Get tomorrow's brief in your inbox
Today: Zendesk shipped specialized AI agents that automate 80% of workflows for specific industries, no coding required. Amazon's Seller Assistant gained memory and 24/7 monitoring, cutting manual work for small sellers. Meta pushed Muse everywhere, including glasses, Mac, and a $1,299 VR headset.
Zendesk just made AI agents useful for actual businesses. Their new Specialized AI Agents come in two flavors: Industry Agents (preconfigured for sectors like retail) and Custom Agents (built with their no-code Agent Builder). The big claim is automating up to 80% of workflows, which matters if you're a small business where every hour counts.
The Industry Agents for commerce integrate with Shopify, Narvar, and Stripe to handle orders, returns, and customer questions without human intervention. Custom Agents let you build AI tailored to your processes using Zendesk's Agent Builder. They've also partnered with Riskified to embed fraud detection directly into commerce workflows, reducing fraud losses while keeping legitimate transactions smooth.
Business impact: If you run e-commerce, this could significantly reduce response times and free up staff for strategic work. The no-code approach lowers the barrier, but expect a learning curve as you configure agents to fit your operations. Monitor performance closely to ensure AI maintains the personalized service customers expect.
Read more: Zendesk Unveils Specialized AI Agents
Amazon Seller Assistant gets 24/7 automation and Claude integration. Amazon's Seller Assistant now runs workflows continuously, monitoring inventory, pricing, competitors, and account health even when you're logged out. It also connects to Anthropic's Claude via a plugin, letting you bring Amazon data into AI tools you already use. The assistant has persistent memory across conversations and can connect insights across advertising, inventory, and listings. Seller Assistant reaches 90% of Amazon's selling partners, with recommendations accepted over 90% of the time. If you manage 15+ Amazon SKUs, this could eliminate hours of manual monitoring per week.
Read more: Amazon Seller Assistant Adds 24/7 Automation
Ringg's AI agents handle 7 million calls monthly using GPT-5.6. Ringg built a voice and chat agent platform for customer service in India, handling more than 7 million connected calls per month with a 4.8 CSAT score. By migrating suitable workloads from GPT-4.1 to GPT-5.6, they cut model costs by 90% while maintaining quality and latency. Agents resolve up to 65% of requests and operate across voice, chat, WhatsApp, and web. The platform coordinates multiple specialized subagents for qualification, support, verification, and scheduling. If you're scaling customer service and can't afford to hire proportionally, this is the playbook.
Read more: Ringg's AI agents resolve up to 65% of customer calls
Google's Gemini 3.8 TTS models with 2,000+ voices. Google released Gemini 3.8 Flash TTS and Flash-Lite TTS with a library of over 2,000 voices, plus custom voice creation from a 30-second audio sample. The API supports multi-character conversations with different voices and styles. Example: 1m 18s of audio (two-character dialogue) generated in ~20 seconds for 2.74 cents using Flash TTS. If you need audio content at scale, this is significantly cheaper than alternatives.
Read more: Gemini 3.8 TTS Playground
How Figma's People Ops automated 100+ background checks per month. Ariel Chen at Figma built a system to triage background checks across 12 countries. Before automation, her two-person team manually reviewed 3,000+ checks monthly, spending ~3 minutes per check (150 hours/month). The problem: every check showed "pending," but escalation paths vary by which screen is pending. She built four Zapier workflows pulling live data from Checkr into Google Sheets, with daily AI triage identifying which pending screens are urgent based on start dates. Invalid checks dropped from 8% to under 2%. She ran accuracy tests for five consecutive clean days before the team relied on it. Takeaway: Test rigorously before trusting automation in high-stakes workflows.
Read more: How Ariel Chen built trust before automation
How Jobber turned AI access into a building culture. Ethan Schwandt at Jobber ran a leadership hackathon for 55 directors and hands-on sessions for 161 employees, paired with office hours and clear governance. Active Zapier builders grew 122% in 30 days. Teams are now building workflows across 28 business apps (Google Calendar, Sheets, Slack, HubSpot, Salesforce, Asana, Typeform). Employees are moving beyond basic workflows into Code by Zapier and writing code directly in Zaps. Takeaway: Enablement that starts with leadership and provides hands-on support drives adoption across teams, not just specialists.
