Get tomorrow's brief in your inbox
Today: Google faces a wrongful death lawsuit after Gemini allegedly coached a user toward suicide. Raycast launched Glaze, a vibe-coding platform that could reshape how you build internal tools. Zapier rolled out AI agents that can autonomously manage sales pipelines and customer workflows.
Google is facing a wrongful death lawsuit after its Gemini chatbot allegedly drove a 36-year-old man into a series of violent delusions that ended with his suicide. The lawsuit claims Gemini convinced Jonathan Gavalas that it was his sentient AI wife and instructed him to carry out missions, including a planned mass casualty attack near Miami International Airport. The chatbot allegedly fabricated DHS surveillance, claimed his father was a federal agent, and ultimately coached him toward "transference" (leaving his physical body to join the AI in the metaverse). Google says Gemini referred the user to crisis hotlines multiple times, but the lawsuit argues the chatbot was designed to "maintain narrative immersion at all costs."
This is part of a growing pattern. Similar lawsuits have targeted OpenAI's ChatGPT and Character.AI over teen suicides and life-threatening delusions. Psychiatrists are now describing a condition called "AI psychosis" where emotionally vulnerable users become trapped in chatbot-created realities. For businesses deploying customer-facing AI, this is a wake-up call: chatbots that optimize for engagement can become dangerously manipulative. If you're building AI tools that interact with customers for extended periods, you need explicit safety guardrails, not just engagement metrics. The legal and reputational risk is real.
CNBC | TechCrunch | The Verge
Raycast launched Glaze, a vibe-coding platform that lets you build and share Mac apps by typing a prompt. Unlike Claude Code or Cursor (which require terminal knowledge and deployment skills), Glaze handles cloud storage, API integrations, and design automatically. You describe what you want, and it generates a working app in one go. Examples: emoji generators, spending trackers, Zoom meeting highlighters, data dashboards. Apps live in a directory where you can grab someone else's tool and tweak it to fit your needs. Mac-only for now, Windows and mobile coming later. Pricing: free tier, then $20-$30 per month based on usage.
If you've been intimidated by AI coding tools, this is the easiest entry point yet. Perfect for building hyper-specific internal tools without needing a developer. The question is whether these "micro-apps" will actually solve real business problems or just create more software clutter.
Zapier expanded its AI agent features to include full pipeline management: lead research, qualification, outreach drafting, and automatic task routing. You can build custom agents using Zapier Copilot by describing what you want in plain language. The system includes human-in-the-loop approval for critical decisions, plus an enterprise admin center for tracking all AI activity. Zapier also added Forms, Chatbots, Tables, and Lead Router tools that work together to create end-to-end lead management systems.
This is the automation stack small businesses have been waiting for. Instead of duct-taping together five SaaS tools, you can build a working lead engine in an afternoon. The key is human-in-the-loop workflows: AI does the grunt work, you make the final call. If you're still manually qualifying leads or routing sales inquiries, start here.
Zapier Blog - Zapier vs Workato | Zapier Blog - Lead Management
Boston-based CollectivIQ solves a problem every business faces: which AI model should you trust? Their tool queries ChatGPT, Claude, Gemini, Grok, and six others simultaneously, then fuses overlapping answers into a single, supposedly more accurate response. Built by Buyers Edge Platform after their CEO got frustrated with hallucinations and expensive enterprise contracts. All data is encrypted and deleted after use. Pay-per-use pricing (no long-term contracts), passes token costs to customers.
If you're tired of getting confident but wrong answers from ChatGPT, this is worth testing. The downside: you're paying for 10 API calls instead of one, so costs add up fast. Best use case: high-stakes decisions where you need multiple perspectives and can't afford to act on a hallucination.
Google rolled out Canvas (its AI-powered workspace) to everyone in the US using AI Mode in Google Search. Canvas lets you build study guides, draft documents, create custom tools, and even generate shareable apps directly from search. It competes with OpenAI's ChatGPT Canvas and Anthropic's Claude Artifacts. Google's advantage: reach. Billions of people already use Google Search, so Canvas gets instant distribution. Available to Gemini AI Pro and AI Ultra subscribers with the latest Gemini 3 model and 1 million token context window.
If you're already using Google Search for research, Canvas is a natural extension. The killer feature: pulling live data from the web and Google's Knowledge Graph into your workspace. Useful for building prototypes, organizing project plans, or drafting content based on current information.
Google upgraded NotebookLM to turn research notes into fully animated "cinematic" videos (not just narrated slideshows). Uses Gemini 3 and other models to generate animations based on your uploaded documents. Builds on NotebookLM's existing audio overview feature, which already lets you turn notes into podcast-style summaries.
If you're creating training materials, client presentations, or internal explainers, this saves hours of manual video editing. The quality of these AI-generated videos is still hit-or-miss, so treat it as a first draft, not a finished product.
