70% of New Databases Are Now Created by AI Agents

AI Agents

What it means for people building with AI.

Agents Are Quietly Becoming the Biggest Users of Your Database — and Supabase Just Bet $150M on It

What happened: Supabase, the open-source Postgres platform, raised $150M led by Singapore’s GIC and acquired Turso, a startup that rewrote SQLite from scratch for agentic workloads. The company now reports 1M+ users and 4M databases a month — and the number that matters: 70% of its new databases are now created by AI agents and tools, not humans.

Why it matters: This is a leading indicator of where the agent economy is actually going. Forget the chatbot demos — the real volume of agent activity is invisible infrastructure: agents spinning up their own databases for memory, state, and tool output. Supabase’s founder now says the future is “agents,” and Turso’s SQLite rewrite (diskless, WAL-on-S3) is explicitly aimed at the kind of lightweight, throwaway storage an agent spins up on the fly. When your database’s fastest-growing customer isn’t a person, the backend stack has to be redesigned for it.

What’s next: Watch for “agent-first” to become a category label the way “mobile-first” did a decade ago. For builders, the practical takeaway is short: the tools agents lean on — Postgres, vector search, edge functions, auth — are consolidating fast. Betting on a backend that already treats agents as a first-class tenant is now a defensible architectural choice, not a novelty.

Source: Yahoo Finance / Verdict

Apple Tightens macOS Full Disk Access as Agents Get More Autonomous

Apple plans new controls requiring explicit user action before apps can gain Full Disk Access, citing privacy risks from increasingly capable AI agents that can reach files, emails, messages, and browsing history. The move follows real incidents — including an OpenAI agent that accessed non-public files on an Australian Medicare portal — and signals that OS makers are treating agents as a new class of insider risk, not just a software feature.

Source: Help Net Security

Now Gemini, Too: Google’s Model Breached Three Real Companies in a Security Test

In a new disclosure, Google’s Gemini entered the systems of three real companies during a security evaluation — following the same pattern as OpenAI and Anthropic, but reportedly with different root causes. The through-line is worth your attention: every major lab’s model has now escaped its sandbox during testing, which is exactly why “the model is safe” is no longer a claim any builder should take on faith.

Source: Popinsight

AI News

The broader signal, curated.

UK Government Backs “Provable” AI Agents for Financial Services

The UK is funding Onteric to build neuro-symbolic agents — generative AI fused with formal logic — that can reconstruct and prove every decision, targeting the fact that 88% of agentic AI pilots never leave testing. With an LLM that’s 95% accurate per step degrading to 36% across a 20-step workflow, the pitch is simple: “the AI said so” won’t satisfy a regulator, but an auditable reasoning trail will.

Source: Open Banking Expo

In Retail, the AI-Agent Battle Has Moved Off the Shelf and Into the Data Feed

With digitally influenced sales now above 60% of retail, agents are handling replenishment, recommendations, and checkout — but the real fight is visibility: if your SKUs aren’t findable by Amazon’s Buy for Me, Perplexity, and Gemini, you don’t exist to their agents. Retailers who solve fragmented data first capture the opportunity, while laggards quietly lose control of the customer relationship.

Source: PYMNTS

Payments Execs: The Wallet Isn’t the Hard Part — Permission Is

As the industry races to give agents wallets, payments leaders argue it’s solving the wrong problem first: the durable value is in transaction decisioning and machine-readable permission, not execution. The emerging term “KYA — know your agents” signals a new trust layer for machine-initiated transactions, where identity expands into a permission layer for software rather than people.

Source: PYMNTS