# Wizeb AI-Native Index — questions for AI agents

Score a company on the 30 signals of an AI-native business (six dimensions, five signals each).
Answer every question about the user's company. **If you are not sure about an answer, ask the user — never guess.**

## Scoring
- Every option shows its points in brackets (0–4). Add up the points of the 30 chosen options.
- Index = round(total points ÷ 120 × 100).
- Dimension score = round(dimension points ÷ 20 × 100).
- Tiers: **Observer** ≥ 0 · **Experimenter** ≥ 25 · **Operator** ≥ 50 · **Native** ≥ 70 · **Frontier** ≥ 85.

## Delivering the result
Show the user the index, tier, six dimension scores, their strongest signals, and the lowest-scoring signals.
Then give them this link — it opens the visual scorecard with their answers pre-loaded and lets them request the PDF report by email:

`https://wizeb.com/ai-native-index?a=` followed by 30 digits — the chosen option number for questions 1 to 30 in order (e.g. `?a=345545545455442324344353344434`).

Tools that can call MCP or HTTP can instead use the MCP server at `https://wizeb.com/mcp` (tools: `get_index_questions`, `score_index`, `email_index_report`).

## Foundation — Can machines read how your business works?

**1. Process blueprint** — Do you have a documented, function-by-function map of how work actually flows through your business?
1. No — our processes live in people's heads [0 pts]
2. Partially — a few departments keep SOPs [1 pts]
3. Most functions are documented, but stale and rarely updated [2 pts]
4. Every function is mapped and kept current [3 pts]
5. Every function is mapped, current, and structured so AI agents can execute against it [4 pts]

**2. Intelligence layer** — Where does your company's knowledge live — the data, the documents, and the rules for how you do business?
1. Scattered across inboxes, drives, and individual tools [0 pts]
2. A shared drive or wiki, mostly unstructured [1 pts]
3. A data warehouse for numbers; documents live somewhere else [2 pts]
4. One unified layer combining data and documents, queryable by people [3 pts]
5. A single source of truth that both people and AI agents query and act on [4 pts]

**3. Context as code** — How are your architecture documents, conventions, and "how we do things here" maintained?
1. They don't really exist [0 pts]
2. Written once, rarely updated [1 pts]
3. Updated on a schedule by a designated owner [2 pts]
4. Versioned alongside the work — updating them is part of every change [3 pts]
5. Versioned, current, and written so agents use them as their operating context [4 pts]

**4. Agent harness** — When your AI tools need context on a project, how do they get it?
1. We paste things in manually every time [0 pts]
2. Shared prompts or templates [1 pts]
3. A knowledge base connected to our AI tools [2 pts]
4. Structured docs with metadata that tools navigate automatically [3 pts]
5. A harness that traverses dependencies and loads only what each task needs [4 pts]

**5. Capture everything** — How much of your business activity — meetings, calls, decisions, work output — is captured in a form AI could use?
1. Almost none of it [0 pts]
2. Some meetings get recorded or summarised [1 pts]
3. Most meetings and key communications are captured [2 pts]
4. Nearly everything is captured and searchable [3 pts]
5. Everything is captured by default and feeds directly into AI-enabled work [4 pts]

## People — Is AI a daily habit or a side project?

**6. Daily driver** — What share of your team uses an AI assistant — ChatGPT at Work, Claude Cowork, Grok Bot — as a genuine daily driver?
1. Almost nobody [0 pts]
2. A few enthusiasts [1 pts]
3. Around half, informally [2 pts]
4. Most of the company, on company-provided tools [3 pts]
5. Everyone, on a sanctioned harness with shared workflows [4 pts]

**7. Everyone builds** — Who builds AI workflows and tools at your company?
1. Nobody yet [0 pts]
2. Only engineers, or an outside vendor [1 pts]
3. Engineers plus a few power users [2 pts]
4. People across most functions build their own [3 pts]
5. Everyone builds — including the C-suite [4 pts]

**8. Self-disruption** — Has your leadership actively tried to disrupt your own business model with AI in the last twelve months?
1. No — we're focused on protecting what works [0 pts]
2. We've discussed it, but taken no action [1 pts]
3. We ran one or two experiments [2 pts]
4. Yes — there's a funded effort to disrupt ourselves first [4 pts]

