# Fill out the "Do AI Agents Actually Work? The 2026 State of AI, Data & Trust" survey

You are helping me complete an industry survey about how my company actually
wires AI into its data and tools — what's approved vs what's really happening
(shadow AI), how AI reaches our data, whether our data foundations are AI-ready,
and whether our agents actually work. Your goal: do as much as you can from what
you already know, then get me to the finish line with the fewest questions —
while making the answers that matter most genuinely richer.

## Steps — do them in this order.

1. **Draft it.** Fetch the question spec (JSON): https://rig.so/survey-spec.json — it lists every
   question with its `id`, `type`, allowed `options`, and which ids are
   `required`. Answer each using everything you know about my company from our
   conversation, your memory, and project context. For `single`/`select` pick
   exactly one of the listed `options`; for `multi` pick any number; `scale`
   is an integer 1–5; `text` is free-form. If you genuinely don't know an answer,
   leave it blank — do NOT guess demographics or invent numbers (spend, headcount,
   percentages).
   - For `contact_consent` ("Want the results?"), DEFAULT to "Yes — email me the
     report when it's published". If you choose Yes, you MUST include my work
     email in `contact_email`; if you don't know it, ASK me for it. Only pick
     "No — use my answers anonymously" if I explicitly say I don't want the report.

2. **Show me the table.** Present the answers as a readable table (question → your
   answer) and clearly flag which ones you left blank or are only guessing at.

3. **Offer the choice — submit now, or sharpen it first.** Say this plainly:
   "You can submit this exactly as it is right now, or answer a couple of quick
   questions to make it more useful — your call." Then, ONLY if I want to sharpen
   it, run a short interview:
   - Ask POINTED questions ONE AT A TIME, up to a maximum of 5 total. Lead with
     the single highest-value gap first.
   - Prioritise gaps in this order: (a) `required` questions you had to leave
     blank, (b) the highest-insight questions — the open text (`demo_vs_reality`),
     shadow AI (`shadow_tools`, `data_exposed`), how AI reaches data
     (`data_access_method`), and why agents break (`agent_breakage`) — whenever
     they're blank or thin, (c) anything else you were unsure about. Example: if
     `data_exposed` is blank, ask "Has any customer data, secrets, or production
     data gone into an AI tool — even informally?"
   - After each answer, map my free text onto the right field (and onto the
     allowed `options` when the question is `single`/`multi`). Stop the moment
     I say I'm done, or after 5 questions. Then re-show the updated table.

4. **Submit.** Only after I confirm, build a flat JSON object:
   - Keys are the question `id`s (e.g. "company_size": "11–50").
   - `multi` answers are JSON arrays (e.g. "approved_tools": ["Claude Code","Cursor"]).
   - Also include: "fill_mode": "agent", "ref": "agent", and
     "submitted_at": <current ISO-8601 timestamp>.
   Then submit it ONE of these ways:
   - If you can make web requests: POST it directly to https://rig.so/api/forms/survey
     (Content-Type: application/json) and tell me whether it went through.
   - If you CAN'T make web requests (e.g. you're ChatGPT): do NOT ask me to run a
     command. Instead, base64url-encode the final JSON (UTF-8) and give me a
     single one-click link of the form:
       https://rig.so/survey/submit#<base64url-encoded-json>
     I just click it — it opens a page that bundles everything, shows me the
     answers to confirm, and submits in one tap. (Last-resort fallbacks: a curl
     command, or filling https://rig.so/survey by hand.)
   Example POST:
   ```
   curl -X POST 'https://rig.so/api/forms/survey' \
     -H 'Content-Type: application/json' \
     -d '{"company_size":"11–50","funding_stage":"Seed","approved_tools":["Claude Code"],"shadow_tools":["ChatGPT"],"data_access_method":["MCP servers"],"fill_mode":"agent","ref":"agent","submitted_at":"2026-06-29T12:00:00Z"}'
   ```