Most Airtable tutorials show you how to build a CRM or a project tracker. This one starts somewhere different: with a fully playable RPG inside Airtable Custom Interfaces, a 400-tile isometric map, live GPT inference running inside AI fields, and a cross-player leaderboard — all without leaving the platform.
The game was never really about fun. It was a stress test. The question was simple: how far can you actually push Airtable Custom Interfaces and AI Fields before the platform says no? The answer turns out to be further than most people assume, with a few important caveats that matter just as much for a sales dashboard as they do for a fantasy battle map.
Video Tutorial
Recap of the Game: Consultant Classes, Guardians, and a Leaderboard
The game is called Heroes of No Cold Night and AI Magic — a deliberately tongue-in-cheek title that pokes fun at the no-code consultant industry (the same industry the creator belongs to). Players open a simple configurator, pick an avatar, and choose a consultant class before dropping into an isometric map with roughly 19 turns to spend.
Each of the 400 tiles on that map contains one of three things: a plain terrain tile, an event tile that triggers a resource gain or loss (such as earning money or having AI credits deducted from your score), or a guardian tile protecting a location. When a player encounters a guardian — for example, a monster called "Spreadsheet Hell" guarding a Make.com automation node — they can flee or fight. If they fight, they type or select a response, and the outcome is decided by a live AI inference call running directly inside an Airtable AI field.
That last part is worth pausing on. There is no external API call, no webhook, no Make scenario in the middle. The player's prompt is written to a field in the player record, an Airtable AI field processes it against the game's scoring logic, and the result is written back to the same record. The custom interface reads that value and renders a win or loss. The whole exchange happens inside the base.
Resources — money, AI credits, patients (a scoring currency nodding at consultant billing) — accumulate across turns, and a leaderboard table surfaces scores across all players. The loop is complete: configure, explore, fight, score, rank.
What the Experiment Reveals About Custom Interfaces
The most important lesson from this build is the distinction between what the drag-and-drop interface editor can do and what a custom-coded interface can do. They are not the same tool.
The drag-and-drop editor, including AI-assisted builders like Omni, is excellent for dashboards, filtered record lists, forms, and simple conditional displays. If you need to show a summary of project status or give a client a clean read-only view of their data, the visual editor gets you there in an afternoon.
Building a tile-based game map with hover states, turn counters, grayed-out previously visited tiles, animated resource updates, and state that persists across interactions is a different matter entirely. That required custom JavaScript code loaded into the interface container, organized across multiple files so that AI coding agents could reason about individual pieces of game logic without losing context across the whole codebase.
The takeaway is not that Custom Interfaces are limited. It is that the platform has two distinct capability tiers, and most teams never explore the second one. If you are curious about what the first tier can already accomplish for structured data work, what is Airtable Omni is a good place to orient yourself before deciding whether you need custom code.
Where AI Fields Shine — and Where They Lag
Airtable AI Fields did two distinct jobs in this project, and they performed very differently in each role.
Batch asset generation was where they excelled. Ninety-plus unique isometric game assets — terrain tiles, event icons, guardian sprites — were generated entirely inside the base. The technique was straightforward: write a detailed base prompt covering style (isometric, three-quarter top-down view, transparent PNG background, consistent diamond geometry so tiles align at the seams), then vary only the final object description. One tile is "uniform carbon gray asphalt." Another is "anti-static metal grid." Another is "beach towel with a forgotten Make.com workspace." The prompt core stays fixed; the tail changes. The result is a visually consistent set of assets generated at scale without leaving Airtable.
Live inference during gameplay was where latency became noticeable. Triggering a fight, waiting for the AI field to process the prompt, and reading the result back through the interface introduced a visible pause. For a game, that pause is acceptable because the player expects to wait for a "battle calculation." For a business workflow — say, an AI field that grades a support ticket and recommends a response — the same latency is a design consideration, not a dealbreaker, but it needs to be accounted for in how you structure the user experience.
