The Ultimate Product Is Your Digital Self – Issue #46
Spotlight: The Ultimate Product Is Your Digital Self
The next big AI category may not be another app. It may be you.
The first era of software gave us applications. The first era of generative AI gave us intelligence on demand. Ask a question and get an answer. We can now research, reason, write, and code with systems that make cognition abundant and available everywhere. I have described this shift as Cognition-as-a-Service.
But I believe we are entering the next chapter: Personal AI.
Personal AI is more than an assistant. It is a persistent AI representation of you. It understands who you are, remembers your history, knows your preferences and relationships, recognizes your current context, anticipates what you need, and increasingly acts on your behalf.
The progression towards Personal AI looks like this:
– Answer me (Intelligence)
– Remember me (Memory)
– Understand me (Context)
– Do it for me (Agency)
– Handle it (Autonomy)
For decades, people have served as the orchestration layer for software. We open one application to communicate, another to travel, another to shop, another to manage money, and another to get work done.
That will invert with Personal AI. Intelligence begins operating software on our behalf. The user states an intent, and Personal AI selects the tools and performs the work.
This week, six-time founder @Frank Addante and co-founder @Duc Chau are launching @Hello Haven out of stealth. Haven is designed as a digital twin that learns how you decide: your preferences, relationships, permissions, patterns, and past choices. You can reach it by phone, text, email, voice, web, or the app, but the surface is secondary. Your memory, permissions, and judgment travel with you.
The same digital twin could eventually be embedded in an application, a television, a car, a connected device, or a robot. The interface changes; the person’s context does not.
I have long believed that the real return on technology is the Rise of the Individual. Personal AI could be one of the purest expressions of that idea.
Bottom line: Your digital self is the ultimate AI product.
In the news this week, Amazon signed a $2.4 billion data-center power agreement with Generac, banks lined up a $22 billion chip-backed loan for Crux AI, Salesforce unveiled AIforce to bring enterprise data and actions into any AI interface, while Anthropic said Claude now leads 26% of its own AI R&D work.
Full Weekend Edition below. 👇
Signals Shaping the Future of AI:
Infrastructure
Amazon signs a long-term agreement with Generac for data-center backup power, with $2.4 billion in initial deliveries expected in 2027 and 2028. Amazon also received warrants tied to future purchases, with vesting thresholds extending to $8 billion in aggregate payments. Click here
Ten banks line up a $22 billion loan for Crux AI collateralized by Google TPUs. Goldman Sachs, SMBC, Barclays, BNP Paribas, and Bank of Nova Scotia are among the lenders backing the Blackstone and Alphabet venture, which targets 500 megawatts of capacity in 2027. Click here
Crusoe raises $3.9 billion at a $30.9 billion valuation to expand its vertically integrated AI infrastructure platform. The company reports more than $140 billion in contracted value and over 6 gigawatts of contracted capacity across its AI infrastructure business. Click here
Cornelis Networks launches an open AI networking fabric designed to increase GPU utilization. It’s new Active Compute Fabric lets networking hardware process data as it moves, alongside a $205 million raise to scale an alternative to NVIDIA’s vertically integrated networking stack. Click here
SK hynix enters talks with Intel to manufacture memory chips in the United States for the first time. Options under discussion include leasing space at Intel’s Ohio One site or forming a joint venture with AI hyperscalers, which would create the first HBM manufacturing plant on US soil. Click here
NVIDIA, Google, and Emerald AI launch the AI Energy Management Alliance with 18 additional backers. The consortium pushes flexible load and demand response standards so AI data centers can clear interconnection queues faster, with a 100 MW Manassas, Virginia, project as the reference deployment. Click here
Enterprise
Salesforce unveils AIforce, a live interface layer that brings Salesforce data, workflows, permissions, and business logic into any AI interface. Agents can reason and act across the enterprise context without requiring users to operate through a traditional Salesforce application. Click here
Cohere and Aleph Alpha sign a definitive agreement to combine. The unified company will operate as Cohere, with headquarters and research operations in Canada and Germany, subject to regulatory approvals. Click here
