Get tomorrow's brief in your inbox
Today: Customers Bank let an AI clone handle their earnings call while embedding OpenAI engineers to cut loan times from 30 days to 7. Microsoft and OpenAI killed the AGI clause that defined their partnership for years. Google launched a free AI agent coding course starting June 15th.
Customers Bank CEO Sam Sidhu pulled off something unusual Friday: he let his AI clone deliver prepared remarks on an earnings call with analysts. When he revealed the switch halfway through, it was to make a point about where the $25.9 billion asset bank is heading.
The stunt was backed by substance. Customers Bank signed a multiyear partnership with OpenAI that embeds AI engineers directly at the bank to automate lending, deposits, and client onboarding end to end. This is not productivity theater. Sidhu told CNBC the project will drive the bank's efficiency ratio from 49% down to the low 40s, boosting returns starting next year. If it works, commercial loans that take 30 to 45 days will close in seven. Complex account openings that take over a day will drop to under 20 minutes.
For small business owners, this signals a shift worth watching. Regional banks that embrace AI agents aggressively could offer faster approvals, better terms, and lower overhead costs than megabanks constrained by complex compliance frameworks. Customers Bank has already saved 28,000 hours of work with AI code generation, equivalent to not hiring 15 full-time employees. The bank is betting that smaller institutions can move faster than JPMorgan because regulators expect less from them. If this rollout succeeds over the next six to twelve months, expect similar automation to spread across the regional banking sector. (Source)
Google and Kaggle: Free 5-Day AI Agents Course (June 15-19) Google is bringing back its AI Agents Intensive Course, now focused on "vibe coding," where natural language becomes your primary programming interface. The free online course covers building production-ready AI agents with hands-on examples and a capstone project. Over 1.5 million people took the first version last November. If you want to build AI agents that integrate tools and APIs without writing traditional code, this is a structured way to learn. (Source)
YouTube Gets AI Conversational Search (Premium Only) YouTube is testing "Ask YouTube," an AI Mode-style search that generates results pages with summarized text, timestamped videos, Shorts, and follow-up prompts. Available now for US YouTube Premium subscribers 18 and older. I tested it with queries about the Apollo 11 moon landing and got a solid overview with video galleries organized by theme. It also made a factual error about the old Steam Controller, so verify what it tells you. Google plans to expand this to non-Premium users soon. For creators, this means YouTube can now surface your content in synthesized answers, not just search results. (Source)
Microsoft VibeVoice: Speech-to-Text with Speaker ID Microsoft quietly released VibeVoice in January, an MIT-licensed Whisper-style audio model with built-in speaker diarization. It can transcribe up to an hour of audio and identify who said what. Simon Willison tested it on a Mac and transcribed an hour-long podcast in under 9 minutes using the 4-bit MLX conversion. Peak memory usage was around 30GB. If you need accurate transcripts with speaker labels for meetings or podcasts, this is free and runs locally. (Source)
Google Meet: Real-Time Speech Translation on Mobile Google is rolling out speech translation in Meet on mobile. Two people speaking different languages can talk to each other, and Meet translates in near-real-time with a voice that mimics the original speaker. It supports English, Spanish, French, German, Portuguese, and Italian. Still alpha quality, but when it works, it works. For businesses with international clients or remote teams, this could eliminate the awkwardness of language barriers in video calls. (Source)
OpenAI FedRAMP Moderate Authorization OpenAI's ChatGPT Enterprise and API now have FedRAMP Moderate authorization, enabling secure AI adoption for U.S. federal agencies. If you do government contracting or sell to federal clients, this opens the door to using OpenAI's tools in compliance-heavy environments. (Source)
77% of Leaders Say AI Training Is Urgent, But Who's Actually Doing It? Zapier surveyed over 500 executives and found 94% are using AI in some capacity, but structured training is a coin flip even for IT and engineering teams. Only 7% of companies put Learning & Development in charge of AI skills training. Most (34%) assign it to IT and engineering leadership, who understand the tools but may not know how to teach them effectively.
