AI digest: Google goes agent-first
Google I/O 2026 showed the search giant betting everything on autonomous agents over chatbots, while enterprise AI platforms mature and Anthropic poaches talent.
Google I/O 2026 was basically one long argument that the future is agents, not chat. Meanwhile the enterprise world is starting to figure out what actually works.
Google bets the farm on agents with Gemini 3.5 Flash
Google launched Gemini 3.5 Flash as their most powerful coding and agentic model yet, capable of autonomously executing complex tasks and building software from scratch. This feels like the clearest signal yet that Google sees the next wave being about AI that does things, not just answers questions. The big test will be whether these agents actually work reliably enough for real production use.
Enterprise agentic AI platforms finally growing up
Enterprise agentic AI has moved from pilots to production with platforms like Salesforce Agentforce and Microsoft Copilot Studio leading the pack. What’s interesting is that we’re finally getting honest constraint analysis instead of just marketing fluff. The shift from “trying things out” to “actually deploying at scale” suggests the technology is getting mature enough to be boring, which is exactly what enterprise buyers want.
Karpathy picks Anthropic over OpenAI comeback
Andrej Karpathy joined Anthropic to get back into frontier LLM research, choosing them over his former home at OpenAI. This matters because Karpathy is one of the few people whose moves actually signal where the cutting edge is heading. His choice suggests Anthropic is building something compelling in the research space, not just playing catch-up.
Cursor’s Composer 2.5 matches frontier models on the cheap
Cursor shipped Composer 2.5, an AI coding model that matches Opus 4.7 and GPT-5.5 on benchmarks at a fraction of the cost. Built on synthetic training tasks, this shows how specialised models can compete with the big players in specific domains. The real question is whether performance on coding benchmarks translates to actual developer productivity gains.