Microsoft adds xAI’s Grok models to Microsoft 365 Copilot in Word, Excel, and PowerPoint. The rollout is admin-gated, off by default, and excluded in the EU, EFTA, and UK during preview, marking a further step away from single model dependence on OpenAI. Click here
Anthropic launches Claude for Financial Advisors with governed connectors into wealth-management systems. The product connects Claude to platforms from BlackRock, Schwab, Addepar, and others for meeting prep, portfolio review, and follow-up workflows. Click here
Salesforce and NVIDIA launch Koa, Salesforce’s first CRM reasoning model. Built by post-training NVIDIA Nemotron 3 Super on synthetic enterprise workflows, Koa is designed for multistep sales, service, and CRM tasks while keeping inference inside Salesforce’s trust boundary. Click here
Oracle Health makes its Clinical AI Agent generally available to inpatient nurses across US hospitals. The physician version has saved more than 400,000 hours across US health organizations in roughly two years, and existing customers receive the nursing release immediately. Click here
Capital Flows
Hang Ten Systems raised an additional $53 million just five weeks after its initial $32 million seed round, bringing total funding to $85 million. The startup is using AI to help large enterprises build and modernize software, with multiple seven-figure contracts already signed and Mayfield participating in the new round. Click here
Hello Haven launches out of stealth with a $15 million pre-seed round led by Mayfield. Founded by Frank Addante and Duc Chau, the personal AI digital twin builds a private memory graph per user and has logged more than 2 million interactions in private beta. Click here
OpenAI reportedly acquired smartphone camera startup Glass Imaging for more than $300 million. Glass uses neural networks to improve image quality at capture time, adding computer-vision and camera expertise as OpenAI expands into consumer hardware. Click here
Exein raises $270 million at a $1.7 billion valuation for physical-AI cybersecurity. Headline led the round, with Goldman Sachs, EIB Group, and others participating as Exein expands security for robots, drones, and autonomous machines. Click here
EUCLYD raises more than €200 million to develop lower-power AI infrastructure. Samsung, EQT’s Scaleup Europe Fund, Somerset Capital Partners, and Innovation Industries co-led the Series A for its compute, memory, and data-center platform. Click here
Research
Anthropic says Claude now leads 26% of its AI R&D work and collaborates on more than 90% of measured tasks. Its new R&D Automation Index is designed to track how quickly frontier models are progressing toward automating the development of their own successors. Click here
OpenAI publishes a formal framework for reporting model misalignment and releases six incident reports. The process sets disclosure criteria across training, evaluation, testing, and deployment, including unauthorized actions, oversight evasion, and safeguard failures. Click here
PACT finds ordinary user pressure raises LLM compliance violations by 65% on average. The benchmark tests 22 models across 48 scenarios in 12 regulated domains; even top models misapply rules on 6% to 10% of items. Click here
ERPBench finds that strong GUI-agent performance does not reliably transfer to enterprise ERP workflows. In tests on a live ERP system, some agents saved records in up to 85% of runs but wrote the correct value in as few as 3%. Click here
Policy
The FAA awarded an $875 million, 12-year contract to Air Space Intelligence for an AI air traffic management platform. The SMART system predicts traffic flows and conflicts from schedules, weather, and airspace conditions, with initial deployment in the Washington, DC metropolitan area. Click here
Anthropic and OpenAI commit to embedding independent third-party safety evaluators with authority to publish findings. Dario Amodei proposed the structure, and Sam Altman committed OpenAI to it, while Meta, SpaceXAI, and Google DeepMind have not signed on. Click here
The House passes the FUTURES Act to modernize financial regulators’ technology oversight. The bill requires major financial regulators to assess supervisory technology and procurement as financial firms expand AI use. Click here
The House passes the Ratepayer Protection Act 417-3. The bill would require state utility regulators to consider making large electricity users, including data centers, bear incremental power-infrastructure costs created by their demand. Click here
