Get tomorrow's brief in your inbox
Today: Salesforce launched Agentforce Contact Center to give AI and human agents unified CRM context. Simon Willison's "profile this user" experiment shows how LLMs can distill identity from comment history. Jensen Huang says engineers should get half their salary again in AI tokens, sparking debate on whether that's a bonus or just infrastructure cost.
Salesforce just dropped Agentforce Contact Center, and it's the clearest example yet of how AI agents and human teams will actually work together in small businesses. The platform lets AI handle routine support tickets automatically but hands off complex cases to humans with full conversation context already loaded. No more "let me pull up your account" dead air. No more asking customers to repeat themselves.
This matters because most small businesses still run support on patchwork systems: email, a phone line, maybe a ticketing tool that doesn't talk to the CRM. Salesforce built Agentforce to unify voice, chat, and email into one interface where supervisors see AI and human agents side by side. The AI transcribes phone calls in real time and surfaces customer sentiment before you pick up. That's useful when you're a three-person team juggling support and everything else.
Early users like Compass Working Capital wanted to scale without losing the personal touch that makes their service work. Ferguson wants to unify engagement channels so customers get consistent answers whether they call, email, or chat. If you run a service business and your support backlog keeps growing, this is the kind of tool that might actually save you time instead of adding another dashboard to ignore. Small Business Trends
HP's AI-powered printing systems are doing more than just printing. They predict maintenance needs, auto-format documents, and track usage patterns so IT teams stop firefighting printer jams and start doing actual IT work. HP's report says 87% of SMB knowledge workers see better workflow efficiency with smart printers, and 60% of IT pros spend less time troubleshooting. The productivity gain could be worth $58 billion across key markets, $25 billion in the U.S. alone. If your team spends more than an hour a week dealing with printer problems, this is worth looking at. Initial cost and training curve are the tradeoffs. Small Business Trends
Amazon's Trainium chip won over Anthropic, OpenAI, and Apple by cutting AI inference costs up to 50% compared to traditional cloud servers. Inference (running a model to generate responses) is the current performance bottleneck, and Trainium3 is breaking records by letting every chip talk to every other chip in a mesh configuration. AWS already has 1.4 million Trainium chips deployed, with Anthropic's Claude running on over 1 million Trainium2 chips. The OpenAI deal includes 2 gigawatts of Trainium capacity, and AWS claims Bedrock could eventually be as big as EC2. If you're running any AI workloads, AWS's Trainium pricing is worth comparing against Nvidia. TechCrunch
Nvidia's new orbital AI platform is designed for space data centers and autonomous spacecraft. Jensen Huang called it "space computing, the final frontier" at GTC 2026. This won't affect your business today, but it signals where AI infrastructure is heading as LEO satellites become strategic assets like ports and cables. SpaceX, Amazon, and Blue Origin are deploying thousands of satellites for orbital data centers. If you're in logistics, mapping, or any field that uses satellite data, this is the foundation for faster, smarter geospatial tools in the next 3-5 years. CNBC
Simon Willison's "profile this user" prompt is a great example of what LLMs can actually do when you feed them enough context. He grabs 1,000 Hacker News comments via the Algolia API, pastes them into Claude Opus 4.6, and gets a detailed profile: working style, technical interests, personality, debate patterns. The profile is "startlingly effective" and feels invasive, which is the point. If you're vetting potential hires, partners, or collaborators who have a public comment history, this technique surfaces patterns you'd miss reading manually. The ethical line is fuzzy here, so use it carefully. Simon Willison
Coding agents are fluent in Git, which means you don't need to memorize advanced commands anymore. Willison's guide on using Git with coding agents shows how to resolve merge conflicts, run git bisect to find bugs, search the reflog for lost code, and rewrite history without breaking things. The key insight: you can be more ambitious with Git because agents handle the tedious parts. If your team uses Git but avoids branching or rebasing because it's a pain, coding agents change that calculation. Simon Willison
