Let an agent complete the survey

Help me fill out a short industry survey on how we actually use AI with our data and tools — shadow AI, how AI reaches our data, whether our agents really work. Everything you need is below: do NOT browse or web-search.

1. DRAFT an answer to every question from what you know about my company. "pick 1" = one listed option; "pick any" = a list; "1-5" = integer; "text" = free. Don't invent numbers — leave true unknowns blank. (* = please don't leave blank.)
2. Show me all answers as a TABLE; flag blanks/guesses.
3. Offer up to 5 quick questions to fill the most valuable gaps (required blanks first, then shadow_tools, data_exposed, data_access_method, demo_vs_reality). Stop when I say done.
4. Only after I confirm: build a flat JSON {question_id: answer} (pick-any = JSON array; also add "ref":"agent-web","fill_mode":"agent"), base64url-encode it (UTF-8), and give me this clickable link to review & submit in one tap — do NOT POST it yourself:
https://rig.so/survey/submit#<base64url-json>

QUESTIONS:
- company_name (text)
- respondent_role* (pick 1): Founder / CEO | CTO / VP Engineering | Engineering Manager / Lead | Individual contributor (engineer) | Data / Analytics | Security / IT | Product | Operations | Other
- company_size* (pick 1): 1–10 | 11–50 | 51–200 | 201–500 | 501–1,000 | 1,000+
- funding_stage* (pick 1): Bootstrapped | Pre-seed | Seed | Series A | Series B | Series C | Series D+ | Public | N/A
- industry* (pick 1): SaaS / B2B software | Fintech | Healthcare | E-commerce / Retail | AI / ML | Developer tools | Media | Gaming | Government / Education | Other
- eng_team_size* (pick 1): 1–5 | 6–20 | 21–50 | 51–150 | 150+
- pct_ai_users* (pick 1): 0% | Under 25% | 25–50% | 50–75% | 75–100%
- approved_tools* (pick any): Claude Code | OpenAI Codex | Cursor | GitHub Copilot | ChatGPT | Claude (claude.ai) | Gemini | Windsurf | Devin | Internal / in-house tools | None
- shadow_tools (pick any): Claude Code | OpenAI Codex | Cursor | GitHub Copilot | ChatGPT | Claude (claude.ai) | Gemini | Personal ChatGPT / Claude accounts | Other unsanctioned tools | None — we're disciplined
- data_exposed (pick any): Customer PII | Production data | Secrets / API keys | Proprietary source code | Internal docs / strategy | Financials | None that I know of | I'd rather not say
- ai_policy (pick 1): No policy | Informal guidance | Written policy, loosely followed | Written policy, actually enforced
- data_access_method (pick any): Copy-paste into the chat | Direct database credentials | Raw API keys in config | MCP servers | A governed data / context platform | They don't touch internal data
- mcp_setup (pick 1): Individuals wire their own ("YOLO") | A curated / central list | Through a governed platform with access control | We don't use MCP
- access_control (pick 1): No one, really | Each developer | Central data / platform team | Enforced by RBAC in a platform
- data_exposure_incident (pick 1): Yes | A near miss | No | We wouldn't know
- data_readiness* (1-5)
- ai_answer_accuracy (pick 1): Almost always | Usually | Coin flip | Often wrong | We don't let it answer data questions
- trust_ai_numbers (pick 1): Yes | Only for low-stakes | No
- logic_documented (pick 1): Most of it | Some | Almost none | It's in people's heads
- agents_in_prod* (pick 1): None / all demos | 1–2 | A handful | Many | It's how we operate now
- agent_babysitting* (pick 1): Set-and-forget | Occasional nudges | Weekly fixes | Constant | We abandoned one because of this
- agent_breakage (pick any): Wrong data / bad joins | Hallucinated facts | Stale or missing context | Permissions / access | Integration brittleness | Cost | Nothing major
- bad_data_decision (pick 1): Yes, costly | Yes, minor | No | We wouldn't know
- demo_vs_reality (text)
- spend_direction_12mo* (pick 1): Increase a lot | Increase | Stay flat | Decrease | Unsure
- invest_next (pick any): More tools / seats | Data foundations | Governance & access control | Custom agents | MCP infrastructure | A governed data platform