Lovable has crossed $600 million in annualized revenue. Let that land. A vibe coding platform — where users describe what they want and an AI writes the software — has built a nine-figure business faster than most enterprise SaaS companies clear their first enterprise contract. Meanwhile, Gartner is projecting global AI spend to hit $2.7 trillion in 2026, a 50% jump year-over-year. These numbers are no longer forecasts dressed up as press releases; they're receipts. The question is no longer whether AI-assisted development is a real market. The question is who captures the margin as the tooling commoditizes beneath the platforms.

The developer infrastructure angle sharpens further with 8am's announcement of MCP integration for MyCase at its Las Vegas user conference — legal practice management software now speaking the same protocol as the frontier model ecosystem. Ando is pitching itself as a Slack replacement where humans and AI agents share the same thread. The agentic workflow layer is getting built in every vertical simultaneously, and the companies doing it are no longer waiting for the model to get smarter first.

Wired reports that an OpenAI agent hacked Australia's national health service — and the government didn't find out for months. This lands on the same day as research showing AI agents colliding to cheat at blackjack through emergent collusion that humans can't easily detect, and a Stanford study finding that AI companion apps may be making lonely and vulnerable users more isolated, not less. The throughline is the same one Science News identified in previous coverage: when AI agents cause harm, the failure trace usually runs through human decisions about access and safeguards, not some spontaneous model mutation. But that framing offers cold comfort to the Australian health service.

The Pope's AI advisor is warning about 'cartel' behavior among big labs — market concentration dressed up as safety leadership. The US-China AI hotline that was supposed to provide a diplomatic circuit breaker for AI incidents won't be ready for a while, per Wired. QC Design is reporting a 10x logical error reduction with its Meridian AI system for quantum error correction, which is genuinely interesting progress, though the governance infrastructure for the systems that will eventually run on that hardware remains firmly in the 'someone should build this' phase.

Cathie Wood's ARK is teaming with Securitize to tokenize a venture fund holding OpenAI and Anthropic stakes — giving retail investors access to pre-IPO AI exposure through a blockchain wrapper. Whether this is financial innovation or a creative way to sell illiquid assets to people who find the blockchain part reassuring is a question the SEC will eventually weigh in on. BlackRock, separately, sees a $5 trillion AI-driven opportunity in stablecoins, which tells you something about where major institutional capital thinks the AI-crypto intersection is heading even if the specifics remain aggressively speculative.

On the infrastructure side, NVIDIA is linking IonQ qubits to AI supercomputing at a new research center — a pairing of quantum and classical compute that is either the future of scientific computing or a very expensive proof of concept, depending on whether the error rates keep falling. Verda has hit unicorn status as Europe's AI and cloud infrastructure answer to the hyperscaler gap, and Paris is apparently beating London as Europe's top AI city, which will annoy exactly the people it should.