May's headlines were about money: reported funding rounds at valuations that would have sounded like typos two years ago. The more useful signal for a working business sat underneath them: where that money is being pointed.
Increasingly, it's pointed at deployment. The labs have noticed what their enterprise customers already knew: the hard part isn't the model, it's making it do useful work inside a real operation.
The deployment gap becomes the product
OpenAI reportedly stood up a dedicated deployment company with billions in initial backing, and acquired an AI consultancy for its engineers and deployment specialists. That is a frontier lab concluding that adoption, not capability, is the bottleneck worth billions.
We'd agree, with one caveat: deployment help from the company selling the model is never neutral. The integration advice that serves you best comes from whoever has to make your systems, your data and your processes work, whichever model that ends up calling.
Search, reshaped in public
Google pushed AI Mode to the centre of Search, calling it the biggest change in twenty-five years, alongside new Gemini releases including a model that generates video from mixed inputs. Apple was reported to be preparing iOS support for third-party AI providers inside its own assistant features.
Two business implications. First: how customers find and evaluate you is changing. AI summaries increasingly sit between your website and your buyer, which raises the value of being the clear, factual source they draw on. Second: even the biggest platforms are hedging on which model wins. Your architecture should hedge the same way.
Security goes collective
Anthropic launched a cross-industry security consortium (reportedly spanning Amazon, Apple, Google, Microsoft and major banks) after its own research models surfaced large numbers of previously unknown software vulnerabilities.
When the companies that build the tools form a joint defence effort, the message for everyone else is simple: patching discipline and supplier security reviews have quietly become AI-era controls. Ask your software suppliers, including us, how they handle vulnerability disclosure. The good ones will have an answer ready.
What to do with all this
May's actions:
- Treat model-vendor deployment services as one bid among several, not the default
- Check what an AI search summary currently says about your business, and fix the sources it draws on
- Keep model choice swappable: logic and data in your systems, the model as a component
- Put supplier security practices on the agenda for your next review cycle
