Bubble Community on Price Transparency and Agent Data Opt-In: The Forum Isn't a Pricing Page, and the Agent Doesn't Go Into Prod Without Consent

On October 6, 2026, Bubble moved Agent 2 to General Availability (GA) for all web and mobile apps. The same release put Bubble MCP into beta and switched in-editor Agent iterations to a per-app AI credits model. Bubble also announced a new Starter Web price, effective October 20, 2026: $39/month on monthly billing or $32/month on annual billing. Until October 20 you can still lock the previous $29/month annual rate. Same topic: Agent 2 GA, MCP Beta,.

The Bubble forum reacted right away. Builders raised price transparency concerns, mainly about paying a higher base rate while sitting on unused Agent credits. There were also isolated reports that log and app-data toggles were on by default on certain existing apps. If you're a non-technical founder building an MVP or testing a startup idea with nocode and vibecoding, clarity and speed are what count. So I want to separate the forum debate from the guardrails the platform actually documents.

The Forum Is Not a Pricing Page: Tariff Realities

Forum threads show what users think. They don't set commercial policy. When I budget a product, I go by official documentation and formal announcements. Here's what they say:

1. Official Starter Web change. Starting October 20, 2026, Starter Web is $39/month on monthly billing and $32/month on annual billing. Existing apps have until October 20 to lock the previous $29/month annual rate.
2. Per-app AI credits. Agent actions inside the editor use monthly credits assigned to that specific app. Manual visual editing stays unmetered. Production app traffic uses standard Workload Units (WU), not AI credits.
3. Bubble MCP limits. Bubble MCP doesn't burn Bubble AI credits. It runs on the limits of the external LLM provider you connect. Same topic: Bubble MCP on Free With.

The complaint I keep seeing is that bundling Agent credits pushes up the entry price for builders who only need core hosting. I get it. But the official policy hasn't moved, so I plan the launch runway around the verified numbers, not around forum speculation.

Agent 2 Data Opt-In and Production Safety

If you're putting an early version in front of customers for feedback, you need to know what the agent can touch. The official documentation lists these controls for Agent 2 and your database:

• Database reads are off by default. Reading records is gated under Settings → AI by the "Allow AI agent access to App Data" toggle, and you can choose individual data types.
• Development data only. The documentation says Agent 2 only accesses the development database, never live production records.
• Admin permissions. Only app owners and administrators can change the data toggles. The Agent inherits the permission level of the user working with it.
• Plan approvals. Before the Agent makes structural database changes, it proposes a plan in the conversation and you have to approve it directly.

The forum tells a less tidy story. Some builders reported that certain existing apps showed the toggles already on after the GA rollout. Others weren't sure where to look: the documentation points to Settings → AI, while peer discussions mention Collaboration panels. That's why I wouldn't trust either source without checking for myself.

Actionable Pre-Launch Checklist for Non-Technical Builders

Before you hand tasks to Agent 2, here's what I'd do to protect the MVP:

1. Check your settings by hand. Don't take the defaults for granted. Open Settings → AI in your editor, look at "Allow AI agent access to App Data", and confirm which data types are on.
2. Keep sensitive information out of the dev database. Live data is isolated, but that's not enough. Fill your development environment with synthetic records only, never customer credentials or private data.
3. Watch your credit burn. During heavy agent-assisted build sprints, keep an eye on the per-app AI credit balance so a workflow doesn't get blocked halfway through.
4. Decide on the lock-in before October 20. If your project needs a longer test phase on Starter Web, consider locking the $29/month annual plan before the $39 price kicks in. Same topic: Workflow engine alpha: “end of.

Checklist item	What to do	What to avoid / watch for
Check settings by hand	Open Settings → AI; review "Allow AI agent access to App Data" and which data types are on	Don't assume defaults are correct
Dev database hygiene	Fill the development environment with synthetic records only	Customer credentials or private data (live data is isolated, but that's not enough)
Credit burn	During heavy agent-assisted build sprints, watch the per-app AI credit balance	Workflows getting blocked halfway through
Starter Web lock-in	If you need a longer test phase on Starter Web, decide before October 20	Missing the $29/month annual plan before the $39 price kicks in