How AI agents should handle infra: What we learned deploying 480k databases for agents

First it was Replit , this week Create XYZ , more big ones in new verticals coming soon. As a category, AI Agents are now creating databases on Neon at more than 4x the rate of developers. We're onboarding 2000 developers per day but Agents are on another level.

A shift in how software can be created has begun. Many creators with no coding background are now building apps using AI and agent platforms. Sure, you can point out that these apps are toys, but in the words of Chris Dixon “The next big thing will start out looking like a toy”

We've been working closely with teams building in this space to determine how best to support them. Here is some of what we've learned so far.

Agent platforms should focus on a specific user archetype

If you're building an agent platform, you're either building for developers or non-developers. Platforms that sit in the middle are too complex for the former and too simple for the latter. Attempting to serve both results in serving neither well.?

Non-developers will gravitate toward the most straightforward solution that works. They'll switch if they find a tool simpler, cheaper, or more effective than what they're currently using. They don't care about implementation details (UI framework, programming language, deployment provider, database, etc.).?

Exposing said details is an anti-pattern. It adds friction and an additional learning curve, especially if users need to sign up for multiple tools and somehow link them together. All infrastructure components, including the database, should be integrated and invisible to best serve those users.?

New text to app platforms have asked if they can integrate by setting users up with their own Neon accounts and we’ve told them no. We want to set these platforms up for success with a fast, reliable, scalable database layer that is INVISIBLE to their customers.

Neon for Agent platforms?

When we built Neon, we made a series of decisions that, in retrospect, helped us create a database platform ideal for AI agents and shipping better end-user experiences.?

• Instant provisioning: Neon spins up a Postgres database in under a second, making AI-generated apps faster to deploy.
• Autoscaling: Compute scales with demand and suspends when idle, cutting costs and enabling free-tier plans.
• Read replicas: Instantly provisioned and cost-effective, allowing AI agents isolated access to shared data.
• Branching & recovery: Instantly create isolated environments or restore databases to a previous state.
• AI-ready APIs: Automate usage limits, billing, and resource management with Neon’s API and MCP server.

Final thoughts

AI agents are now creating more databases on Neon than developers - and this trend shows no signs of slowing down. As Agent platforms democratize app creation, we're doubling down on building the best Postgres platform to power this new way of building software.

If you're building AI agents that need databases, and you want to make this complexity genuinely invisible to your end-users, reach out or message me on LinkedIn we'd love to partner.