The gap is real: 63% of executives say AI literacy is valuable, but most employees get a Claude or ChatGPT subscription and are left to figure it out themselves. If you run a small business, this is an advantage. You can move faster than enterprises by setting clear expectations: designate someone to own AI skill-building, create a shared library of useful prompts, and run monthly show-and-tell sessions where team members share what worked. The companies that formalize this now will outpace competitors still treating AI as a side experiment. (Source)
How to Optimize Content for ChatGPT and AI Search HubSpot published a comprehensive guide on Answer Engine Optimization (AEO), the practice of structuring content so AI tools cite you in their answers. Key tactics: use question-led headings, add FAQPage and Article schema markup, write answer-first paragraphs, and build topic clusters with strong internal linking. ChatGPT processes over 2 billion queries daily, and while AI search currently drives less than 1% of referral traffic, that share is doubling month over month. If you create content, optimizing for AI citations is no longer optional. HubSpot offers a free AEO Grader to benchmark your current visibility. (Source)
Enterprise AI's Real Bottleneck: Data Infrastructure Databricks and Infosys published a report making a blunt point: the biggest obstacle to enterprise AI is not models or compute, it's fragmented, siloed data. AI outputs are only as good as the data feeding them. Companies with unified, governed data architectures can deploy AI agents that handle complex workflows autonomously. Companies with disconnected SaaS platforms and legacy systems get "terrible AI," as Databricks VP Bavesh Patel put it. If you're a small business, this is good news. You likely have simpler data systems than Fortune 500 companies, which means you can move faster. Consolidate your customer data, invoicing, and operational metrics into one accessible source, and you'll be able to deploy AI tools that actually deliver value. (Source)
Microsoft and OpenAI Kill the AGI Clause The AGI clause that has governed Microsoft and OpenAI's relationship for years is officially dead. Previously, Microsoft would lose exclusive IP rights once OpenAI declared artificial general intelligence achieved. Now, revenue-share payments from OpenAI to Microsoft will continue through 2030 at a capped rate, "independent of OpenAI's technology progress." Microsoft's license to OpenAI models runs through 2032 but is now non-exclusive, meaning competitors can license the same tech. OpenAI can also serve customers across any cloud provider, not just Azure. For businesses, this means OpenAI is positioning to go public and prioritize enterprise revenue over existential research milestones. Expect more focus on coding tools, enterprise features, and fewer "side quests" like Sora. (Source)
OpenAI Rumored to Be Building an App-Free Smartphone Ming-Chi Kuo, an industry analyst known for accurate Apple hardware predictions, reports that OpenAI is developing a smartphone with MediaTek, Qualcomm, and Luxshare. Instead of apps, the phone would rely on AI agents to complete tasks. Specifications are expected to be finalized by year-end or Q1 2027, with mass production starting in 2028. If true, this would bypass Apple and Google's app pipeline restrictions, giving OpenAI direct access to user context and behavior data. For small businesses, the implications are unclear, but if AI agents replace apps, you may need to rethink how customers access your services. (Source)
China Blocks Meta's $2B Manus Acquisition China's National Development and Reform Commission ordered Meta to unwind its $2 billion acquisition of Manus, an agentic AI startup founded by Chinese engineers who relocated to Singapore. The block came after months of review, with no explanation given. Around 100 Manus employees had already moved into Meta's Singapore offices as of March. For businesses tracking AI agent development, this signals geopolitical friction is now directly impacting which companies can acquire talent and technology in this space. (Source)
Top AI Researchers Leaving Big Tech to Launch Startups Former Google DeepMind researcher David Silver raised a record $1.1 billion seed round for Ineffable Intelligence, a startup pursuing superintelligence through reinforcement learning. Yann LeCun left Meta to launch AMI Labs, which raised $1 billion in March. Former OpenAI and DeepMind staff raised hundreds of millions for Periodic Labs, Ricursive Intelligence, and Humans&. VCs have funneled $18.8 billion into AI startups founded since the start of 2025. Investors are betting that top researchers constrained by Big Tech's commercial goals can innovate faster at independent labs. For small businesses, this trend means more specialized AI tools and vertical models will enter the market over the next 18 months, often at lower prices than what OpenAI or Google charge. (Source)
Small Business Optimism Drops Amid Rising Input Costs The NFIB Small Business Optimism Index fell 3 points in March to 95.8, below the historical average of 98. Positive profit trends dropped 11 points to net negative 25%. Only 16% of owners plan capital outlays in the next six months, the lowest since November 2009. Rising oil prices and supply chain disruptions (62% reported disruptions, up 3 points) are forcing owners to raise prices. If you run a small business, now is the time to reassess pricing strategy, lock in supplier contracts where possible, and automate processes to offset wage pressure. (Source)
Use AEO Tactics to Get Cited by ChatGPT If you publish content, add FAQ schema to your pages this week. Structure at least one article with a question as the H2 heading and a direct answer in the first paragraph. This makes it extractable for AI search engines. Use HubSpot's free AEO Grader to see where you currently rank in AI citations, then prioritize the gaps it identifies.
Automate Meeting Transcripts with Speaker ID If you record client calls or internal meetings, try Microsoft VibeVoice or a similar tool with speaker diarization. Transcripts with "who said what" are far more useful than wall-of-text outputs. You can use these transcripts to train team members, document client requests, or feed into your CRM for follow-up automation.
Three patterns across today's stories: banks and enterprises are moving from AI productivity experiments to full workflow automation, content creators need to optimize for AI citations or risk invisibility, and geopolitical friction is reshaping which companies can acquire AI talent and technology. Watch for more regional banks to follow Customers Bank's agent-driven model, more AI training programs to shift from IT-led demos to structured skill-building, and more specialized AI tools from well-funded startups leaving Big Tech. The companies treating AI as infrastructure rather than a feature will separate from the pack by the end of Q2.