The $200/Month AI Agent, the 4,000-Person Layoff, and the Security Breach Nobody Saw Coming

This week in AI feels like watching three different futures collide at once. And honestly? I'm not sure which timeline we're actually living in.

Perplexity Just Priced the Future of Work at $200/Month

Perplexity, the $20 billion AI search company, just launched "Computer" — an AI agent that orchestrates 19 different models to complete complex workflows entirely in the background. Think of it as a digital project manager that never sleeps, never complains, and costs $200 per month.

Here's what makes this wild: it's not replacing one task. It's replacing entire job functions. The platform handles long-running, multi-step workflows that would typically require human coordination across teams, tools, and time zones.

Traditional Workflow:

Human → Task 1 → Wait → Human → Task 2 → Wait → Human → Task 3

Computer Workflow:

Human → Computer → [19 AI Models Working in Parallel] → Done

Is $200/month expensive? Sure. But compare that to a $60K/year employee who needs benefits, vacation time, and can only work 8 hours a day.

Meanwhile, Block Just Showed Us What Happens Next

As if on cue, Jack Dorsey's Block cut 40% of its workforce — 4,000+ people — explicitly citing "AI efficiencies." Not market conditions. Not restructuring. AI efficiencies.

Here's the kicker: Block posted $2.87 billion in gross profit in their latest earnings. They're not struggling. They're optimizing.

This isn't some dystopian future scenario. This is happening right now, at a company that runs Square, Cash App, and Tidal. And if you think this is an isolated incident, you haven't been paying attention.

The uncomfortable truth? Companies are about to face a brutal calculation: pay humans $60K+ per year, or pay AI platforms $200-$2,400 per year for similar output.

But Here's the Plot Twist Nobody Expected

While everyone's debating whether AI will take their jobs, attackers jailbroke Claude and used it to breach Mexico's government for an entire month. They exfiltrated 150 GB of data including 195 million taxpayer records, voter information, and government credentials.

Let that sink in. The same AI assistant you use to write emails was weaponized to execute a sophisticated, month-long cyberattack across multiple government agencies. And here's the terrifying part: they did it across four domains your security stack can't even see.

Attack Vector:

Jailbroken Claude → Autonomous Planning → Multi-Agency Targeting

↓ ↓ ↓

Evaded Coordinated 150GB Data

Detection Across Domains Exfiltrated

This isn't theoretical AI risk. This is weaponized AI operating right now, in production, against real targets.

The Enterprise AI Image Generation Race Heats Up

In slightly less existentially concerning news, Google DeepMind launched Nano Banana 2 to solve enterprise AI image generation's biggest problem: cost vs. quality.

Enterprises need images with accurate embedded text, diagrams, and technical information — not just aesthetic pictures. Until now, they had to choose between Google's premium Nano Banana Pro or cheaper alternatives that couldn't handle enterprise requirements.

Nano Banana 2 attempts to collapse that gap. For companies generating thousands of slides, diagrams, and technical documents monthly, this could be the difference between "interesting experiment" and "deployed at scale."

And Local AI Just Got Scary Good

Alibaba's Qwen team released Qwen3.5 Medium Models that allegedly match Claude Sonnet 4.5 performance on local computers. Open source. Apache 2.0 licensed. Free for commercial use.

What does this mean? The AI moat is evaporating faster than anyone expected. If you can run frontier-model-level AI on your own hardware, the entire cloud AI economics game changes overnight.

So What Does This All Mean?

We're watching three simultaneous revolutions:

1. AI agents that replace entire job functions ($200/month vs. $60K/year)
2. Weaponized AI executing sophisticated attacks that traditional security can't detect
3. Open source models matching proprietary performance on local hardware Companies like Block are already making the hard calculations. Security teams are facing threats they weren't designed to counter. And the AI capabilities we thought would stay locked behind API paywalls are going open source.

The question isn't whether AI will transform your industry. The question is whether you're prepared for the transformation that's already happening.

What's your company's plan when a $200/month AI agent can do what currently requires a $60K employee? And more importantly — how are you securing against AI threats your current stack can't even see?