--- name: al-663-when-every-customer-needs-different-fields-let-an-ai-agen description: "Working Surface (Nov 5, 2026 issue; summarizing a Japanese LayerX Ai Workforce engineering write-up by suna dated around Nov 4) describes a trial design where an AI agent designs and creates the PostgreSQL tables that store each customer's business data for work such as invoice checks and contract review. Customer field needs often appear only after work starts; the old flexible storage made types hard to enforce and forced the agent to scan every record to match invoices to purchase orders. In the new design the agent proposes real tables, asks clarifying questions, and creates them only after a person approves. The agent may add tables and columns but not delete or rename; every record change is shown first, approved by the user, and reported back with prior values. Working Surface notes there is no release date; treat this as an engineering account of a trial, not a shipped product promise. Use when the user wants help with human-approved agent schema design (operations) for a…" --- # When every customer needs different fields, let an AI agent propose the database tables from the conversation--but only create them after a person approves, and only allow adding tables and columns, never deletes Working Surface (Nov 5, 2026 issue; summarizing a Japanese LayerX Ai Workforce engineering write-up by suna dated around Nov 4) describes a trial design where an AI agent designs and creates the PostgreSQL tables that store each customer's business data for work such as invoice checks and contract review. Customer field needs often appear only after work starts; the old flexible storage made types hard to enforce and forced the agent to scan every record to match invoices to purchase orders. In the new design the agent proposes real tables, asks clarifying questions, and creates them only after a person approves. The agent may add tables and columns but not delete or rename; every record change is shown first, approved by the user, and reported back with prior values. Working Surface notes there is no release date; treat this as an engineering account of a trial, not a shipped product promise. ## Inputs to collect first Ask the user for anything below that they haven't already given you. Don't guess missing facts. - E.g. invoice check / contract review - E.g. invoice to PO - Your own material (text, notes or data) ## Instructions Do the task below with the user's inputs in place of each [bracket]. Help me let an AI agent design storage for a workflow where each customer needs different fields. Workflow: [e.g. invoice check / contract review]. Examples of fields we've seen: [list]. Rules I want: human approves schema; agent may add tables/columns only; no deletes or renames; every row change needs approval and must show old values. 1) Interview me with up to 8 questions to learn entities, keys, and which fields must be typed vs free text. 2) Propose a first schema (tables, columns, types, indexes) in plain language and as SQL CREATE statements. 3) List what you are unsure about and will not create until I answer. 4) Write the approval card text I should see before any CREATE runs. 5) Give a safe migration plan from a flexible/EAV store to these tables, including how matching [e.g. invoice to PO] gets faster with real columns. Do not run DDL or change data yourself. ## Safety checks (share these with the user) - Check every figure, price and total against your own records before anything is sent, posted or booked. A person approves refunds, discounts, payments and price changes. - Don't let AI make the final call on hiring, pay, performance or discipline. Use it to organise information, and have a person decide using criteria you can explain. - Anything legal, tax or compliance-related is a first draft for a qualified person to check. It is not professional advice, and clause references and deadlines must be checked against the source. - Remove names, contact details, account numbers and other personal data the task doesn't need before you paste, and check your AI tool's data and training settings. - Say clearly which parts need a person to review or approve before anything is sent, paid, posted or filed. ## About this skill - Works with: LayerX Ai Workforce (trial pattern), ChatGPT, Claude, PostgreSQL - Based on this real-world example: https://workingsurface.ai/links/2026-11-05-02/ - Free prompt from AgentPlaybook Real Estate: https://agentic-library.netlify.app/prompt/al-663-when-every-customer-needs-different-fields-let-an-ai-agent-p/ - To install in an agent that supports Agent Skills, save this file as SKILL.md in a folder with the same name as `name` above (for example in your skills directory). Install steps for Claude.ai, Claude Code, Cursor and Codex: https://agentic-library.netlify.app/skills/