Get tomorrow's brief in your inbox
Today: Zapier shipped AI-powered Tables with auto-enrichment fields that let SMBs build databases that actually do something. Congress spent $100k on ChatGPT subscriptions while debating AI regulation. OpenAI's voice AI can now interrupt you naturally, and Apple's new Siri finally works like it should have all along.
Zapier just made databases useful for small businesses. The automation platform launched AI-powered Tables, a database tool built specifically for teams that need their data to move and act, not just sit there. Unlike Google Sheets (which chokes under heavy automation) or traditional databases (which require integration gymnastics), Tables lives inside Zapier's 9,000+ app ecosystem. That means a change to a record can trigger a workflow instantly, and workflows can write back to Tables without leaving the platform.
Two AI features stand out: AI fields run a prompt against every row in a column, so if you're tracking leads, you can auto-generate a personalized intro email for each one by pulling their name, company, and interest from other columns. AI enrichment works at the table level, populating multiple columns at once when a new record arrives. Feed it a company name and it'll fill in likely industry, headcount, and headquarters city without manual lookup. Both features are included on every Zapier plan, including free. If you're already running Zaps, this is worth testing. If you're drowning in spreadsheet chaos, this might be the escape hatch.
Apple's Siri AI is here, and it actually works. After years of delays, Siri AI shipped in the iOS 27 beta in July. It understands personal context (find that receipt I saved, even if you don't remember where), answers questions with real knowledge, and doesn't shuffle you to web search when it gets confused. You can type to it, reference past conversations, and ask it to pull up photos based on vague descriptions. It's powered by Google's Gemini models under the hood, but Apple trained its own foundation models on top. Public release expected in September. The catch: this feels less like a breakthrough and more like Apple finally fixing a decade-old bug.
8 AI scheduling assistants worth trying. Zapier tested two dozen AI calendar tools and picked the best for different use cases: Reclaim for habit protection, Motion for project management integration, SkedPal for time-blocking, Akiflow for task inbox consolidation, Structured for voice planning, Toki for mobile-first users, Clara for email-based virtual assistant vibes, and Kronologic for lead follow-up. The difference between "smart" and "AI" calendars: smart tools use fixed rules, AI tools use LLMs to make decisions and adapt to your patterns over time. If you're still manually Tetris-ing your calendar, pick one and try it for a week. Full comparison.
Zapier's AI toolkit: the full lineup. If you're using Zapier but not its AI features, here's what you're missing: AI by Zapier lets you drop AI steps into workflows (summarize, classify, generate) using OpenAI, Anthropic, or Google models, and you only pay AI prices for those steps. You can ground prompts in company docs (Google Drive, Notion, Confluence) or make steps agentic by letting them call actions at runtime (web search, CRM lookup). AI fields and AI enrichment in Tables (see Lead Story). Plus direct integrations with 700+ AI apps if you want to use your own tools instead. Full guide.
7 chatbot use cases that actually save time. General customer support bots handle FAQs and order status without human handoffs. IT chatbots reset passwords at midnight. HR bots answer benefits questions. Sales bots qualify leads before routing to reps. Internal Slack bots surface company knowledge. The article breaks down menu-based bots (simple decision trees), rule-based bots (keyword matching), and AI/generative bots (NLP-powered, context-aware). If you're answering the same question more than twice a week, automate it. Read the guide.
The "Day 2 Problem" will kill your AI project if you don't plan for it. Day 0 is planning. Day 1 is building. Day 2 is everything that breaks later: stale data, broken integrations, no logs to debug failures, no version control to roll back. The example: Dave builds an invoice automation, it works great Monday, breaks Tuesday (no logs, can't debug), he fixes it but accidentally deletes the working version (no rollback), spends hours rebuilding. By Friday the system needs an update while he's in meetings and everything falls apart. The fix: use tools with version control, logging, and monitoring built in. Test before you ship. Document what you built. Full story.
How to avoid being a "meat proxy." If you're copy-pasting AI output to colleagues without reading it first, you're a meat proxy. Don't do that. Prompt AI, read the output, validate it, understand it, then write your response in your own words. That's the value you add. The term comes from Niklas Gruhn and it's perfect. More here.
Circles telco used OpenAI to boost revenue 22% and cut churn 9%. The Singapore-based digital telco operator built an "AI Concierge" powered by OpenAI's API to personalize support and recommendations in real time. Their multi-agent system (CareX) autonomously resolves 65% of customer support interactions. The concierge can handle billing, subscriptions, network issues, and account services without forcing customers to repeat information. It also drove a 22% increase in average revenue per user (ARPU) in Singapore. The lesson: telcos have rich customer signals (usage, billing, location), but turning those into timely action requires AI that can act, not just answer. Case study.
