9 June 2026: Legal AI agents become the PM signal

The useful AI story this afternoon is not a louder chatbot. It is the steady move of AI into work that has rules, cost limits and consequences, from legal request handling to app building and Apple’s device level assistant plans.

Sandstone is pitching legal AI agents as infrastructure for in-house teams, not as a courtroom gimmick. TechCrunch reported that Sandstone has raised a $30 million Series A led by Lightspeed, with Sequoia also participating after backing its seed round. The company’s own site describes Sandstone as a platform for managing and resolving in-house legal requests with agents, which is the practical part of the story. The pitch is not that AI replaces legal judgement. It is that routine intake, triage, routing and document workflow can be handled more consistently before a lawyer has to spend time on the judgement call.

For UK readers and small firms, the signal is bigger than one funding round. Legal work is one of the places where AI adoption cannot rely on “close enough” answers, because mistakes can create contract, privacy or compliance problems. That makes Sandstone a useful test case for AI governance in normal business software. If legal AI tools can show clear audit trails, human review points and source handling, they may become easier to trust than broad assistants used informally inside teams.

Lovable says its AI app builder has reached $500 million in annualised run rate revenue. The figure is vendor reported through TechCrunch, so it should be read as the company’s claim rather than an independently audited result. Lovable also says users are creating one million new projects a week, with the product positioned as a way to build apps, websites and internal tools without deep coding skills.

The reason this matters is that AI coding is moving from developer productivity into small business operations. A cafe, agency or local services firm may not need a full engineering team to test a booking tool, client portal or stock tracker. But it still needs to understand what has been generated, what data the tool touches and who maintains it when it breaks. That is where human oversight in AI becomes less theoretical and more operational.

Apple’s new AI architecture is starting to look less like a race for the biggest model and more like a fight over where AI should live. MacRumors reported that Apple’s updated Apple Intelligence architecture is built around Google Gemini models, while TechCrunch framed the broader WWDC message as a slower, more integrated AI strategy. Apple says its Apple Intelligence and Siri AI features are designed for iPhone, iPad, Mac, Apple Watch and Vision Pro, which puts the emphasis on device level actions, personal context and everyday workflows.

That is a different bet from asking users to live inside one general purpose chatbot. The consumer question is whether Siri AI can reliably understand what is on the device, what app action is needed and when to hand the job to a stronger cloud model. The business question is whether Apple can make this feel predictable enough for people who do not want to manage separate AI subscriptions for every small task.

Apple is also trying to lower the cost of AI experimentation for smaller developers. TechCrunch reported that Apple is waiving cloud API costs for developers with fewer than two million first time App Store downloads. That detail matters because AI experiments often fail on economics before they fail on imagination. A feature can be technically impressive and still be too expensive to run for a small app with uncertain demand.

If Apple keeps that cost support narrow and temporary, developers will still need to design features that survive once normal pricing applies. If it becomes a durable part of the platform, smaller teams may get more room to test AI search, summarisation, voice interfaces and personal assistant features. Either way, the important metric is not how many demos ship this week. It is how many useful AI features remain affordable after the launch window.

There is also a sourcing lesson in this afternoon’s brief. Sandstone and Lovable are both useful signals, but their most eye-catching figures come from company claims reported by TechCrunch, not from independent audits. That does not make them useless. It means readers should separate adoption momentum from proof of product quality. For small teams choosing tools, the first question is still whether the product reduces a real workflow cost without creating hidden review, privacy or maintenance work.

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

• Apple’s WWDC demos looked more cautious this year. TechCrunch said the demos appeared more grounded after earlier criticism of overpromised AI features.
• A text to CAD research paper surfaced in the brief. The arXiv paper points to continuing work on turning language prompts into controllable design objects, but it is research rather than a finished product.
• Mercor’s valuation dispute stayed out of the main five. The brief included a funding market argument, but it did not change what readers can do with AI today.

The thing to watch next is whether legal AI tools and app builders can prove reliability after the first impressive demo. Funding and revenue claims get attention, but the durable signal will be repeat use: teams trusting these systems with real requests, real data and clear human review.

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 weekday afternoon.