Salesforce Pitches AI for Deportations, Microsoft Resurrects Failed Voice Assistant
Good Morning from San Francisco, Salesforce answered an RFP you didn't know existed. The pitch: use AI to
Good Morning from San Francisco,
Salesforce answered an RFP you didn't know existed. The pitch: use AI to nearly triple ICE's workforce through 'aggressive, high-yield marketing.' Sales reps celebrated with fire emojis. Ron Conway quit the foundation hours later, ending 25 years with Marc Benioff over what he called 'willful ignorance' about ICE raids.
Meanwhile, Microsoft wants you to talk to your PC. Again. Copilot Voice launches this week with the same promises Cortana made in 2015. Desktop voice assistants keep failing for one reason: talking to your computer in an office signals you're not working. Corporate IT will disable it by Monday.
Stay curious,
Marcus Schuler
Salesforce pitched Immigration and Customs Enforcement on using AI to recruit 10,000 officers and agents through "aggressive, high-yield marketing," according to internal documents obtained exclusively by The New York Times.
Hours later, Ron Conway quit the Salesforce Foundation board after 25 years, telling Marc Benioff his Trump support and "willful ignorance" about ICE raids broke their friendship.
The August 26 memo answered an ICE request showing how Salesforce could "nearly triple" the agency's workforce. When a sales exec posted it was "out the door," colleagues replied with fire emojis. The same week, Benioff—who hosted a 2016 Clinton fundraiser—called for National Guard troops in San Francisco and dined with Trump in England.
The substrate: U.S. government is now Salesforce's "largest customer" at billions annually. The company forecasts $60 billion revenue by 2030 while its filings warn government ties risk "reputational harm." That's the bind. Federal contracts fund growth; selling optimization to enforcement draws fire in a sanctuary city.
Why this matters:
• Optimization tools don't distinguish between selling bracelets and staffing deportations. Technical capability outpaces democratic oversight.
• Federal revenue dependence reshapes corporate politics. Billions in contracts mean absorbing policy backlash as cost of business.
Prompt:
This hyper-realistic selfie shows Vincent van Gogh painting "Starry Night" on a riverside lawn. Van Gogh stands at arm's length from the camera, his arm clearly extended toward it, as if holding a phone (though the phone is nowhere to be seen). He has flaming red, messy hair and a beard, and wears a pristine blue coat. In the background, an easel holds the canvas of "Starry Night" in progress, surrounded by plants and scattered rocks. Fireflies also cling to the plants. Above, a starry sky shimmers with vibrant patterns. The image captures the casual, iPhone-esque framing of the selfie, the cinematic night lighting, and the realistic texture of the skin.
Microsoft rolled out Copilot Voice, Vision, and Actions for all Windows 11 PCs on October 16—three days after Windows 10's end-of-support deadline.
Users can now say "Hey Copilot" to activate voice commands, grant the AI screen-reading permissions through Vision, and (eventually) let experimental agents manipulate local files through Actions.
The twist: these features work on any Windows 11 machine, not just the Copilot+ hardware Microsoft pushed all year. That Zoom scheduling integration is now the only Copilot+ exclusive.
Microsoft tried this in 2015 with Cortana—same promises about natural interaction, task automation, and transforming PC use. By 2021, under 1% of Windows 10 users touched it regularly. The company killed Cortana entirely. Voice assistants fail on desktop because talking to your computer in an office signals "not actually working." Phones normalized voice through mobility. Desktop use happens at a desk, with a keyboard, near other people.
Microsoft's bootstrapping voice-first interaction where ergonomics are worst. It's backwards. Enterprise IT won't enable Actions—experimental agents that edit files introduce cross-prompt injection risks without clear automation payoff.
Why this matters:
Lorikeet handles customer support requests across chat, email, voice, and SMS by connecting to your systems and solving problems from start to finish. It follows your workflows and actually takes action—processing refunds, rescheduling appointments, or updating orders—instead of just providing FAQ answers.
Tutorial:
URL: https://www.lorikeetcx.ai/
Create a 1:1 meeting template for my direct reports. Cover these topics:
Use a bulleted agenda. Include 2-3 questions for each topic that prompt real conversation, not yes/no answers.
Anthropic has launched Claude Skills, a new capability system for its AI models that offers a conceptually simpler approach compared to existing solutions. According to technology analyst Simon Willison, this new feature may prove more significant than Anthropic's Model Context Protocol (MCP), which faces challenges due to high token usage that limits its practical applications.
Wikimedia Foundation reports that Wikipedia is experiencing a significant decline in human traffic as more users obtain the site's information through AI chatbots rather than visiting the platform directly. The organization warns this trend could threaten Wikipedia's sustainability model, as reduced site visits may lead to fewer volunteer contributors to maintain and expand content, as well as decreased individual donations that support the platform's operations.
U.S. memory chip manufacturer Micron Technology plans to stop selling server chips to data centers in China following a 2023 ban that prohibited the use of Micron products in critical infrastructure. Despite withdrawing from the server market, the company will continue selling memory chips to China's automotive and smartphone sectors, according to sources familiar with the matter.
OpenAI announced Thursday that it has temporarily suspended Sora's capability to generate videos depicting Martin Luther King Jr., responding to a request from the civil rights leader's estate after users created inappropriate content. The pause comes after some users produced what the company described as "disrespectful depictions" of the iconic civil rights activist using the AI video generation tool.
