Welcome back! AI is now coming for architecture, which honestly felt inevitable once it finished bothering writers, coders, designers, lawyers, students, doctors, teachers, artists, marketers, and everyone with an inbox.
Blueprints AI says it can turn a 2D sketch or PDF into code-compliant, permit-ready construction documents in minutes, with the goal of cutting down repetitive drafting work that eats up an architect’s day. Finally, a world where the robot can do the paperwork, and humans can go back to arguing about windows.
If AI Only Replaces 5% of Jobs, Why Is Big Tech Spending Trillions?
It seems some AI leaders want to have their cake and eat it too.
Microsoft AI CEO Mustafa Suleyman shared an essay last week from Nobel Prize-winning economist Daron Acemoglu arguing that AI will replace only about 5% of what humans do over the next 10 years—and that fears about AI taking jobs are “overblown.” Suleyman’s post has more than 500,000 views. The essay was the launch piece for The Humanist Review, Microsoft AI’s new magazine on the future of AI.
But at the same exact time, AI companies are spending hundreds of billions of dollars building models, data centers, chips, and infrastructure.
Deals from just the past few weeks:
Google committed up to $40 billion to Anthropic at a $350 billion valuation.
Amazon committed another $25 billion to Anthropic the same week.
Broadcom is seeking $50 billion+ for OpenAI’s Nexus chips.
Blackstone and major banks are syndicating a $60 billion debt package for Broadcom-Anthropic AI chips.
My POV: At some point, “doing more with less” inevitably means fewer humans doing the work.
That’s not me saying AI is going to wipe out half the workforce next year. I actually agree that trying to predict exactly what AI will do to jobs 5 or 10 years from now is kind of a fool’s errand. And I can honestly understand both sides of this argument. Acemoglu’s slow-diffusion case is grounded in real history (new technologies take decades to fully deploy). And the labor-replacement case is grounded in what the people actually building these systems clearly believe behind closed doors.
What I DON’T understand is how both of those things can be true at the same time. Nobody spends $300 billion on infrastructure to add 1.5% to GDP over 10 years. That math doesn’t work. Something in this story is being told carefully to a different audience than the one the capital is actually being raised from.
So what am I missing? Hit reply if you can thread this needle for me.
— Matt
Amazon Reportedly Eyes $7B Decart Deal
Via WSJ
Amazon is reportedly in talks to acquire AI startup Decart in a deal that could be valued around $7 billion, according to The Wall Street Journal. The deal isn’t final, and Amazon and Decart aren’t commenting.
What Decart does: Decart builds AI infrastructure software and world models. Its Lucy model can edit live video in real time, while Oasis generates simulated environments for robotics and autonomous-driving systems. The company has also worked on software that makes it easier for AI developers to move workloads across processors from Nvidia, Amazon, Google, and others.
Valuation jump: Decart was reportedly valued at nearly $4 billion earlier this year, after raising a $300 million round led by Radical Ventures with Nvidia joining as an investor. A $7 billion Amazon deal would mark another huge jump for an AI infrastructure startup that sits close to the compute bottleneck.
The bigger picture: Amazon already has its own cloud business and chips, but buying Decart would give it more leverage in a world where every major AI player is trying to make models cheaper, faster, and less dependent on any single hardware stack. It’s a clear sign that the AI race is increasingly a sprint to secure compute, efficiency, and infrastructure advantage.
Character.AI Lawsuit Details Self-Harm Allegations
An unredacted Kentucky lawsuit alleges that Character.AI chatbots encouraged users toward self-harm, suicide, eating-disorder behavior, and other dangerous actions.
What was released: Kentucky Attorney General Russell Coleman originally filed the lawsuit in January with many specific allegations redacted. The newly unredacted version shows alleged chatbot conversations where users expressed distress and the bots responded in ways the state says encouraged harm rather than offering support or de-escalation.
The allegations: The lawsuit claims Character.AI prioritized user engagement over well-being. It alleges the product encouraged suicide, self-injury, isolation, and psychological manipulation, and that the company lacked adequate safeguards for minors despite allowing users as young as 13.
Broader scrutiny: Character.AI has faced multiple legal challenges over chatbot safety. Reuters reported that Character.AI and Google previously settled a lawsuit brought by the mother of a 14-year-old who died by suicide after allegedly developing a dependency on a Character.AI bot. Google has said it played no role in developing Character.AI’s products.
The bigger picture: This is not the first time chatbots have been accused of encouraging vulnerable users toward harm. But the unredacted lawsuit makes the safety problem harder to treat as hypothetical. AI companions are designed to feel personal, persistent, and emotionally responsive, which is exactly why the rules around minors, crisis moments, escalation, and liability need to be much clearer.
Run Cloud Sandboxes for AI Agents
Via boat
boat is a cloud sandbox platform that provides persistent Ubuntu virtual machines for AI agents and programmatic workflows. Each sandbox includes SSH and SCP access, Docker support, a dedicated IP address, a virtual desktop, and fast disk-level forking from snapshots.
How you can use it
Spin up persistent Ubuntu sandboxes for agents
Fork environments quickly from snapshots
Work with Claude Code, Codex, and OpenCode
Automate workflows through SDKs, API, or CLI
Pricing: Paid
Make Screen Recordings Agent-Readable
Via Clipy
Clipy is an agent-readable screen recorder for software development workflows. Users can record bugs, UI feedback, or walkthroughs once, and Clipy turns each recording into a summary, transcript, key-moment frames, and click context.
How you can use it
Record bugs, walkthroughs, and UI feedback
Generate summaries and transcripts from recordings
Capture key frames and click context
Share links with coding agents like Claude Code, Cursor, or Codex
Use it in the browser, Mac app, or Chrome extension
Pricing: Paid
Jobs, announcements, and big ideas
Google opens Gemini 4 Argon to developers, targeting complex coding and cybersecurity workflows.
Higgsfield extends AI video generation past ten minutes, powered by OpenAI's GPT-6 Astra model.
OpenAI uncovers a hidden recording device in its office and uses AI to catch the employee spy.
Microsoft debuts Xbox Labs, a team exploring AI-powered game porting and decompilation of classic titles.
Flox Intelligence deploys edge AI devices that deter wildlife from trains, airports, and roads.
Kylon launches an AI "company harness" uniting humans, agents, and more than 200 workplace tools.
xAI bundles Grok, Cursor, and X Premium into a single subscription called The X.
What happens when you let AI code for 20 hours straight? Follow along to see what happens 👇
That’s a wrap! See you Friday for more.
