As 2025 wraps up, one thing is clear: AI stopped being “a tool you open” and started becoming “a capability embedded into how work gets done”. For leaders, that shift matters more than any single model launch.
Below is a clear, practical recap of the biggest AI shifts of 2025, and what they mean as you plan for 2026.
Key Shifts in AI in 2025
1) AI moved from chat to action (agents that do the work)
This year, the big change has been AI moving from “answering” to taking actions across tools and the web.
- OpenAI’s Operator evolved into ChatGPT agent inside ChatGPT, with an “agent mode” that can use a virtual computer to complete tasks. If you want a practical walkthrough, check out the last edition of FutureShifts.
What’s new here is not just capability, it’s expectation: people increasingly assume AI should plan, execute, and return a finished output, not just a suggestion.
This year’s launches have put AI directly inside the browser experience. We saw agentic browsers bring LLM capabilities into day-to-day browsing, for example Perplexity’s Comet and OpenAI’s ChatGPT Atlas. (Perplexity; ChatGPT Atlas) Below is a slide from my keynote on AI agents at the AI Horizons Summit hosted by Arab Open University (AOU) in Kuwait this year.
2) Tool access started to standardise (MCP became a real inflection point)
As agents grew, so did the need to connect them safely to internal systems.
- In 2024, Anthropic introduced the Model Context Protocol (MCP) as an open standard for connecting AI tools to data sources and services. (Anthropic)
- By late 2025, MCP had become a cross-industry topic, with major vendors converging around it as a common “connector layer”. (The Verge)
Why this matters: It reduces the “custom integration for everything tax”, which is one of the biggest blockers to scaling agentic workflows.
3) Shopping and transactions became agent-ready
We moved from “AI recommends what to buy” to “AI can complete the purchase”.
- OpenAI announced Instant Checkout in ChatGPT, built with Stripe via the Agentic Commerce Protocol (ACP). (OpenAI)
This is an early signal of a broader shift: conversations becoming conversion, with purchase flows happening inside AI interfaces.
4) Orchestration and automation platforms drew serious money
The market rewarded “AI that reliably runs workflows”, not just clever demos.
- n8n announced a $180m Series C and a $2.5bn valuation, positioning workflow orchestration as core infrastructure for AI-enabled operations. (n8n Blog)
5) On-device AI accelerated (AI moved closer to the user)
In 2025, more AI capability shifted onto personal devices (phones and PCs), reducing latency and keeping more processing local, while enabling more proactive “ambient” experiences.
- Google positioned the Pixel 10 around “advanced on-device AI” with Gemini built in, powered by the Tensor G5 and on-device models such as Gemini Nano, including proactive features like Magic Cue. (Google)
6) Regulation stayed noisy, but the direction of travel was clear
2025 did not bring global alignment. It brought more pressure, more scrutiny, and less patience for “we’ll fix it later”.
- US federal posture: Executive Order 14179 aimed to remove perceived barriers to US AI leadership. (The White House)
- In December 2025, the White House issued a further executive action focused on limiting state-level obstruction of national AI policy, with wide public debate about legality and impact. (The White House)
- EU AI Act: no blanket “pause”, but targeted delays are on the table. The Act remains in force, with key obligations already applying, including rules on prohibited practices and AI literacy from 2 February 2025, and general-purpose AI obligations from 2 August 2025. The main application date remains 2 August 2026, with longer transitions for some areas. (AI Act Service Desk)
7) Frontier model competition accelerated (especially around “work outputs” and agents)
Model progress in 2025 was not just bigger benchmarks. It focused on usefulness for professional work and long-running agent tasks.
- OpenAI released GPT-5, then GPT-5.1, and then GPT-5.2, positioning the latest series explicitly for professional work and agents. (OpenAI ChatGPT 5; OpenAI ChatGPT 5.2)
- Reuters reported GPT-5.2’s launch in the context of intensified competition, including Google’s Gemini 3, positioned by Google as state-of-the-art for reasoning, multimodal understanding and coding. (Reuters)
8) “Vibe coding” went mainstream, and then got a reality check
Building with AI got dramatically easier for non-engineers and business teams. Airtable explicitly leaned into the “AI-native” app-building narrative, including “vibe coding” language. At the same time, experienced builders increasingly cautioned that vibe coding is not a substitute for maintainable software engineering. (Airtable)
Why it matters: The opportunity is real, but so is the risk of fragile internal tools if teams skip testing, documentation, and ownership.
9) Adoption constraints were mostly human
Multiple reputable reports continued to emphasise that tools are not the main blocker, leadership, operating model design, incentives, and resistance to change are. (McKinsey; Microsoft)
Why it matters: Your 2026 advantage is less about “which model” and more about “how you roll it out”.
10) Multimodal creation went mainstream, especially video (and it brought provenance and rights to the forefront).
2025 was a step-change year for video generation and multimodal content tools, with major releases from both OpenAI and Google:
- OpenAI launched Sora 2 (Sept 2025) and continued expanding Sora features through late 2025. (OpenAI)
- Google introduced Veo 3.1 (Oct 2025) and positioned it around higher-quality video plus richer native audio and creative controls, including API support. (Google Blog)
Merry Christmas, and a practical wish for 2026 🎄
My wish for you this Christmas is simple: that 2026 is the year your organisation moves from “AI pilots” to repeatable, trusted outcomes, with the right guardrails and the right people enabled to use them.
Merry Christmas, and wishing you a restful end to the year.
Connect with GenFutures Lab
At GenFutures Lab, we help organisations harness AI responsibly and effectively from strategy and adoption to workforce upskilling.
If you’re exploring how AI agents, automation, or training could transform your organisation, we’d love to hear from you. 👉 Book a consultation or connect with us