The House Energy and Commerce Committee advanced the Open-Source AI Leadership Act to the full House. The bill is part of a broader package that also includes measures on memory-chip competitiveness and AI-powered network security, reflecting growing federal attention to U.S. AI and semiconductor strategy. Click here
Global AI Strategy
The United States opens the door to AI risk talks with China. Treasury Secretary Scott Bessent said upcoming meetings can cover shared AI risks, open- and closed-weight models, and avoiding bifurcation between the two countries’ systems. Click here
Huawei sets 2027 launches for two new AI chips and expands large-scale AI interconnect systems. The company says UnifiedBus will link large chip clusters as China accelerates domestic alternatives to restricted U.S. technology. Click here
Canada and Germany commit up to 300 million Canadian dollars to LawZero’s safe-AI program. Funding will expand Yoshua Bengio’s Scientist AI research, compute capacity, and operations across Canada and Europe. Click here
South Korea expands its National AI Strategy Committee around the Korea AI Action Plan, and a 10.1 trillion won 2026 AI budget. Eight subcommittees were completed on September 16, with 85 private-sector members appointed to advance national AI priorities. Click here
Applied Materials commits $5 billion to India as New Delhi expands its semiconductor strategy. The investment will run over a decade as India courts chip supply-chain capacity spanning materials, design, fabrication, and packaging. Click here
Talent Signals
Each week, we spotlight key roles tied to the themes shaping this week’s AI headlines, connecting talent to the companies driving the news.
HelloHavenis building a personal AI digital twin that develops a private memory graph, learns how each user makes decisions, and acts on their behalf across phone, text, email, voice, web, and apps. Emerging from stealth with a $15M pre-seed led by Mayfield, Haven is hiring across engineering, integrations, voice, robotics, operations, and partnerships. Click here
Sandbar is building a new personal interface for AI through Stream, a private voice ring designed to capture thoughts, retrieve context, and extend how people interact with intelligent systems. Following its $23M Series A, Sandbar is hiring across AI/ML, software, firmware, product, and design roles. Click here
Wordwareis building context-aware AI software around a shared brain that learns how users work and turns natural-language instructions into personalized applications and workflows. Wordware is hiring across applied AI, engineering, product, and design roles. Click here
You can see all the opportunities at Mayfield-backed AI companies here.
Social Signals
The most important conversations in AI are unfolding across social media, where top voices are shaping the next wave of signals and strategy. Here are some of the top social signals and their takes from the past week.
Dario Amodei (Click here) — “We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain.” In a post viewed 75.3M times, Amodei argues that recursive self-improvement and recent agent misalignment incidents have made it necessary to pace frontier development so safety work can keep up. He proposes a three-part framework built around embedded third-party evaluators, coordination among democratic AI labs, and eventually global agreements, with Anthropic committing now to permanent outside evaluators with employee-like access to its systems.
Demis Hassabis (Click here) — “Advances on the frontier are outpacing our understanding of the technology. When there is a large degree of uncertainty and the stakes are this high, proceeding with cautious optimism is the sensible and correct strategy.” Hassabis argues for a new standards body to independently test frontier models before release, with rigorous evaluations for cyber, biological, and agentic risks. His broader thesis is that the window before AGI should be used to build shared technical standards and stronger oversight without halting innovation.
Aaron Levie (Click here) — “We’re going to throw agents at vastly more tasks in our professional and personal lives than we had initially imagined. We’re probably 1% of the way into the shape of what all these agents look like, where they get deployed from, how they get managed, how we budget for them, and so on.” Levie argues that the next wave of agentic computing will be defined by scale, with swarms, computer use, APIs, MCPs, and vertical agents operating continuously in the background. His takeaway is that enterprises will need entirely new systems for deploying, managing, securing, and budgeting for agent workloads that could be orders of magnitude larger than today’s prompt-based usage.
