet, keep it.

This page makes no requests of you. No analytics, no cookies, no sign-in, no email field. Your scores are saved on this device and nowhere else.

01

The redefinition

The same two letters, four times.

The first pair points at a machine. Each one after moves the question closer to a person — which is where the risk, the ethics and the value have been sitting the whole time.

  1. 01
    Technical drawing of a sealed timber shipping crate, banded and corner-braced, with a blank plate where the contents should be listed.

    Artificial Intelligence

    The framing everyone handed you. Notice what it does: it points you — the leader — at the technology. Licenses. Tools. A procurement problem.

  2. 02
    Technical drawing of a hand steadying a workpiece in a drilling jig, the tool guided rather than replaced.

    Augmented Intelligence

    Your people and systems are already intelligent. The question is what you amplify — not what you replace.

  3. 03
    Cutaway technical drawing of a tachometer, its gear train and governor exposed behind a single sweeping needle.

    Accelerated Inference

    The language model itself doesn’t calculate — it infers. “There is a high probability this is true,” at a speed you have never had.

  4. 04
    Technical drawing of a hand pressing an inspection stamp, tethered by a wire to the finished part it signs off.

    Accountable Interpretation

    An accelerated inference has no meaning until a person interprets it — and that person has a name. Risk lives there. Ethics lives there. Value lives there.

    Same two letters. The first pointed at a machine. This one points at a person.

02

The mechanics

The bucket overflows quietly.

48 seconds, no narration — watch what nobody tells you about.

Silent by design. There is no alarm in the video because there is no alarm in the system. 00:48
Technical drawing of a screw press bearing down on a stack of sheets, the stack compressed to a fraction of its loose height.

Context is a bounded space, and it does not start empty — the model’s instructions, its guardrails, its tools and your institution’s policy layer are already in there before anyone types. Overfill it and it compacts: to make room, it squashes what it was already holding, and the fidelity you loaded is quietly lost. That is context rot, and everything produced after it is compromised while nothing in the output looks any different.

The failure mode to remember: your team believes the system knows everything they told it about how you work, long after it stopped remembering.

Access is not attention.

03

The scorecard

Are your teams building judgment — or building dependence?

Nine principles. Not one of them names a model. Read the row, then score it — honestly, as it is today, not as you intend it to be by Q4.

AUGMENT

What are we augmenting — and who decided?

People

Name the human whose judgment is amplified, not the headcount being saved.

Process

What context goes in is a management decision, not an individual’s.

Tools

Curated context beats more context. Plan for what it silently forgets.

1 · more noise, faster
— / 5 higher-order problems · 5

ACCELERATE

What pace can we defend?

People

Your verification capacity — not your build capacity — sets your real pace.

Process

Diagnose every miss by cause: augmentation, acceleration, or interpretation.

Tools

Measure what ships and survives, not what gets produced.

1 · the old loop, faster
— / 5 shrinking loops · 5

INTERPRET

Who is accountable for the answer?

People

A named expert signs off. “The AI said so” is not a sign-off.

Process

Source → Authority → Own-Eyes, before anything moves.

Tools

Pair the language model with things that actually compute.

1 · blind trust
— / 5 practised judgment · 5
Technical drawing of a chain in line: heavy end links, and one visibly thinner link that sets what the whole chain can carry.

Your lowest row isn’t a weakness. It’s your pace.

Score all three rows to read your pace.

04

The ladder · read bottom to top

Five rungs. You only need to climb one.

Rung one produces no asset. Every rung above it leaves something behind — and that something is the compounding. The right-hand column is the whole argument.

  1. 5

    DELEGATE

    The system works while you judge.

    Running work — what happens without you in the room.

    If you’re here, do this

    Hand one of your artifacts to a colleague and watch them try to use it. Whatever breaks is the real work.

  2. 4

    VERIFY

    It checks its own work, and shows receipts.

    Checks — what must be true before anything ships.

    If you’re here, do this

    Pick one recurring task you do on a schedule. Let the system draft it, and keep your check in front of the output.

  3. 3

    REMEMBER

    It knows what you decided, and why.

    Decisions — what was chosen · what was rejected · why.

    If you’re here, do this

    Write down one thing that must be true before work ships. Make the system check it and tell you.

Everything above this line outlives you.

The test: could someone else on your team use what you built yesterday? If no — you have a habit. If yes — you have an asset.

Technical drawing of a twin-gauge manifold: two dials reading the same line, so a claim can be checked against a second instrument.
  1. 2

    ANCHOR

    It knows who you are before you type.

    Context — who you are · how you work · what good looks like.

    If you’re here, do this

    Start one file of decisions — what you chose, what you rejected, why. Three lines each. Point the system at it.

  2. 1

    ASK

    Every chat starts from zero.

    nothing persists

    If you’re here, do this

    Write down who you are and what you’re working on, once. Paste it at the top of your next three conversations. That’s rung two.

Chats evaporate. Systems compound.

05

Leadership instruments

Four things to say out loud.

Not prompts. Sentences and questions for the room you are actually in — a steering committee, a review, a conversation with someone who wants to know why this is taking so long. Copy any of them.

The pace script

A defensible pace has to survive pressure from both directions, with the same posture.

“Why aren’t you moving faster?”
Because our verification capacity sets our pace — and here is what we’re doing to raise it, and who owns that.

“Why did that ship before it was ready?”
It didn’t. Here is the named person who validated it, and the check they ran.

Three sentences, filled from your scorecard:

1. “We’re moving at the pace of [your lowest row] — not the pace of the tool.”

2. “Here’s what we’ve already verified: [what’s working].”

3. “Here’s what unlocks the next step, and who owns it: [the unlock].”

The 3-Check

Three minutes per document, before it moves.

Source — where did this actually come from?

Authority — is that source the one that governs this?

Own-Eyes — has a human who knows this area read it?

The point is not the checking. It is that the ritual is defensible: when someone asks how you knew — and someone does ask — there is an answer, and the answer has a name attached to it.

Three questions for any AI-produced assessment

1. What is it actually reading? The thing — or the language about the thing.

2. Does it say the same thing twice? If not, it isn’t a control. It’s an opinion with a timestamp.

3. What did it not look at? It reports on the whole. It assessed a part.

Ask them in that order. None of the answers will arrive as an error message — silent failure is the shape of all of it.

The value test

First, positively. A meaningful outcome names someone who is better off:

Because of this, WHO can now DO WHAT — and we’ll know because WHAT CHANGED.

Can’t name the person? You have an output. Can’t name what changed? You have a hope. Can only name yourselves? You have an efficiency — which may be fine. Say so.

Then, negatively. AI is often better at producing the evidence of work than at doing the work:

Could a system satisfy this definition of success while producing nothing of value?

If yes, you have defined an artifact, not an outcome. “The process was followed” used to be a good proxy for value. That link is now severed.