Anthropic shipped a plugin. Two trillion dollars vanished.

One legal automation tool from Claude triggered the biggest AI-driven stock selloff in history. ServiceNow, Salesforce, Thomson Reuters—all cratered. Then OpenAI launched a platform to deploy AI “employees” inside your company. The SaaS industry is being repriced in real time. Below: what’s actually happening, who’s exposed, and the one prompt that maps your personal risk.

In Today’s 5-Minute AI Digest. You will get:

1. The MOST important AI News & research
2. AI Prompt of the week
3. AI Tool of the week
4. AI Tip of the week

…all in a FREE Weekly newsletter. 

OpenAI’s new platform treats AI agents like hires—with onboarding, permissions, and performance reviews. It plugs into your CRM, databases, and internal apps. Uber, Intuit, and State Farm are already live. The play: OpenAI wants to become the operating system sitting between your team and every tool they touch.

[Read our deep dive article]

Anthropic’s Cowork plugins sparked the largest AI-driven selloff in market history. The S&P software index hit its worst non-recessionary drawdown in 30 years. Thomson Reuters down 28%. Microsoft down 13%.

The split: JPMorgan calls it oversold. Goldman warns it’s “the end of the beginning.”

Baseten, DeepInfra, Fireworks AI, and Together AI reported 4–10x cost-per-token cuts by pairing Blackwell GPUs with open-source models. Healthcare AI startup Sully.ai slashed costs 90% overnight.

Translation: The economics that blocked enterprise AI deployment six months ago just disappeared

The second International AI Safety Report—authored by 100+ researchers including Yoshua Bengio, backed by 30+ countries—launched alongside India’s AI Impact Summit this week. Key finding: AI-generated scam content and non-consensual imagery are surging faster than guardrails can contain.

Worth watching: This is becoming regulation’s playbook.

The “AI vs. software” framing is wrong. Software isn’t being replaced—it’s being pushed down the stack.

OpenAI’s Frontier and Anthropic’s Cowork don’t compete with Salesforce. They sit on top of it, mediating every interaction. When an AI agent logs into Salesforce for you, Salesforce becomes plumbing. The AI layer becomes the product.

That’s why the market isn’t just repricing revenue—it’s repricing who owns the enterprise relationship. SaaS survivors will be the ones that build their own agent layers before they become commodity infrastructure underneath someone else’s AI. Watch Salesforce Agentforce and ServiceNow’s next moves—the counter-offensive starts in the next 60 days.

[Read our deep dive article]

What it does: Maps which tools in your stack are most vulnerable to AI replacement—and which are safe—so you act on data, not panic.

The prompt:
I use these software tools at work: [list your tools—e.g., Salesforce, Notion, Jira, Figma, Slack]. For each one, assess: (1) How likely is it that an AI agent could handle 50%+ of what I use this tool for within 18 months? Rate 1–10. (2) What’s the switching cost if I moved to an AI-native alternative? (High/Medium/Low) (3) What’s one concrete change I should make with this tool right now to future-proof my workflow? Present results as a table sorted by disruption risk, highest first. Flag any tool that scores 7+ risk with Low switching cost—those are my immediate action items.

Why it works: The switching-cost dimension is the key. High-risk tools with low switching costs are your triage list—where you can act this week. High-risk tools with high switching costs need a 90-day migration plan, not a weekend panic. This turns abstract market fear into a prioritized to-do list.
Real-world use: Run it, then share the table with your manager. You’ll be the person who showed up with a plan while everyone else was refreshing stock tickers.

What it is: A feature that runs your query across three frontier models (Claude, GPT-5.2, Gemini 3) simultaneously and synthesizes one unified answer.

Why you need it: Every model has blind spots. Model Council shows where they agree (move fast) and where they disagree (dig deeper). It turns AI’s biggest weakness—inconsistency—into a verification layer.

One-liner: “A second and third opinion on every AI answer—automatically.”

Rating: ⭐⭐⭐⭐ (4/5) — Excellent for high-stakes research. Requires Max ($20/mo). Web only for now.

Best for: Investment research, competitive analysis, vendor evaluations—any decision where being confidently wrong is expensive.

Try it: perplexity.ai

The tip: Never use an AI’s first output. Instead, take its response and prompt: “Now act as a skeptical competitor who just saw this. Identify the three weakest assumptions and rewrite the entire thing to survive those attacks.” Use that version.

Why it works: AI models suffer from self-reinforcement bias—when you ask them to “improve” their own work, they polish instead of challenge. Switching to a competitor persona forces genuinely adversarial reasoning. It’s the same principle behind Perplexity’s Model Council, but applied to a single model.

Limitations: Works best on strategic content (plans, proposals, positioning). Less useful for creative writing or code. If the critique feels soft, add “be ruthless—your job depends on finding the flaw.”

Pro move: Chain three rounds: generate → attack → rebuild. The third version is almost always dramatically stronger than the first. I used this on the Our Angle section above.

•       AI platforms are positioning themselves above your SaaS stack—and the market just noticed

•       Inference costs dropped 10x—the economics blocking enterprise AI just evaporated

•       Multi-model verification is the new standard for any AI work where being wrong has a cost

Now implement one.

The person your CEO asks “are we exposed?” won’t be the one who read about the selloff. It’ll be the one who showed up with the audit.

Stay curious—and stay paranoid.

— R. Lauritsen

Share this newsletter:

P.S. Forward this to the colleague who’s still paying for software an AI agent could handle. They’ll owe you one before Q2 planning.