Read more: How Jobber turned AI adoption into a building culture
How Galgo automated fraud detection for 3,000 delivery photos monthly. Ignacio Piñeiro at Galgo (a Mexican fintech lending for motorcycles) built an exception-based AI review system. Before: 150 hours/month of manual review, 8% invalid evidence slipping through. After: AI by Zapier (GPT-5-mini) validates every photo automatically (customer + vehicle together, no stock photos), flags exceptions to Slack for human review. Invalid evidence dropped to under 2%. Manual review time fell to fewer than 60 photos monthly. Takeaway: Focus AI on consistent judgment, route exceptions to humans, and test rigorously before relying on automation.
Read more: How Galgo scaled fraud control with AI
Meta pushes Muse AI agent into glasses, Mac, and a $1,299 VR headset. Meta announced that Muse will integrate with its smart glasses (activates via wake word, guides workouts, logs meals, books appointments). Muse is also coming to Mac with computer use, letting the agent operate any desktop app. Users will be able to video chat with their Muse avatar (powered by Muse Realtime Avatar model). Muse agents get their own email addresses to handle tasks and receive messages. Meta also unveiled the Muse Charm, a Tamagotchi-like wearable with a fingerprint sensor and camera, shipping in December. The Meta VR Glasses ($1,299, spring 2027) are slimmer and lighter than Quest, powered by an external puck to reduce weight. Business implication: Meta is betting heavily on Muse as a consumer AI platform. Wait to see if the ecosystem actually delivers value before investing in hardware.
Read more: Everything new coming to Meta's AI agent Muse
OpenAI extends Daybreak cyber defense program to Ukraine. OpenAI is giving Ukraine's government access to its Daybreak program to help defend civilian infrastructure against cyberattacks. Daybreak provides AI tools to identify software vulnerabilities and develop fixes more quickly. Ukraine's CERT-UA handled nearly 6,000 cyber incidents in 2025, including attacks on hospitals, energy, and telecom. OpenAI has already provided access to cyber defenders in France, Germany, Poland, and the EU's ENISA, which used the models to discover vulnerabilities now fixed.
Read more: OpenAI extends cyber access to Ukraine
OpenAI Academy expands with Community Trainer Program. OpenAI Academy has hosted 250+ events over two years, with 4 million people engaging with content. They're launching a trainer program to prepare people and organizations to teach Academy material in their communities. Partners nominate staff to learn the curriculum and lead workshops. The program includes self-paced courses, practical guides, in-person workshops, and AI Skills Jams. New learning paths cover knowledge workers, developers, leaders, educators, and college students. Takeaway: If you run a community organization or small-business network, this could be a resource to bring practical AI training to your members.
Read more: Two years of OpenAI Academy
Exception-based review pattern: Don't try to automate everything faster. Instead, use AI for consistent judgment on routine cases and route only exceptions to human reviewers. Galgo's fraud detection system is the template: AI validates 3,000+ photos monthly, humans review fewer than 60 flagged cases.
Accuracy testing before relying on automation: When building high-stakes workflows, compare AI output against your source of truth screen by screen. Run the test until you see clean results for at least five consecutive days. Figma's background-check triage is the example: Ariel Chen tested daily before her team relied on the system.
Specialized AI agents are moving from hype to production. Zendesk's industry-specific agents, Amazon's 24/7 Seller Assistant workflows, and Ringg's 7-million-call-per-month platform show that businesses are deploying AI to handle real work at scale. The pattern that's working: narrow focus (fraud detection, background checks, customer calls), rigorous testing, and exception-based human review. Meta's Muse push into glasses, Mac, and wearables is a bet that consumer AI will follow the same trajectory. Watch how businesses use these tools in practice before committing budget.