Get tomorrow's brief in your inbox
Today: Elgato Stream Deck now lets AI push your buttons through MCP support. Anthropic accidentally nuked 8,100 GitHub repos trying to pull leaked Claude Code source. Google expanded Search Live globally and Gemini got deeper personal intelligence.
AI agents can now control your Stream Deck, and that matters more than you think. Elgato's Stream Deck 7.4 update adds Model Context Protocol (MCP) support, letting Claude, ChatGPT, and other AI assistants trigger any macro or action you've set up on your Stream Deck. This is bigger than just hands-free button pushing. MCP is becoming the universal connector for AI tools, and when your AI assistant can control physical devices and software macros, you're getting a preview of what agentic AI will look like in six months.
For small businesses, this is the pattern to watch. You won't be building custom integrations anymore. You'll be setting up actions once, then telling your AI assistant to handle them. If you run a content business, podcast, or do any kind of production work, you're already using Stream Deck for scene switching, audio routing, and workflow shortcuts. Now your AI can do that work for you based on voice commands or written requests. The setup requires Node.js and some technical configuration, so it's not turnkey yet. But this is where workflow automation is headed: less manual configuration, more natural language control.
Zapier published the definitive guide to AI business automation - and if you've been wondering how to actually use AI without hiring a tech team, this is it. The guide breaks down where AI automation delivers the fastest ROI: customer support (ticket routing, response drafting), sales (lead qualification, CRM updates), and operations (invoice processing, data entry). The key insight is that AI handles judgment calls while traditional automation handles execution. You don't need AI to create a CRM contact when a form is submitted. You need AI to read an unstructured email and decide if it's a sales inquiry, support request, or spam, then route it correctly. Start with high-volume, predictable tasks that involve some interpretation. Read the full guide
Holo3 sets a new benchmark for desktop automation at a fraction of the cost. This open-source AI model scored 78.85% on the OSWorld desktop benchmark with only 10B active parameters, meaning it can navigate software interfaces and execute multi-step workflows cheaper than GPT 5.4 or Opus 4.6. It's trained on synthetic enterprise environments, so it knows how to handle real business software. The 35B-A3B model is free under Apache2 license and available through their inference API. If you've been waiting for affordable AI that can actually drive desktop software instead of just answering questions, this is worth testing. Holo3 announcement
Google's March AI updates brought practical features you can use today. Search Live is now available in 200+ countries, letting you talk to Google using your camera feed for real-time help (think troubleshooting equipment or identifying parts). Gemini in Docs, Sheets, Slides, and Drive got smarter for AI Ultra and Pro subscribers, with state-of-the-art performance in Sheets for data analysis. Google Maps added "Ask Maps" for conversational queries like "Where can I charge my phone without waiting for coffee?" and can book reservations on the go. Personal Intelligence connects your Gmail, Photos, and other Google apps to give you recommendations that actually match your preferences. If you're already paying for Google Workspace, the Gemini add-on is worth the cost for Sheets alone. Full update list
HP launched HP IQ, an on-device AI layer for workplace collaboration. It turns HP devices into context-aware partners with features like Ask IQ (voice and text queries answered locally) and HP NearSense (spatial intelligence for file sharing between nearby devices). The catch: it's launching on the HP EliteBook X G2 PCs first, with other devices to follow. If you're spec'ing new hardware for your team and you run an HP shop, this might tip the decision. The local-first architecture means most processing happens on-device, reducing cloud dependency and security exposure. Integrates with Microsoft Intune for enterprise management. HP IQ announcement
Building safer AI agents with Zapier MCP: a case study. On March 18, Zapier hosted a webinar showing how to connect OpenClaw (an always-on AI assistant that works through WhatsApp, Telegram, Slack, and iMessage) with Zapier MCP to give it access to 9,000+ app integrations through a single secure connection. The demo included a live security test against prompt injection attacks. The advantage: managed authentication, scoped permissions (your agent can create Gmail drafts but not send emails), and simplified credential management. If you're experimenting with AI agents but worried about security, this architecture is how you do it. You govern what runs through Zapier's platform while your agent handles the logic. Watch the webinar recap
Microsoft Power Automate is worth considering if you live in the Microsoft ecosystem. It's a workflow automation platform built into Microsoft 365 that connects Teams, SharePoint, Dynamics, and Outlook without third-party tools. Best use cases: approval workflows (expense reports, PTO requests), document processing with AI Builder, and connecting Microsoft apps to each other. The limitation: once you step outside the Microsoft world, the integrations get limited fast. If your tech stack is diverse, Zapier or Make are better options. But if you're already paying for Microsoft 365 and your workflows mostly stay inside that ecosystem, Power Automate is right there waiting. What is Power Automate?
