The Week AI Stopped Talking and Started Doing
If last week was about bigger and cheaper AI, this week was about something more important for the rest of us: AI that actually does things.
The headlines moved away from raw model benchmarks and toward agents, actions, and adoption. Here's what stood out — and what it means for businesses outside the Silicon Valley bubble.
1. Google Launched a Real Agent Stack
On April 22, Google Cloud unveiled a full set of tools for building, deploying, and tracking AI agents inside companies — a direct shot at OpenAI and Anthropic in the agent-platform race.
The distinction matters. Until recently, "AI agents" was mostly a marketing term wrapped around a chatbot. The new wave — from Google, Anthropic, and OpenAI — is genuinely different: software that can plan a task, use tools, take actions across systems, and report back. With proper observability so you can see what it actually did.
For businesses, this is the tooling layer that makes serious automation possible without a full engineering team behind it.
2. Yelp Became a Buying Engine
Yelp expanded its AI assistant so users can move from discovery to action in a single conversation — booking a reservation, ordering food, scheduling a service — all without leaving the chat.
This is a small story with a big implication. The default behaviour of AI assistants is shifting from "answer my question" to "complete my task." If your business depends on customers finding you and then completing a booking, payment, or order somewhere else, that gap is starting to disappear.
The practical question for any service business: when an AI agent tries to book or buy from you, can it succeed? If your booking flow only works for humans clicking through a website, you're about to be invisible to a fast-growing slice of demand.
3. DeepSeek's V4 Pushed Context to a Million Tokens
DeepSeek released preview versions of its V4 Flash and V4 Pro models — strong on coding and agentic benchmarks, and notably pushing the context window to 1 million tokens.
Why this matters in plain terms: a million-token context means an entire codebase, a full year of email correspondence, or every contract in a small business can be handed to the model in a single prompt. No clever chunking. No retrieval pipelines. Just "here's everything — now help."
For SMEs, that collapses a lot of complexity. Many of the AI projects that died on the vine eighteen months ago died because the model couldn't see enough at once. That bottleneck is gone.
4. The Compute Layer Keeps Doubling Down
- Google Cloud debuted its newest TPU generation, positioning itself as a serious challenger to Nvidia for AI workloads.
- Anthropic and Amazon expanded their partnership for another 5 gigawatts of new compute.
- Microsoft announced a four-year, $10 billion investment in Japan covering AI data centre expansion with SoftBank and Sakura Internet.
The takeaway hasn't changed from last week, but it keeps getting reinforced: the infrastructure being built now is generational. Whatever you assume AI can do today, assume it'll do meaningfully more in twelve months — at lower cost.
5. The SME Adoption Problem Gets Named
A quietly important thread this week: a wave of new research and programmes focused specifically on small and mid-sized businesses.
- DBS expanded its Spark GenAI programme on April 24, explicitly to reduce friction around AI adoption for SMEs.
- New industry research found close to 2 in 5 SMEs (39%) are actively seeking expert advice on integrating AI.
- The consistent finding across reports: the biggest barrier isn't tools or cost anymore. It's knowing where to start.
The phrase that keeps showing up in interviews with business owners is some version of: "We know AI might help. We just don't know where to begin."
That gap — between availability and applicability — is now the real story.
What This Means for Your Business
A few practical takeaways from the week:
- Agents aren't a future thing. The tools to deploy them seriously dropped this week. The early adopters are six months ahead, not six years.
- Make sure machines can buy from you. If your booking, ordering, or quoting flow assumes a human with a mouse, audit it. AI assistants are increasingly completing transactions on behalf of users.
- Long context changes what's possible. Things that were previously "too messy" for AI — full inboxes, full archives, full contract sets — are now fair game.
- The hard part isn't the tech anymore. It's identifying which task in your business should be the first one. That's a strategy question, not a technology one.
At The Empyrean, we help businesses cut through the noise around AI and connect it to the systems they already have — without the overhaul. If you'd like an honest read on where AI could actually help your business, we're happy to take a look.