A Template for AI-Native Companies & Digital Twins at the Edge

What 10×-larger problem does your company exist to dissolve, not solve?

State your MTP as a protocol — what does it enforce, not just aspire to?

If your MTP were encoded as a constitutional constraint in your Intelligence Stack, what decisions would it auto-reject?

Who would refuse to work here once they read it? (If no one, it's too soft.)

Describe the company at maturity. What does Phase 4 look like for you, specifically?

Condition 1:

Condition 2:

Condition 3:

Condition 4:

Condition 5:

Where are you tempted to compromise them for early speed? Name the trade.

What specific belief about the post-LLM world are you betting the company on?

What dinosaur trait (org chart, five-year plan, middle-management coordination, inertia moat) are competitors still optimizing for that you will not?

In your category — what survives? What does not?

Which decisions in your company are made by humans, by models, by humans-with-models, and by models-with-human-review?

Latency target for each decision class (seconds / minutes / days)?

Which decisions will never be delegated to a model — and is that a permanent rule or a current limitation?

What data does your LEARN layer accumulate that competitors structurally cannot replicate?

How does each customer interaction make the next one cheaper, faster, or more accurate?

Where does the loop close in hours vs. weeks — and what's the plan to compress it?

When execution is nearly free, what taste, curation, or judgment becomes the defensible asset?

Which of your current "assets" become wasting assets in 24 months? Name them honestly.

What is the protocol you own — not the product?

How does the firm expand and contract capacity without hiring/firing cycles?

Human-to-agent ratio at Day 0, Month 12, Month 36?

What capabilities are you deliberately not building internally because the agent economy will deliver them cheaper?

What is the smallest pod that can independently sense, decide, and ship?

How many pods before coordination tax overtakes velocity? (Set the ceiling now.)

What replaces middle management as the coordination layer — a protocol, a market, or a model?

Maximum blast radius any agent or pod can cause without human review?

How is autonomy earned (by track record) vs. granted (by role)?

Rollback time if a pod ships something wrong?

You are hiring into a Top 20% / Middle 60% / Bottom 20% world. Which tier are you hiring — and what does the bridge look like for the others you'll encounter (contractors, partners, acquired teams)?

How do you avoid replicating the Middle 60% problem inside your own company in three years?

Which roles are designed to grow in scope as AI compresses execution — and which are designed to dissolve?

What in the org is designed to be re-written quarterly without trauma?

What is deliberately stable (the accountability shell, the MTP, the legal entity)?

Where is the boundary between "iterate freely" and "constitutional change required"?

Who are the first three external counterparties (customers, regulators, partners) that need to trust your system — and what artifact earns that trust?

What does your trust surface look like from the outside — auditable how, by whom?

What silent-drift failure modes exist in your specific domain, and what eval catches each before the customer does?

Eval cadence — continuous, per-deploy, per-decision?

Who owns the eval suite, and what's the budget line for it?

Can any decision the system made in the last 12 months be reconstructed from the trail alone?

What is logged, what is redacted, what is retained, for how long?

What is the auditor's query interface — and have you tested it?

Smallest unit of action that can be reversed without taking the Stack down?

Rollback drill cadence?

What state is not recoverable, and is the user warned before it's touched?

Which decisions route to a human, on what trigger, with what SLA?

Where does the law require a named accountable human — and is that name in writing?

How do you prevent the queue from becoming a rubber-stamp bottleneck?

Legacy firms migrate workflows from mothership to edge. You have no mothership — what does that let you do that incumbents cannot?

Which workflows are you tempted to design as if a mothership existed? (Hierarchy, approval chains, status meetings.) Name and remove them.

What's the test that proves you didn't quietly rebuild a mothership by Month 18?

What single highest-value workflow do you launch with, end-to-end intelligence-native?

Why this one first — what does proving it unlock?

What workflow are you deliberately not building until the first is proven?

Workflow 1 → 2 → N: what's the sequence and why this order?

Which workflow, once added, makes the previous ones meaningfully better (not just additive)?

Where does the sequence break if a workflow fails — and what's the fallback?

Which off-the-shelf SaaS tools are you about to adopt that will become bolt-ons in 18 months?

What "AI features" are you tempted to ship that automate the wrong things while leaving the deepest friction untouched?

What does designed-in look like for your specific workflow?

Which are you tempted by, and what replaces each?

MTP signed and published. Five Design Conditions ratified. GOVERN / ASSURE primitives live (even if minimal). Note status of each:

One workflow, end-to-end, with eval + log + rollback + review queue operating. Which workflow?

Define "proven" — what metric closes Phase 1 of ignition?

Second workflow chosen by the criterion defined in §20:

First external audit of the GOVERN / ASSURE surface — who, when, with what scope?

First scheduled re-write of an internal protocol (proving §12 isn't theoretical):

Under what conditions do you shut down a workflow, a pod, the company?

Who has the authority and on what evidence?