AI tokens as compensation is now a real trend. Jensen Huang floated the idea at GTC that engineers should get roughly half their base salary again in AI tokens. His top engineers might burn through $250,000 a year in compute. VC Tomasz Tunguz says top-quartile software engineers already cost $375,000, and adding $100,000 in tokens brings the fully loaded cost to $475,000. The pitch is that tokens make engineers more productive, so it's an investment. The reality is that if your company is funding a second engineer's worth of compute on your behalf, you're expected to produce at twice the rate. Engineers at Meta and OpenAI are already competing on internal token consumption leaderboards. If you're hiring engineers or managing an engineering team, this is the new comp structure you'll start seeing in offers. TechCrunch
Delve, a Y Combinator compliance startup, is accused of falsely convincing hundreds of customers they were compliant with HIPAA and GDPR by generating fake evidence and rubber-stamping reports. An anonymous Substack post from a former client claims Delve produces fabricated board meetings, skips framework requirements, and places itself as both implementer and examiner. Delve denies the claims and says it's an automation platform, not an audit firm. If you're buying compliance-as-a-service from any vendor, this is a reminder to verify the auditor is independent and the evidence is real. TechCrunch
Nvidia's GTC keynote didn't impress Wall Street. The stock dropped during Jensen Huang's 2.5-hour speech despite him announcing $1 trillion in expected purchase orders for Blackwell and Vera Rubin chips by end of 2027. Analysts say the market hates uncertainty, and AI is moving so fast that investors don't know how to price it. Nvidia's revenue was up 73% year-over-year last quarter, and Amazon just ordered 1 million GPUs for AWS. The disconnect between Nvidia's performance and investor confidence is weird, but it doesn't change the fact that Nvidia chips run the rails for most AI businesses. TechCrunch
Indie game developers are rejecting generative AI. At GDC 2026, nearly every indie dev said they wouldn't use AI in their projects because it detracts from the human element. A GDC survey found 52% of respondents think generative AI is having a negative impact on the game industry, up from 30% in 2025 and 18% in 2024. Developers said AI-made games don't connect with audiences, look cheap, and raise legal problems around copyright and training data. The takeaway: if you're building creative tools or services, your customers may actively avoid AI-generated work, even if it's cheaper or faster. The Verge
Hachette pulled the horror novel "Shy Girl" over concerns that AI was used to generate the text. The author denies using AI and blames an editor she hired, but reviewers on GoodReads and YouTube had already flagged the book as likely AI-generated. U.S. publishers rarely do extensive editing on titles that were already self-published, which is how this slipped through. If you're in publishing, editing, or content production, AI-generated text is now a quality control issue you need to check for. TechCrunch
OpenAI is pulling back from building data centers. CEO Sam Altman acknowledged that data centers at this scale are hard, citing supply chain challenges, tight deadlines, and weather events that brought things down temporarily. The company is retreating from hefty spending plans and accepting its role as a purchaser of cloud capacity rather than a builder of infrastructure. The shift is about showing fiscal responsibility before an IPO later this year. For businesses using OpenAI, this means the company is prioritizing stability and profitability over expansion, which is probably good for long-term reliability. CNBC
Turn your handwriting into a custom font using Claude. Upload a sample of your handwriting, and Claude can generate a usable font file. This is a good example of a one-off AI task that saves you time if you need a personal touch on documents or branding materials. Inc.com
Cold outreach is dying, and AI-driven relationship management is replacing it. Instead of blasting generic emails, use AI tools to track when prospects engage with your content, identify buying signals, and personalize follow-up timing. This is the smarter way to generate high-quality leads without annoying people. Inc.com
The split between AI hype and AI skepticism is widening. Nvidia's stock drops despite strong fundamentals. Game developers reject generative AI even as vendors pitch it at conferences. Engineers are getting paid in tokens, but nobody's sure if that's compensation or just infrastructure cost. Meanwhile, practical tools like Salesforce Agentforce and HP's smart printers are quietly delivering ROI by solving boring problems.
The pattern: AI is useful when it automates specific workflows with clear before-and-after metrics. It's overhyped when vendors pitch vague productivity gains or creative magic. If you're deciding where to invest in AI, follow the money on what businesses are actually buying, not what CEOs are promising. Watch for tools that integrate with systems you already use and solve problems you already have.