Nearly 3 in 4 enterprises admit losing their AI vendor would disrupt core operations. A Zapier survey of 500 U.S. enterprise executives found that 47% would see at least one key business function stop working if their primary AI vendor disappeared, and 27% are completely reliant. Despite 89% claiming they could switch vendors within a month, 58% of those who've actually tried say it either failed or required significantly more effort than expected. The lesson: plan for flexibility before you need it. Don't build everything on a single AI provider. If you're using GPT-4 for summaries, try Claude for a week and make sure your prompts work on both. Vendor lock-in happens slowly, then suddenly. Full survey results
Anthropic accidentally took down 8,100 GitHub repos trying to remove leaked Claude Code source. On Tuesday, Anthropic discovered it had accidentally included access to the source code for Claude Code in a recent release. When the company issued a DMCA takedown notice, it hit 8,100 repositories, including legitimate forks of Anthropic's own public repo. The company retracted the bulk of the notices and restored access. This is a bad look as Anthropic reportedly plans an IPO. For small businesses: if you're relying on AI tools for production work, make sure you have fallback options. Corporate execution matters. Full story
Oracle is cutting thousands of jobs to fund AI data center infrastructure. The company is freeing up cash flow to build out cloud capacity for customers like Nvidia, Meta, OpenAI, AMD, and xAI. Oracle's stock is down 25% year-to-date despite investor optimism about the company's AI buildout. For small businesses: this is the infrastructure play behind the AI services you're using. Expect pricing volatility as cloud providers figure out how to pay for these massive data center expansions. Oracle layoffs story
Cognichip raised $60M to use AI for designing AI chips. The company is building a deep learning model to help engineers design computer chips faster and cheaper, claiming it can reduce chip development costs by 75% and cut timelines by more than half. Intel CEO Lip-Bu Tan joined the board. The company can't yet point to a chip designed with its system, but the funding round signals serious investor interest. For small businesses: this is a long-term bet, but if it works, it means cheaper, faster AI hardware in 3-5 years. Cognichip funding
Meta's Louisiana data center will consume as much power as South Dakota. The company announced it's funding 10 natural gas power plants (7.5 gigawatts total) to power its $27 billion Hyperion AI data center. These plants will dump 12.4 million metric tons of CO2 annually, 50% more than Meta's entire 2024 carbon footprint. Natural gas as a "bridge fuel" is looking more like permanent infrastructure. For small businesses: the environmental cost of AI is real and growing. If sustainability matters to your brand, choose AI services from providers with credible renewable energy commitments. Meta's natural gas expansion
Set up a Python automation script this week. If you're doing repetitive tasks like downloading files, reformatting data, or moving information between systems, Python can handle it automatically. You don't need to be a programmer. Zapier published 9 starter scripts covering API data retrieval, file operations, and task consolidation. Download Python (free), create a .py file, paste a script, and run it from your terminal. Start with something simple like automatically renaming files or pulling data from an API. Once you see it work, you'll find more tasks to automate. 9 Python automation scripts
Test different AI image generators before committing to one. ChatGPT (best overall), Midjourney (artistic results), Ideogram (accurate text), and FLUX (customization) all have different strengths. If you're creating social media graphics, marketing materials, or presentations, spending an hour testing prompts across 2-3 platforms will show you which one matches your workflow. Most offer free trials. Don't assume the first one you try is the best fit. Best AI image generators 2026
The theme this week is infrastructure and integration. MCP is becoming the USB cable for AI, Stream Deck showed how physical devices connect to AI assistants, and enterprises are realizing vendor lock-in is real. The gap between "AI can do this in a demo" and "AI can do this reliably in production" is narrowing, but execution still matters. Anthropic's GitHub mess proves that even leading AI companies make operational mistakes. For small businesses, this means the tools are getting better and easier to integrate, but you still need fallback plans and vendor diversity. Watch for more MCP integrations this month as the standard gains momentum.