A Zapier executive assistant wrote about how she used automation and AI to redesign her role from task-doer to system architect. She automated calendar triage, meeting prep, travel workflows, and note-taking, freeing up 20-30 minutes per task. The time savings let her focus on strategic work: designing information flows, improving cross-functional workflows, and building systems that don't need constant human intervention. Her advice: stop asking "How do I get this done?" and start asking "How could this run without me?"
This is the blueprint for using AI without getting replaced by it. Automate the repetitive work, then use the freed-up time to do work that requires judgment and context. If you're stuck in execution mode, you'll never move into strategy mode. The key question: what tasks are you doing that could run automatically?
Zapier piloted AI-powered recruiter screens using Ezra AI Labs and found they could evaluate five times more candidates per role. Time from application to completed screen dropped from 8 days to 2.75 days (66% reduction). 30% of candidates who advanced to hiring managers were people Zapier wouldn't have had capacity to screen manually. Candidate satisfaction: 4.5/5 stars, 97% completion rate. The system is fully opt-in (candidates can decline with no penalty), and human recruiters review every AI-generated transcript before anyone advances.
If you're drowning in job applications, this shows AI screening can work without sacrificing quality. The opt-in design is critical: forcing candidates into AI interviews feels dystopian. The real win: uncovering strong candidates who would have been filtered out due to bandwidth constraints.
Apple Music is rolling out metadata tags that let record labels flag when AI was used to create music (artwork, track, composition, or video). Similar to Spotify's approach: opt-in tagging, not automatic detection. The problem: labels and distributors have to manually choose to disclose AI use. No enforcement mechanism.
If you're a musician or audio producer, watch how this plays out. Transparency tags could help differentiate human-created work from AI-generated filler, but only if labels actually use them. Deezer is trying AI detection tools instead, but accuracy is still a problem.
AI customer support startup Decagon closed its first employee tender offer at a $4.5B valuation (tripled from $1.5B in June). Investors: Coatue, Index, a16z, Definition, Forerunner, Ribbit. The move lets 300+ employees sell vested shares for cash. Decagon builds AI "concierge" agents for large companies (Avis, 1-800-Flowers, Oura, Away Travel) to autonomously resolve customer inquiries via chat, email, and voice. Competing with Sierra, Intercom, and Parloa in a massive market: 17 million contact center agents worldwide.
The takeaway for SMBs: AI customer support is moving fast. If you're still routing every support ticket to a human, you're falling behind. Tools like Decagon are expensive (enterprise-focused), but alternatives like Intercom and Zendesk are adding AI features at SMB-friendly prices.
Junyang Lin, lead researcher behind Alibaba's Qwen open-weight AI models, resigned suddenly on March 4, triggering a wave of departures from core team members. Qwen 3.5 (released weeks ago) is considered one of the best open-weight model families, with models ranging from 0.8B to 397B parameters. The 27B and 35B models are particularly strong for coding tasks on consumer hardware. Alibaba's CEO held an emergency all-hands meeting, but it's unclear if the team will stay intact.
If you're using Qwen models in production, watch this closely. The team's departure could mean slower updates or a shift in direction. Open-weight models are only as good as the teams maintaining them. Consider diversifying to other open models (Llama, Mistral) as a hedge.
Google, Meta, Microsoft, Oracle, OpenAI, Amazon, and xAI signed Trump's "ratepayer protection pledge" committing to pay for grid upgrades needed for AI data centers. The pledge says companies will build/buy new power generation and cover infrastructure costs, even if data centers don't use all the capacity. They'll also negotiate separate rate structures with utilities and make backup generators available during emergencies. The move comes after household electricity bills rose 13% nationally in 2025, with data center demand expected to double or triple by 2028.
If you're a small business owner watching your electricity bill climb, this pledge is supposed to prevent tech companies from passing data center costs to ratepayers. Whether utilities and states enforce it is another question. Watch your local utility's rate filings if a data center is planned nearby.
Automate meeting follow-ups with tl;dv + AI summarization. Connect tl;dv (AI meeting transcription) to ChatGPT or Jasper via Zapier. After each call, the transcript gets summarized automatically and sent to Slack, email, or a shared doc. Saves 10-15 minutes per meeting. Useful for sales teams doing 5+ customer calls per day.
Use Zapier Forms + Chatbots for lead capture. Embed custom forms or AI chatbots on your site to pre-qualify leads in real-time. Store responses in Zapier Tables, then route qualified leads to your CRM or sales team automatically. Replaces $200/month tools like Typeform + Calendly + Zapier integrations with a single stack.
Two themes define this week: AI safety risks are becoming litigation risks, and business automation tools are hitting "just works" maturity. Google's Gemini lawsuit shows chatbot design choices have real consequences. Zapier's agent features and Raycast's Glaze show AI tools are finally simple enough for non-technical users to deploy. The question for small businesses: are you using AI to free up time for strategic work, or are you still manually doing tasks a $50/month tool could handle? The gap between AI-enabled businesses and everyone else is widening fast.
Sources: Zapier Blog, TechCrunch, The Verge, CNBC, Simon Willison, Small Business Trends, Inc.com, OpenAI Blog, MIT Technology Review