AI This Week: Agentic AI Goes Mid-Market, and a Ransomware Wake-Up Call

Welcome to the first issue of AI This Week — our short, no-hype roundup of the AI news that actually matters if you run a business and care about getting work done. We skip the model-benchmark drama unless it changes what you can do on Monday morning, and we tell you why each story matters rather than just that it happened.

Here’s what caught our eye this week.

Agentic AI arrives for mid-market companies

The biggest practical story this week is about who gets access, not who has the smartest model. Accenture and Google Cloud announced a suite of pre-built agentic AI solutions aimed squarely at mid-market companies — the kind with revenue between a few hundred million and a few billion, which have historically been too small for bespoke enterprise AI projects and too big for off-the-shelf apps.

Why it matters: for most of the last two years, “AI agents” were something only large enterprises with dedicated teams could realistically deploy. Pre-built agent suites change the math. If you’ve been reading about autonomous agents and assuming they were out of reach, that gap is closing faster than expected. Our guide on AI agents for business walks through what these systems actually do and where they fit.

Deloitte: three in four companies plan to deploy agents within two years

Backing up that first story, Deloitte’s 2026 State of AI in the Enterprise report landed this week with a striking figure: nearly three in four companies plan to deploy agentic AI within the next two years.

Treat any survey number with a pinch of salt — “plan to deploy” is not the same as “successfully deployed.” But the direction is clear, and it lines up with what we’re seeing on the ground. The interesting question is no longer whether businesses will adopt agents, but which processes they’ll point them at first. Our take, based on what actually works: start narrow, keep a human approving anything high-risk, and log everything. The companies that skip those guardrails are the ones that end up with silent errors piling up.

A sobering first: fully autonomous AI ransomware

Not all AI news is good news. This week brought the first documented case of a fully autonomous AI ransomware attack — reportedly hundreds of payloads deployed with no human directing the operation in real time.

We’re flagging this not to fearmonger, but because it changes the security conversation for anyone deploying automation. As you connect AI systems to your tools, email, and data, the same capabilities that make agents useful make them a bigger target — and a bigger risk if misconfigured. If you’re automating business processes this year, security can’t be an afterthought. Least-privilege access, human approval on sensitive actions, and monitoring aren’t red tape; they’re what keeps an automation project from becoming a liability.

The quiet trend underneath it all: cheaper inference

One theme tied several stories together this week: the cost of running AI keeps falling, and that’s what’s making everything else possible. Analysts continue to point to the dramatic drop in inference cost over the past few years as the quiet force behind the current wave of AI features. When running a model gets an order of magnitude cheaper, use cases that were financially impossible suddenly pencil out.

For businesses, the takeaway is practical: features and automations you priced out a year ago may be worth re-costing now. If you want to run those numbers, our AI cost calculator and automation ROI calculator can help you sanity-check whether a project makes sense today.

What we’re watching next week

The industry is clearly shifting from “how big is the model” to “how well does it complete real work without supervision.” That reframing — usefulness over raw size — is the through-line we’ll keep following. Expect more pre-built agent offerings, more scrutiny on security, and more pressure on vendors to prove their tools finish tasks rather than just demo well.

That’s it for this week. If a story here sparked a question about your own workflow, that’s exactly the kind of thing we dig into in our guides — and we’ll be back next week with the next issue.