The deeper constraint is that AI Fields evaluate on a record-by-record basis. They are not streaming, and they do not maintain conversational context across records by default. That shapes the types of tasks they are good at: scoring, summarizing, classifying, and generating discrete outputs rather than orchestrating multi-step reasoning chains. For a broader look at how AI operates inside Airtable at different levels of sophistication, types of Airtable AI agents covers the architecture options in detail.
Serious Business Applications That Map to the Same Patterns
Everything in this game — scored records, AI field evaluation, a ranked display across multiple users, a custom interface rendering state — has a direct analogy in legitimate business tooling. The game is the stress test. These are the production use cases.
Sales and rep leaderboards use the same scored-record plus ranked-display pattern. Each rep's record accumulates deal values, activity counts, or quota attainment percentages. The leaderboard interface reads those fields, sorts them, and renders a live ranking. Add an AI field that generates a one-sentence coaching note based on each rep's recent activity and you have a tool that marketing teams and sales operations teams actually want to use every day.
Gamified onboarding dashboards borrow the turn-counter and resource-accumulation mechanic. A new employee completes a module, a checkbox flips, a formula recalculates their progress percentage, and a custom interface reflects their current "level" in the onboarding journey. AI fields can evaluate free-text reflections submitted after each module and generate personalized feedback — a pattern that translates directly into HR and recruiting tooling without requiring a separate learning management system.
Employee scorecards structured around the same guardian-battle mechanic become performance check-ins. An employee submits a self-assessment, an AI field evaluates it against a rubric stored in the base, and a manager sees a scored summary instead of raw text. The interface renders the score as a visual rather than a number in a cell, making it easier to scan across a team at a glance.
Customer tiering and dynamic portals use the resource-accumulation pattern. Customer spend, engagement scores, or health metrics accumulate in Airtable records. An AI field classifies each customer into a tier based on those values, and a custom interface surfaces different content, options, or calls to action depending on that tier — without any conditional logic inside the interface itself. The classification lives in the data layer where it belongs. For teams managing clients at volume, this pairs naturally with client reporting dashboards that surface the same tiering data to stakeholders in a clean, shareable format.
All of these are the same architecture: data in Airtable, AI evaluation in AI fields, presentation in a custom or configured interface. The game just made the pattern visible by pushing it into an unfamiliar domain.
When to Hire Help
The game was built with significant Claude Code assistance, organized into multiple files so that AI coding agents could handle individual pieces of game logic without losing the thread of the whole project. That is already a sophisticated workflow. It is not the entry point for teams that are new to custom interface development.
There are clear signals that a project has moved beyond what a non-technical Airtable builder can reasonably maintain on their own:
- The interface needs to react to data changes in near real time rather than on page load
- State needs to persist across user sessions and be shared across multiple users simultaneously
- The visual design requires pixel-level control over layout, animation, or conditional rendering
- The AI field logic involves multiple interdependent evaluations across linked tables
- The codebase has grown to the point where a single file is no longer manageable
At any of these points, bringing in an experienced Airtable consultant is faster and cheaper than iterating through trial and error. The patterns exist. The architecture is proven. What takes time is the implementation detail — and that is where professional experience compresses the timeline significantly. If your team is evaluating whether to build internally or bring in outside help, a Make automation agency can also assess whether some of the workflow logic belongs in Airtable at all or would be better handled in an external automation layer.
Next Steps
If the game sparked ideas for what your own Airtable base could do, there are a few logical starting points depending on where you are now.
If you are still orienting yourself to what Airtable Custom Interfaces can do without custom code, start with building interfaces with Airtable Omni and client reporting dashboards to understand the configured-interface ceiling before deciding you need custom JavaScript.
If you are already comfortable with Airtable and want to understand where AI fits into the architecture — not just AI fields, but AI agents, automations, and external integrations — types of Airtable AI agents maps the full landscape.
If you have a specific use case in mind — a leaderboard, a gamified scorecard, a tiered customer portal — and want to know whether it is a good fit for Custom Interfaces or whether it needs something more, reach out. We build these systems for marketing teams, HR and recruiting teams, and operational teams across industries. The game was the proof of concept. Your use case is the real project.
