6 June 2026: Vibe coding meets database scale (AM)

AI’s weekend signal is not another abstract model race. The useful stories are about infrastructure, security settings and business software that decide whether AI actually fits into everyday work. For UK readers and small firms, the question is becoming less about trying every new tool and more about knowing which parts of the stack are changing underneath them.

Supabase says AI coding tools are now driving a large share of new database demand, and its answer is a new open source Postgres scaling layer called Multigres. In a company blog post, Supabase said it had raised a $500 million Series F at a $10 billion pre money valuation and that database launches on its platform had grown 600% over the past year. The company also said more than 60% of new databases are now launched by some form of AI tool, with Claude Code and Codex credited as major growth drivers.

The practical point is that vibe coding has moved from demo culture into backend load. If more non specialist builders are using AI tools to create real apps, the pressure moves quickly to authentication, data storage, backups and scaling. Supabase describes Multigres v0.1 alpha as self hostable and not yet production ready, so the claim to watch is not whether it sounds ambitious. It is whether it gives smaller teams a safer path from prototype to durable software without forcing an early database migration. For readers building with agents, Cristoniq’s guide to what can go wrong when AI agents act on your behalf is the useful companion question.

OpenAI has made two everyday ChatGPT changes that matter more than their small release note footprint suggests: broader Lockdown Mode and a sunset path for older paid models. According to OpenAI’s ChatGPT release notes, Lockdown Mode is now available to all logged in users as an optional security setting that restricts network enabled features such as browsing, deep research, agent mode and file downloads. OpenAI also says o3 will leave ChatGPT on 26 August 2026, while GPT-4.5 will leave ChatGPT on 27 June 2026. The company says those retirement dates apply to ChatGPT only, not the API.

For consumers, Lockdown Mode is a plain warning that connected AI is not just a smarter chatbot. It can touch web pages, files and external services, which makes prompt injection a real operational risk rather than a niche security term. If you use ChatGPT for work documents, supplier research or account data, the setting is worth testing before you need it. Cristoniq’s explainer on prompt injection covers why hostile instructions can be hidden in ordinary looking content.

Google has reportedly signed a huge SpaceX compute deal, underlining how much AI capacity is becoming a board level constraint. TechCrunch reported that Google will pay SpaceX $920 million per month from October 2026 to June 2029 for access to roughly 110,000 Nvidia GPUs plus associated CPUs, memory and components, citing a regulatory filing. The article notes that the deal resembles a recent Anthropic compute arrangement with SpaceX and includes a cancellation option after 31 December 2026.

This is not a direct consumer feature, but it explains why AI product pricing keeps moving. Every assistant, coding agent and automated workflow sits on a physical compute bill somewhere. When the biggest firms are securing multi year GPU capacity at this scale, smaller businesses should expect more tiering, usage limits and model routing in the tools they buy. The user facing version of this story is simple: the cheapest AI plan may not stay cheap if your team starts using agents heavily. Cristoniq’s guide to AI data centres explains why the infrastructure layer matters.

Meta’s Business Agent is now available globally inside WhatsApp Business and Instagram DMs, pushing AI customer support into the channels many small firms already use. TechCrunch reported that Meta is making the bot available after tests in markets including India and Mexico. According to Meta, as reported by TechCrunch, the agent can answer customer questions, recommend products, book appointments, qualify leads and route queries to a person when needed.

The UK angle is not that every small business should turn it on immediately. It is that customer service automation is moving into familiar messaging apps rather than separate helpdesk software. That lowers the adoption barrier, but it also raises sharper questions about disclosure, accuracy and escalation. A customer asking about delivery, returns or a booking may not care whether the answer came from a human. They will care if the answer is wrong, or if no one takes responsibility for it. The tool to watch is the handoff from bot to person, not the demo answer.

Coralogix has raised $200 million on the bet that companies will need monitoring tools for AI agents as they move into production. TechCrunch reported that the Series F values the software monitoring company at $1.6 billion post money. The company sells observability tools, meaning systems that collect logs, metrics and traces so teams can understand what software is doing and why it fails.

Agent monitoring sounds dull until an automated system makes a bad call. If businesses are going to let AI investigate incidents, write code or query operational data, they need records of what the system saw, which tool it used and when a human stepped in. That is why AI governance is becoming a small company issue too. It is not only about regulators or big model labs. It is about having enough visibility to explain what happened when automated work touches customers, money or production systems.

Here is everything else worth knowing from today’s AI news.

• Google wrapped its May AI announcements into one official recap; the company highlighted Gemini 3.5, Gemini Omni, agentic Search features and new health hardware, which is useful as a reference point after a noisy I/O cycle.
• AirTrunk plans a major India data centre expansion; TechCrunch reported a $30 billion commitment for 5GW of capacity by 2030, another sign that AI infrastructure is spreading beyond the usual US cloud regions.
• Google’s Dreambeans is testing a more personal AI feed; TechCrunch reported that the experimental app turns connected Google data into limited daily story suggestions for eligible US Google AI Ultra users.

The thing to watch next is whether these separate stories converge. A builder may use an AI coding assistant, deploy on Supabase, serve customers through a messaging agent, and need monitoring when the workflow breaks. That chain is where AI becomes useful, expensive or risky in practice.

This is a daily news update for informational purposes only. AI products and policies change rapidly. Verify details directly with providers before making decisions. Nothing here is financial or legal advice.

AI Daily is Cristoniq’s daily guide to developments in artificial intelligence, published every morning and evening.