· 6 min read

AppSheet's Real Cost at 20 Users — and Why an AI Agent Can't Rescue It

AppSheet costs $5–$20 per user per month — but the bigger problem is invisible: its click-only configuration can't be operated by an AI agent. Real cost math, and the config-as-code exit path.

TL;DR

QuestionShort answer
What does AppSheet cost at a real team size?15 users on Core ≈ $150/mo ($1,800/yr); 25 users on Enterprise Plus ≈ $500/mo ($6,000/yr)
Is per-seat the real problem?No — it’s manageable at 10 users, painful at 25+
What’s the problem nobody prices in?Every change is manual GUI work — no config-as-code, so no AI agent can operate your app for you
When is AppSheet still right?≤10 users, standard workflow, someone internally enjoys owning it

AppSheet pricing is public and simple: Starter $5, Core $10, Enterprise Plus $20 per user per month (pricing, August 2026), with free prototyping for up to 10 users. But two costs never make it into the subscription comparison: the seat creep as your team grows, and the maintenance bottleneck that defines 2026 — nearly every other kind of software got AI-maintainable, but a click-configured AppSheet app didn’t.

(One nuance before the math: if you’re on a paid Google Workspace edition, check which editions include AppSheet Core — some bundle it. That changes the math for the Workspace-tier users on your team.)

The seat math, run honestly

TeamTier you’ll realistically needMonthlyPer yearOver 3 years
5 usersStarter$25$300$900
10 usersCore$100$1,200$3,600
15 usersCore$150$1,800$5,400
25 usersEnterprise Plus$500$6,000$18,000
40 usersEnterprise Plus$800$9,600$28,800

Two things hide inside that table:

  • You rarely stay on Starter. It lacks security controls and app management — the moment roles or data ownership matter, you’re on Core.
  • APIs and real databases are Enterprise Plus territory. Connecting AppSheet to PostgreSQL or external services lives at the $20 tier. If your data is outgrowing Sheets (which is why you’re reading this), that’s the tier you’re heading for.

At 25 users you’re paying $18,000 over three years for the right to use the platform — before anyone’s time is counted. And someone’s time is the next cost.

The 2026 problem: your app can’t be operated by an agent

Here’s what changed recently, and why it matters more than the seat math.

Almost every other form of business software became agent-maintainable: if your system is a normal codebase, you can tell an AI agent “change the discount approval rule to require manager sign-off above $500” — and the agent reads the code, makes a small, reviewable change, and updates the tests. Minutes of work, no developer on staff.

AppSheet has no equivalent path. Its configuration — tables, slices, bots, behaviors, security filters, virtual columns — lives in a web editor you operate by clicking. There is no config-as-code, no CLI, no API for modifying the application itself. So when you ask an AI to change your AppSheet app, the honest answer is: it can’t. A human has to open the editor, navigate to the right screen among dozens, edit the expression, save, and test on a device. Multiply by every change request your business generates.

The result is the AppHouse pattern I keep finding: an office manager who “knows the app” and has quietly become an unacknowledged developer — every workflow change routed through their calendar. The platform markets itself as no-code; the operational reality is code-by-clicking, and clicking doesn’t scale and can’t be delegated to software.

What I did about it: AppSheet config as agent-readable code

I recently migrated a complex AppSheet configuration — the messy, years-in-the-making kind with intertwined slices, bots, and security filters — into a conventional codebase. Here’s the honest mechanics of that path, because the fear is always “we’ll lose the app we have”:

  • What AppSheet gives you on export: a folder of definition files (YAML/JSON-ish) describing tables, columns, and rules. It’s data, but it’s not readable — semantics scattered across files, no ordering, no explanation of why a rule exists.
  • The mapping that works: tables → database schema; slices → queries/views; bots → scheduled workflows; virtual columns → computed fields; security filters → row-level access control in one place you can audit.
  • What the team keeps: the same workflow, same screens-by-purpose, same data. What changes is where the logic lives — in code a person or an AI agent can read, diff, change in one place, and verify against a test suite.

The day-one difference is mundane but decisive: a rule change that used to be a half-day of careful GUI surgery (open the right editor screen, don’t touch the neighboring expression, re-test manually) becomes a small code change an agent applies in minutes and a test suite validates. Maintenance stops being a specialist clicking exercise and becomes reviewable work.

That is the actual moat in 2026 — not the database, not the UI framework, but whether the artifact is readable by the agent you’ll inevitably use to maintain it. Prompt-evolved code and click-configured apps both fail this; deliberately structured code passes it.

When AppSheet is still the right answer

The same honesty as always: small team (≤10 users), standard workflow (forms, lists, approvals that look like templates), and someone internally who wants to own the app. For that profile, AppSheet at $50–100/month is a fair deal, and I say so — the alternatives piece covers when to stay.

Leave when any of these become true: seats are creeping past 10–15, you need API/database connections (Enterprise Plus pricing), changes queue behind one person’s calendar, or the Sheets underneath are at 50k+ rows.

What no per-seat subscription includes

AppSheet, Glide, Retool — every platform in this class rents you capability. None of them include the human layer: someone who studies how your operation actually runs, migrates the data with its history intact, deploys on infrastructure you control (a modest VPS runs a real ops platform fine — it doesn’t need a cloud bill), and stays accountable when something breaks. That layer isn’t sold as a subscription by the platform vendors. It’s a person, and it’s what I do.

Sources & further reading