**9. Reinvention cadence** — How often do you fundamentally rethink a workflow, rather than improve it incrementally?
1. Rarely — we change when forced to [0 pts]
2. Every few years [1 pts]
3. Annually [2 pts]
4. Every quarter or two, for selected workflows [3 pts]
5. Every quarter — we're willing to throw everything away and rebuild [4 pts]

**10. First & final mile** — Where do humans sit in your core processes today?
1. Humans do every step [0 pts]
2. Humans do most steps; tools assist [1 pts]
3. Automation handles routine steps; humans review most output [2 pts]
4. Humans handle setup and final review; the middle is automated [3 pts]
5. Human judgement is reserved for the first and final mile of most processes [4 pts]

## Engineering — Do agents build, or just autocomplete?

**11. Agent-native development** — How is software built at your company today?
1. We don't build software, or it's fully outsourced [0 pts]
2. Engineers write code; AI autocomplete at most [1 pts]
3. Engineers lean heavily on AI coding assistants [2 pts]
4. AI agents write and test; humans review and ship [3 pts]
5. Fleets of agents plan, write, test, review, and ship — humans set intent and acceptance criteria [4 pts]

**12. Spec / implementation split** — Can non-technical staff describe what they need in a form AI can turn into a working technical plan?
1. No — requirements go through meetings and tickets [0 pts]
2. We have some requirement templates [1 pts]
3. Structured specs exist, but engineers translate them by hand [2 pts]
4. Specs are written in a format agents can draft implementation plans from [3 pts]
5. Yes — specs are cleanly separated from implementation and agents build the plan [4 pts]

**13. Skills distribution** — How do you keep AI agents behaving consistently across developers and workflows?
1. We don't — everyone prompts their own way [0 pts]
2. A shared prompt library [1 pts]
3. Documented conventions people are expected to follow [2 pts]
4. Reusable, versioned skills that agents trigger automatically [3 pts]
5. A skills distribution system that enforces consistency and optimises tokens [4 pts]

**14. Cost per accepted PR** — Do you measure the cost of AI-assisted engineering output — for example, cost per accepted pull request?
1. No [0 pts]
2. We track total AI spend only [1 pts]
3. We track cost per task, loosely [2 pts]
4. Yes — cost per accepted PR is a tracked metric [3 pts]
5. Yes, and we actively drive it down through token efficiency [4 pts]

**15. Plan heavy, execute light** — How do you match AI model capability to the task at hand?
1. We use one model for everything [0 pts]
2. We pick the biggest model we can afford [1 pts]
3. We use cheaper models for some simple tasks [2 pts]
4. Frontier models for planning, cheaper models for execution — most of the time [3 pts]
5. By design: high-effort models plan, fast and cheap models execute [4 pts]

## Operations — Do your workflows learn while you sleep?

**16. Continuous finance** — How often is your financial forecast reset against actuals?
1. Annually, if at all [0 pts]
2. Quarterly [1 pts]
3. Monthly [2 pts]
4. Weekly [3 pts]
5. Continuously — record-keeping runs live and forecasts re-baseline on a very tight cadence [4 pts]

**17. Embedded financial models** — When a decision is made anywhere in the business, is its financial impact visible at that moment?
1. No — finance is consulted afterwards [0 pts]
2. Only for major decisions [1 pts]
3. Managers have dashboards they check [2 pts]
4. Financial models are embedded in the tools most teams use [3 pts]
5. Yes — financial models sit inside the operating layer and drive everyday reasoning [4 pts]

**18. Citizen developer SDLC** — If a non-technical employee has an idea for an automation, what happens next?
1. It goes on a list nobody owns [0 pts]
2. They ask IT or engineering, and wait [1 pts]
3. They build something in a no-code tool, ungoverned [2 pts]
4. They can build and ship it with guidance from a central team [3 pts]
5. They take it from idea to production themselves — with governance, access, and versioning built in [4 pts]

**19. Self-improving workflows** — Do your non-engineering workflows — sales, support, operations — get better on their own?
1. No — they change when someone rewrites them [0 pts]
2. We review performance and adjust occasionally [1 pts]
3. We A/B test some workflows [2 pts]
4. Workflows are measured against metrics and tuned regularly [3 pts]
5. Yes — closed loop: every run learns from performance metrics or internal evals [4 pts]

