🏕️ Welcome to Airtable Summer Camp: five days, five product showcases, Aug 10–14

Summer Camp Day 1: Watch an app build itself

First day of camp, and we're not starting small.

Today Airtable showed what happens when your AI tools don't just talk about your work. They build it, inside Airtable:

• Automations via MCP: your AI assistant can create real Airtable automations. Codified, centralized workflows your whole team can see and govern, instead of prompts and routines trapped in one person's chat session.
• Interface creation via MCP: AI-generated interfaces built on your live base data, so the deliverable your agent produces is always the current one. No more stale documents floating around drives and Slack threads.
• Skills library: frontier models get deep Airtable expertise out of the box, so they get more done with less hand-holding.

Why it matters: until now, AI-assisted building has been a solo sport. The work happened in personal Claude or ChatGPT sessions that teammates couldn't see or control. Day 1 moves that work into Airtable, where it becomes shared, inspectable infrastructure.

Read the deep-dive: Airtable's MCP server now builds interface and automations

Miss a day? The whole week lives here: https://lp.airtable.com/summer-camp 

Over to you:

• What's the first workflow you'd hand to an agent to build?
• If your AI could generate an interface from your base today, what's the first one you'd ask for?
• Builders who already connected Claude, ChatGPT, or Cursor to a base: how's it going?

Tomorrow: using Airtable without opening Airtable. 

Summer Camp Day 2: Use Airtable without opening Airtable

Day 2 flips the usual direction. Instead of going to your data, your data comes to where you already work:

• Form submissions via MCP: your AI assistant can submit and collect data through Airtable forms, following the structure your team set up, so data lands in your base the way your business actually models it.
• Interface reading via MCP: your assistant reads your live interfaces, so it always works from the most up-to-date view of your data. Not a stale export, not a screenshot from last week.

Why it matters: the hardest part of putting AI to work isn't the model. It's getting the model the right data in the right shape, with governance intact. Airtable is where companies model how their business actually works; Day 2 lets frontier models plug straight into that, from whatever tool you already live in.

▶️ Watch the Day 2 showcase: https://www.airtable.com/articles/mcp-forms-and-interface-permissions

🔗 Catch up on the week: https://lp.airtable.com/summer-camp

Over to you:

• Which AI tool is your daily driver, and what's the first Airtable workflow you'd run from it?
• What data-collection process at your org would instantly improve if your assistant could submit it through a proper form?

Tomorrow: the midweek main event. A new look for your base. ⛺

Summer Camp Day 3: Meet Canvas - Build an app on top of your base.

The midweek marquee. Today we got a first look at Canvas.

Describe the app you want in plain language, and Canvas builds the whole thing from the conversation — the base, the workflows, and the surface — zero configuration, no pre-existing base required.

Why it matters: vibe coding made it trivially easy to spin up an app — but those apps usually lack the structured data and governance to survive at enterprise scale, or even to be maintained next quarter. Canvas keeps the speed and drops the fragility: prompt-driven building that lands on real Airtable foundations from the first sentence.

🎬 Watch the video: https://www.airtable.com/articles/apps-dashboards-and-decks 

✍️ Join the Canvas waitlist: https://airtable-labs.com/canvas/ 

🔗 The full week: https://lp.airtable.com/summer-camp 

Summer Camp Day 4: Top speed, full control

Every day this week has shown agents doing more. Day 4 answers the question every admin and security team asks next: "okay, but who's watching them?"

Today's showcase:

• Agent audit log, reaching customers: visibility into what your agents actually did, action by action.
• Permission-sets (beta): finer-grained control over who can do what, human or agent.
• AI for HIPAA early access: AI workflows in regulated environments.
• SSO & sharing controls: tighter command over how work gets shared.

Why it matters: agentic work has been opaque, spread across everyone's personal ChatGPT and Claude sessions. Invisible, ungoverned, one bad move from wrecking a workflow nobody could trace. In Airtable, humans and agents act under one permission model; every action lands in the same audit log and the same approval flows. Agents operate inside your rules, and the org stays in control. That's what makes teams comfortable deploying AI at scale: speed because of the guardrails, not instead of them.

▶️ Watch the Day 4 showcase: https://www.airtable.com/articles/agent-audit-log

🔗 The full week: https://lp.airtable.com/summer-camp 

Over to you:

• Admins: which matters most, the audit log, permission sets, or HIPAA? Why?
• How would real agent governance help your org? What have you been holding back from deploying?
• Builders: how are you thinking about handing work to agents vs. keeping it human?

Last day of camp tomorrow, and it's everything you asked for. Literally.

I can’t complete exercise in the the MCP quickstart page, related to Day One … 

I’m using the Airtable MCP connector in ChatGPT Enterprise/Codex with the dummy CSV.

Here’s what ChatGPT says is the going on … 
 

What works:

• My Enterprise admin enabled all Airtable connector actions.

• I have owner access to the Airtable workspace.

• create_base, create_table, and create_field work.

• search_bases and list_tables_for_base correctly return newly created resources.

• I successfully created a fresh base with four tables and linked relationships.

The problem is create_records_for_table. ChatGPT followed the documented sequence:

1. Search for the base.

2. List the tables.

3. Call create_records_for_table using the returned base, table, and field IDs.

The request is rejected by ChatGPT’s connector validation before it reaches Airtable. The validator appears to accept only a fixed allowlist associated with an unrelated example/test base, rather than the IDs returned by the discovery tools.

The tool reference says create_records_for_table should accept IDs obtained from search_bases and list_tables_for_base, and support batches of up to 50 records. Reading records from the new base works, but all tables remain empty because the write is rejected before execution.

ChatGPT saw the same behavior with an earlier attempt, so it does not appear to be specific to one base.

Is this a known Airtable MCP or connector-schema bug? How can the create_records_for_table action’s stale ID allowlist be refreshed or removed so it accepts newly created bases and tables?

Summer Camp Day 5: Everything you asked for

Last day of camp, and this one's yours. Today's drop is built from the request list — features this community has asked for, upvoted, and (politely) not let us forget:

• Staged edit forms ensure any edits made to a record are validated and complete before the information is saved.
• Autoschedule replans a whole timeline automatically when one date slips, instead of leaving downstream tasks stale.
• Omni, Airtable's AI assistant, is now available in the mobile app, so you can check status and make quick edits on the go.

Why it matters: not every launch needs to be a platform shift. Day 5 is the proof that the feedback loop works — community requests in, shipped features out. If you've ever posted a feature request here, today is your receipt.

📖 Read the deep-dive: https://www.airtable.com/articles/omni-mobile-autoschedule-staged-edit-forms 

🔗 Rewatch the whole week: https://lp.airtable.com/summer-camp 

Over to you

• Which Day 5 ship is your favorite — and what were you using as a workaround before?
• What should we build next? Drop your top request below (this thread will be read).
• And the week in review: which of the five days was your day?

Thanks for coming to camp. Keep building and keep the requests coming. ⛺☀️