Agent mode. Let it score you.

You are scoring how AI-native a company actually is, on behalf of someone who works there. AI-native is an operating model where agents do the work and people make the calls.

Use the company context you can actually access: documents, wikis, tickets, CRM, code, conversations. For each question, choose the answer best supported by what you find. If you cannot see enough context to answer a question, score it 0, because inaccessible context is itself the finding. Do not flatter.

Q1 (Workflow legibility): Pick the workflow that would hurt most if it broke tomorrow. If a new hire had to run it next week, what would they work from?
  0. Whatever lives in people’s heads. They’d shadow someone for a month.
  1. Scattered notes, an outdated Notion page, a few Slack threads.
  2. A written SOP exists and is mostly accurate.
  3. A documented workflow with named owners, triggers, exceptions and an escalation path.

Q2 (Workflow legibility): How does work actually move between your people and tools?
  0. Forwarded emails, Slack DMs, and someone remembering to chase.
  1. Real tools, but people copy data between them by hand.
  2. Mostly structured tools. Handoffs are defined but enforced by habit.
  3. Defined handoffs inside systems that show where any piece of work is right now.

Q3 (Context readiness): Your most experienced operator resigns tomorrow. What walks out the door with them?
  0. The operating manual. It exists only in their head.
  1. A lot. Some of what they know is written down somewhere.
  2. A few painful weeks. Most decision rules are documented.
  3. Very little. Decision criteria, exceptions and edge cases are explicit and maintained.

Q4 (Context readiness): Could a competent outsider make your routine decisions from written rules alone? Approve or reject, route, prioritise.
  0. No. “It depends”, on things we’ve never put into words.
  1. Partially, for the simple cases.
  2. For most decisions, with some effort.
  3. Largely. Thresholds, routing logic and approval rules exist as explicit criteria.

Q5 (Data and systems): Two of your systems disagree about the same customer, order or ticket. Which one is right?
  0. Unclear. We’d ask around.
  1. Depends who you ask. The spreadsheet usually wins.
  2. We know the system of record for most things.
  3. One defined source of truth per domain. Everything else syncs from it.

Q6 (Data and systems): Can your tools talk to each other?
  0. Mostly closed tools. Manual exports, if anything.
  1. Some Zapier-style glue. It breaks quietly.
  2. Core tools are connected. The gaps are filled by hand.
  3. API access across the core stack. Data moves without a person routing it.

Q7 (Human–agent boundaries): If you deployed an AI agent into one workflow tomorrow, could you state exactly what it may do alone, what it only drafts, and what it must never touch?
  0. We’ve never thought in those terms.
  1. Rough intuitions. Nothing written.
  2. We could define it per workflow if pushed.
  3. Yes. We already define explicit boundaries, including what triggers an escalation.

Q8 (Human–agent boundaries): Who can access what in your stack, and would an AI agent inherit sane permissions?
  0. Everyone has access to everything. Shared logins exist.
  1. Some role-based access, plenty of exceptions.
  2. Mostly clean. An agent could be scoped, with work.
  3. Role-based, auditable access. Scoping an agent to least privilege would be straightforward.

Q9 (Production controls): An automated system in your operations makes a mistake. How do you find out?
  0. A customer or employee complains. Eventually.
  1. Someone might notice, if they happen to be paying attention.
  2. Spot checks and ad hoc review.
  3. Monitoring, logs and failure thresholds surface it before anyone stumbles on it.

Q10 (Production controls): How do you decide whether an automation’s output is good enough to act on?
  0. Gut feel.
  1. We’d eyeball a few examples.
  2. Defined acceptance criteria, for some workflows.
  3. Test sets and evals with acceptance thresholds, before anything touches production.

Q11 (Adoption and ownership): How does your team actually use AI today?
  0. A few people paste things into ChatGPT occasionally.
  1. Most people have subscriptions. Usage is individual and invisible.
  2. Real individual productivity gains, and some shared prompts and templates.
  3. Teams own repeatable AI workflows that keep working when one person is on holiday.

Q12 (Adoption and ownership): Who owns a workflow end to end, including the authority to improve it?
  0. Nobody. Workflows are everyone’s, which means no one’s.
  1. The founder. By default. For all of them.
  2. Informal owners exist for the big ones.
  3. Named owners who can change the rules without filing an engineering ticket.

Q13 (Business impact): What has AI measurably changed in your business so far?
  0. Nothing measurable. Vibes.
  1. People say they save time. We can’t see it in any number.
  2. One or two metrics actually moved, such as response time or cycle time.
  3. Tracked impact on cost, cycle time or revenue, attributable to specific workflows.

Q14 (Business impact): How do you pick what to automate next?
  0. Whatever the last impressive demo was.
  1. Loudest pain wins.
  2. A rough sense of volume and pain.
  3. Scored against frequency, measurable pain, and whether the context exists to do it safely.

Reply with, in this order:
1. One line exactly like this, with your fourteen chosen digits in question order: AYAIN2:2,1,3,0,2,1,2,3,1,2,0,1,3,2
2. For each of the seven dimensions (Workflow legibility; Context readiness; Data and systems; Human–agent boundaries; Production controls; Adoption and ownership; Business impact), one sentence naming the strongest evidence behind its score.
3. Which questions, if any, you scored 0 for lack of accessible context, named explicitly.
4. This link, so the user can open their full result, with the fourteen digits replaced by your own in question order and no commas: https://areyouainative.com/agent#a=21302120130121&by=agent