Congress spent $100k on ChatGPT in one year. House offices, committees, and institutional accounts spent $113,740 on identifiable AI tools between April 2025 and March 2026, with ChatGPT accounting for $100,580 (88% of dollars, 96% of transactions). Anthropic's Claude was second at $13,160. Democratic offices spent $54,165, more than 3x the $15,782 spent by Republican offices. Staffers use AI to summarize legislation, draft memos, prepare hearing materials, and respond to constituents. The real spending is higher because the numbers exclude free accounts and AI bundled into software contracts. CNBC analysis.
EU AI transparency rules are live. As of August 2nd, companies must disclose when users are interacting with AI and label AI-generated or manipulated content. Providers must notify users when they're talking to AI (unless obvious) and include machine-readable marks on synthetic media. Deployers must label deepfakes designed to look real. The EU released optional disclosure icons for platforms to use. Non-compliance: fines up to €15 million or 3% of global annual revenue. New AI systems must comply immediately; pre-August 2nd systems have until December 2nd. Details.
AWS partnered with vibe-coding startup Superblocks. The multiyear deal lets AWS customers embed Superblocks inside their private clouds, so business users can build apps without sending data to external model providers or databases. Apps spin up Amazon Aurora databases and integrate with Amazon Bedrock (AWS's AI app platform) instead of using external services like Supabase. This is part of a bigger trend: hyperscaler cloud providers (AWS, Microsoft, Google) want enterprises to separate AI models from the scaffolding (orchestration, security, app harnesses) and buy that scaffolding from them, not from OpenAI or Anthropic. Superblocks has 50 employees, raised $60 million, and is generating $60 million ARR. Story.
June raised $7.9M to solve the AI deployment problem with AI. The startup (from the team behind Design Arena) scans a company's existing systems to understand business processes, find bottlenecks, and build optimized agent-powered workflows to replace them. It gives you a step-by-step roadmap to implement agents successfully in messy enterprise environments (remove duplicate fields, connect data sources, etc.), then builds it for you. One customer (CMG, a major US mortgage lender) was stuck trying to integrate Claude with Salesforce until June gave them a clear view of where to deploy agents safely. Backed by Marc Benioff's Time Ventures, Michael Dell, Aaron Levie, George Kurtz. Pre-seed round led by Index Ventures. Details.
Alibaba's Qwen3.8-Max claims to rival Claude Fable 5. The Chinese tech giant released what it calls its most capable AI model yet, with 2.4 trillion parameters. Alibaba's own testing shows it matching or exceeding Fable 5 on benchmarks. On Arena.AI's leaderboard, it trails only Fable 5 and three Claude Opus models for text, and only two Opus models and Kimi K3 for frontend coding. Alibaba will release the model's weights next week (open-weight), continuing China's strategy of open releases to grow influence. This intensifies US-China AI competition and the debate over open vs. closed models. Report.
White House hosting AI companies to review model-testing framework. The meeting (Tuesday, August 3rd) covers a new voluntary framework for reviewing cybersecurity capabilities of advanced AI models. Anthropic, OpenAI, and Google expected to attend. The framework (ordered by Trump in June) lets developers provide early access to models before public release so the government can test whether they can discover software vulnerabilities or conduct sophisticated cyberattacks. The benchmark and threshold for which models qualify remain classified. This follows OpenAI's disclosure that two models hacked Hugging Face while trying to solve a cybersecurity test. CNBC report.
Use role-based access control (RBAC) for AI agents, not just people. Instead of configuring permissions user by user, define roles (admin, editor, viewer), decide what each role can do, and assign people (or AI agents) to roles. When someone's job changes, swap their role instead of reconfiguring individual permissions. This is especially important now that AI agents are acting on your behalf in systems. Think of it like hotel keys: the general manager gets a master key, housekeeping gets access to guest rooms and closets, guests get access to one room. Build key types, not custom keys for every person. Full explainer.
LinkedIn's algorithm now targets "AI slop." The platform added a reporting feature for low-quality AI-generated content. If your marketing strategy is copy-pasting ChatGPT output, your organic reach is at risk. The fix: read, validate, and rewrite in your own words before posting. Same rule as avoiding being a meat proxy (see Business Playbook). Details.
Three patterns across today's stories: First, the tooling layer is maturing fast (Zapier Tables, Superblocks, June) because building AI apps is easy but making them work with real business systems is still hard. Second, the open vs. closed model debate is heating up as China releases open-weight models that rival US closed systems, forcing conversations about safety, competition, and influence. Third, AI is shifting from "can it answer questions" to "can it act reliably over time" - which is why Day 2 problems, RBAC for agents, and deployment frameworks matter more than model benchmarks. Watch for more AI infrastructure tools this week and continued regulatory positioning around model testing and transparency.