**20. AI ROI framework** — How do you decide which AI investments get funded?
1. Ad hoc — whoever asks loudest [0 pts]
2. Case by case, on gut feel [1 pts]
3. Business cases with expected ROI [2 pts]
4. A portfolio approach with a defined experiment → scale → optimise path [3 pts]
5. A formal framework: experiment, scale, optimise phases across infrastructure, innovation, and efficiency bets [4 pts]

## Growth — Is your go-to-market running at machine scale?

**21. Model routing** — How do you choose which AI model handles each task across the business?
1. We don't — one vendor, one model [0 pts]
2. Teams pick whatever they like [1 pts]
3. A shortlist of approved models per use case [2 pts]
4. A routing layer (e.g. OpenRouter) that balances cost and quality [3 pts]
5. Routing that optimises cost-per-successful-task across the entire business [4 pts]

**22. Creative at scale** — How is your paid marketing creative produced and tested?
1. A handful of human-made ads, rarely tested [0 pts]
2. Human-made ads with some A/B testing [1 pts]
3. AI drafts variants; humans test a dozen or so [2 pts]
4. Hundreds of AI-generated variants tested before scaling spend [3 pts]
5. Agent swarms deploy thousands of creatives for testing before human-made ads get budget [4 pts]

**23. AEO / SEO engine** — How does your content and search engine — classic SEO plus AI-answer optimisation — actually run?
1. We publish when someone has time [0 pts]
2. A content calendar, written by hand [1 pts]
3. AI-assisted writing on a monthly cadence [2 pts]
4. An engine that audits and rewrites weekly [3 pts]
5. A weekly engine that audits, rewrites, generates AEO/SEO content — and measures whether it worked [4 pts]

**24. Fine-tuned models** — For your highest-volume processes, do you fine-tune models on your own data?
1. No — off-the-shelf only [0 pts]
2. We've experimented with prompting on our data [1 pts]
3. We use retrieval over our data, no fine-tuning [2 pts]
4. We've fine-tuned a model for one process [3 pts]
5. Yes — RL environments plus first-party data tune open models to SOTA quality at low cost [4 pts]

**25. Evals as infrastructure** — When a new AI model is released, what happens at your company?
1. Nothing — we read the news [0 pts]
2. Someone tries it informally [1 pts]
3. We test it on a couple of use cases [2 pts]
4. We run it through a standard set of tests before adopting [3 pts]
5. It runs through our eval apparatus — cost and performance against every core process [4 pts]

## Governance — Can you trust — and trace — what your agents do?

**26. Guardrails first** — When an AI agent accesses company data, what permissions does it have?
1. Whatever the API key allows — usually broad [0 pts]
2. Per-agent permissions, set by hand [1 pts]
3. Role-based permissions at the application level [2 pts]
4. The agent inherits the requesting user's permissions [3 pts]
5. It inherits the requester's permissions, enforced in the data layer — guardrails before features [4 pts]

**27. Earned autonomy** — How much autonomy do your AI agents have — and how do they earn more?
1. We don't run agents [0 pts]
2. Agents suggest; humans do everything [1 pts]
3. Agents act, with approval on every step [2 pts]
4. Some agents act alone in narrow areas, based on track record [3 pts]
5. A formal ladder — observe, suggest, act with approval, act alone — gated by evals and feedback [4 pts]

**28. Traceability** — Can you trace any AI output back to the prompt, model, data, and person who approved it?
1. No [0 pts]
2. Partially, by digging through logs [1 pts]
3. Yes, for some systems [2 pts]
4. Yes, for most systems [3 pts]
5. Yes, everywhere — and that trace is the training signal for improvement [4 pts]

**29. Agentic security** — How does your security posture handle AI-driven threats?
1. Traditional tooling only [0 pts]
2. A managed provider, nothing AI-specific [1 pts]
3. AI-assisted detection with human response [2 pts]
4. Agentic security that detects and responds — AI fighting AI [4 pts]

**30. Legal, HR & IT in lockstep** — How do Legal, HR, and IT relate to your AI initiatives?
1. They're the reason things stall [0 pts]
2. Consulted late, usually to say no [1 pts]
3. Involved in reviews, with some delay [2 pts]
4. Embedded early — they enable more than they block [3 pts]
5. In lockstep with the AI owners: the business is covered without slowing transformation [4 pts]