OpenAI is collaborating with Broadcom to develop custom AI chips, expecting to reduce costs by 20% to 30% compared to Nvidia's offerings. The move represents a strategic shift for the company as it seeks alternatives to Nvidia's GPUs, which face significant supply backlogs and delivery delays.
Kayak has launched a new ChatGPT-powered chatbot feature on both its desktop and mobile websites, enabling users to ask natural language travel questions and receive assistance with comparing flights, hotels, and car rentals. The AI integration represents the travel search engine's effort to streamline the trip research and booking process by allowing customers to interact conversationally with the platform rather than relying solely on traditional search filters.
Reddit announced Thursday that it is expanding its Google Gemini-powered search experience to five new languages: French, German, Spanish, Italian, and Portuguese. This multilingual expansion represents Reddit's effort to make its AI-enhanced search functionality accessible to a broader international user base beyond English-speaking markets.
The FBI's investigation into the coerced suicide livestream of a 13-year-old boy found dead in a Gig Harbor, Washington parking lot has resulted in the arrest of a German member of the notorious online predator group 764. The case highlights the dangerous reach of international online predators who manipulate vulnerable minors into extreme acts of self-harm through digital platforms.
Three labor unions have filed a federal lawsuit in New York to prevent the Trump administration from conducting social media surveillance of visa holders, targeting posts the government considers hostile or threatening. The legal challenge, filed Thursday, represents the latest effort to block the administration's expanded screening policies that monitor foreign workers' online activities as part of visa processing and renewal procedures.
A hacker group calling itself "Scattered LAPSUS$ Hunters" has published the alleged personal information of hundreds of federal law enforcement officials, including names, phone numbers, and addresses of personnel from the Department of Homeland Security, Immigration and Customs Enforcement, FBI, and Department of Justice. The mass doxxing incident represents a significant security breach targeting U.S. federal agencies and puts the personal safety of government employees at potential risk.
Meta announced plans to implement new Instagram safety features in 2026 that will give parents greater control over their teenagers' interactions with AI characters on the platform. The upcoming tools will allow parents to completely block their teens from chatting with AI characters and will provide parents with insights derived from their teens' chat conversations.
Samsung Electronics plans to showcase its highly-anticipated trifold smartphone at this month's APEC summit, marking the company's entry into the emerging tri-fold device market. The prototype will be displayed under glass protection, with attendees unable to physically interact with the device during the demonstration to world leaders and global dignitaries.
EssilorLuxottica's stock jumped over 10% following the company's announcement of impressive third-quarter results, with revenue climbing 11.7% year-over-year to €6.9 billion. The eyewear giant credited its partnership with Meta, particularly the Ray-Ban Meta smart glasses, for contributing 4 percentage points to the quarter's growth, highlighting the strong market demand for wearable technology products.
Wikipedia updated bot detection in May, reclassified six months of traffic data, and found an 8% year-over-year decline in human visitors.
While AI models trained on its articles now answer questions directly without requiring anyone to visit the source.
From Wikipedia's angle: We're watching value extraction without reciprocity. Bot bandwidth jumped 50% since January; scrapers consume 65% of expensive infrastructure traffic. Donations and volunteers depend on visits that AI answers eliminate.
From platforms' view: Wikipedia's open licensing makes it ideal training data. Why drive traffic back when we can synthesize answers directly?
From users' perspective: Pew found only 1% click through from AI summaries. Faster answers win.
The bind: Wikipedia's openness made it essential. That same openness now breaks its funding model—attribution without traffic severs the volunteer-donor flywheel.
Why this matters:
Aboon wants to turn 401(k) setup from a months-long paper nightmare into a days-long software flow. The New York startup raised $17.5M to build an AI-powered TPA that runs through financial advisors instead of fighting them.
The Founders
Nick Gavronsky (CEO) and Amy Ouellette (COO) launched in 2023. Gavronsky's dad—a South African immigrant running a small business—couldn't figure out retirement plans, which lit the fuse. Gavronsky brings Betterment product chops and startup reps (Trade Coffee, Welcome). Ouellette ran retirement ops at Betterment for Business and Vestwell, so she knows where the compliance bodies are buried. No employee count disclosed yet. 🗽
The Product
Aboon sits in the TPA lane—plan design, compliance testing, annual filings—and plugs into recordkeepers like Capital Group, Empower, and Hancock. Advisors punch in employer details, the AI spits out plan options with tax projections, and onboarding happens in days instead of the usual 10–12 weeks. Human consultants back the software when things get weird. Speed is the pitch. The claim: proposals in minutes, live plans fast enough to beat your competitors' paperwork.
The Competition
Vestwell, Guideline, Human Interest, and Ascensus all want the same small-business wallet. Most bundle recordkeeping with admin or sell direct to employers. Aboon stays narrow—TPA only—and runs exclusively through advisors. That modularity could be a moat or a margin trap. Edward Jones signed on early, giving access to tens of thousands of reps.
Financing
Bain Capital Ventures led the $17.5M seed. Edward Jones Ventures, Altai, Runyon, Outpost, and EJF joined. Valuation undisclosed. The Edward Jones tie matters more than the dollars—it's distribution on day one.
The Future ⭐⭐⭐⭐
If Aboon ships what it promises, advisors will use it everywhere. The wedge is real: state mandates push adoption, SECURE 2.0 sweetens the ROI, and nobody loves the legacy process. Risk? Edge cases blow up timelines, and bundled competitors have more marketing muscle. Execution decides everything.
Get the 5-minute Silicon Valley AI briefing, every weekday morning — free.