CTO.new Coordinates Tens of Thousands of AI Agents with a Database per Project on Turso

CTO.new, built by Engine Labs, gives users teams of AI agents that plan, assign, and complete engineering work autonomously. Every project on the platform, which CTO.new calls an AI business, is a team of agents working together inside its own sandboxed infrastructure. Keeping those agents coordinated at scale required a data layer that could multiply as fast as the projects did.

Today, CTO.new runs tens of thousands of Turso databases, one for every project on the platform, supporting roughly over 50,000 users on a $500 per month plan.

#A Return Customer

CTO.new first adopted Turso in the earliest days of the vibe coding market. Users would generate an application, and the platform would provision backend infrastructure for each one. At the time, that meant a Supabase instance per user at $20 per user per month, a cost structure that stopped making sense once the market settled on $20 per month as the standard base plan price.

"We got to the point where instead of spinning up a Supabase project at $20 per user per month, we spun up a Turso database and handled auth ourselves for maybe a dollar a user," recalls Simon Spurrier, Founder of Engine Labs and CTO.new. "Marginal database costs went from $20 to basically nil. Migrating onto Turso cut our costs by about 95%."

When CTO.new pivoted toward completing engineering tickets directly on customer codebases, the need for managed per-user databases fell out of the stack entirely. They never replaced Turso with an alternative. The requirement simply disappeared.

Then the product evolved again, back toward infrastructure management, and a new database requirement emerged: tracking the workloads of autonomous agents. The use case had changed, but the shape of the solution had not.

"We already had a good experience with Turso on scalability and cost," says Spurrier. "When we were looking for something to fill this new use case, the shape of a Turso database made total sense. We still had our account. We logged back in and started using it again. We didn't even need to consider alternatives."

#The Challenge: Where Does Agent State Live?

Every CTO.new project spins up its own sandbox where agents do their work. The question was where to keep track of the agent workloads running inside that infrastructure. CTO.new evaluated three options.

The core production database. Rejected on security grounds. Agents need full access to their database, including the ability to manipulate schemas. Granting that level of access to shared production infrastructure was a risk the team refused to take.

A database inside the sandbox. Cheap and contained, but the sandbox has to be running for the database to be queryable. Sandboxes cost cents per minute, and at scale, spinning one up just to check what an agent plans to do tomorrow becomes a meaningful line item.

A lightweight database per project, outside both. This was the requirement that emerged: fast, cheap, fully hosted, provisioned individually for every project, accessible to agents without touching core infrastructure, and available even when the sandbox is cold.

"We wanted to manage tens of thousands of databases without managing any of the infrastructure ourselves," says Spurrier. "When we asked the question that way, we already knew the answer."

#The Solution: A Sophisticated To-Do List for Agent Teams

Each AI business on CTO.new gets one Turso database shared by its team of agents. The database functions as the coordination layer for the whole team.

"If you imagine a kanban board, it's basically a sophisticated to-do list that the agents can read and write to whenever they need to," explains Spurrier. Tasks carry states like to-do, urgent, and in progress, along with metadata. Agents managing work and agents doing work both operate against the same database, and a team lead agent assigns tasks and monitors progress through it.

The economics of the query pattern matter here. "Instead of having to run a query against some big blob of context, which might be expensive, the agent can just query the database, which is basically free," says Spurrier.

Because the database lives outside the sandbox, task state is always available. Checking the status of an agent team never requires paying to wake up compute.

Database lifecycle is deliberately simple: databases are created alongside new projects and never deleted unless a customer explicitly requests it. "Some projects go dormant, but we've had customers come back and ask us to restore things," says Spurrier. "It's just so cheap that it doesn't matter. We don't need to delete anything for cost."

At peak, CTO.new runs on the order of 2,000 concurrent, actively queried databases every day.

#Results: What Changes When the Database Is Free

The direct cost comparison is stark. Running tens of thousands of traditional database instances at industry benchmark pricing for managed Postgres would cost many multiples of CTO.new's $500 per month Turso bill, even if we assume the cost of $0.50 per database. Or worse, it would require significant backend infrastructure to multiplex many logical databases into the same physical entity, while running the risk of isolation problems.

On the face of that, Spurrier argues the more important shift is architectural. "When the marginal cost of a database goes to zero, you no longer have to think about how much you can cram into one database and how to make that work. You can just create a new database. That's easier, and typically more ideal, when you want to keep logical boundaries between the data you're storing."

Those boundaries matter most when agents are the ones touching the infrastructure. "When you let agents interact with any piece of infrastructure, there's some risk. So you should ring-fence that as much as possible," says Spurrier. A single shared multi-tenant database for all agent workloads would have put every customer's state in one blast radius.

The savings flow directly to CTO.new's users. Avoiding unnecessary sandbox starts keeps one of the platform's largest cost line items in check, which funds something rare among AI products: a free forever tier. "Any marginal improvement we get on costs helps us offer more for free," says Spurrier. "The product probably wouldn't work in its current model without it. What users see is that they get more free AI usage than they otherwise would."

His advice to engineers who reach for Postgres on AWS by default: "Fine, if you only need one database. We still use Postgres for our core infrastructure. But if you need lots of databases, imagine the database is free. What would you do then? You would never in a million years spin up a Postgres database per user. The fact that the costs are not absurd opens up so many more use cases."

#Looking Forward

CTO.new plans to be running millions of databases on Turso, with thousands to tens of thousands concurrent. "At anything like our current growth rate, it won't be too long before that happens," says Spurrier. "We're confident Turso can support it, partly because the shape of the database is so lightweight, and partly because the team just seems to know what they're doing."

Building agent infrastructure that needs a database for every team, tenant, or task? Get started with Turso Cloud today.