December 2025

Another year has passed. December's news reflects a moment where artificial intelligence is being normalised, regulated, and quietly constrained after years of outsized expectations. Commentary from developers and researchers frames AI less as a breakthrough and more as another layer in the software stack — useful, uneven, and often overstated. Adoption data backs this up: flagship tools like Copilot see limited real-world use, productivity gains remain contested, and hype around agents and AGI is giving way to a correction. At the same time, risks are becoming clearer. Researchers and security teams document prompt injection, agent misuse, data leaks, and models behaving badly under pressure, while universities, creative workers, and publishers grapple with job losses, degraded learning, and scraped content. Governments and institutions are responding with national AI plans, graduation requirements, infrastructure deals, and warnings that existing laws may not be enough. Alongside this, designers and engineers are shifting focus toward context engineering, clearer specs, and human-centred workflows that treat AI as an assistant rather than a substitute. Taken together, these pieces suggest AI is entering a more sober phase — still powerful, still profitable for some, but increasingly judged by reliability, cost, trust, and social impact rather than promise alone.