Meta Launches Muse AI Agent to Book Travel and Pay Bills
ALSO: Google buys Spirit’s data, AI now writes 42% of code
Welcome to Guru99 AI Report!
Top Story: Would you trust an AI to book your flights, answer your emails, and spend your money? Meta is betting you will. This week we open up Muse, watch AI turn coders into code reviewers, and follow Google as it buys a bankrupt airline’s data to train AI. Let’s dig in.
🤖 Meta’s Muse: An AI Agent That Runs Your Errands
Brief Buzz:
Meta has released Muse, an AI agent aimed at users in the United States. Hand it the keys and it will book your travel, send emails, fill out forms, and make purchases on your behalf. The catch: you have to give it access to your email, your calendar, and your payment accounts.
- It’s free for up to 100 million tokens a week, after which it costs $20 or $100 a month depending on usage.
- Every user gets their own isolated virtual machine, with a separate Sentinel agent guarding sensitive actions such as payments.
- Meta’s own internal tests reportedly surfaced reliability failures involving private iCloud photos before launch.
- The launch lands weeks after Meta reached an $18 billion agreement over social media harms to minors.
- Meta says Muse never sees your passwords and shares nothing with its advertising systems.
💡 Why Should You Care?
Meta’s biggest bet on consumer AI asks the least-trusted of the big tech companies for its most sweeping access yet. The convenience is real — and so are the risks if it slips.
🤖 AI Writes the Code. Reviewing It Is the New Job
Brief Buzz:
AI can now write huge amounts of software in minutes — but a human still has to check it. That flood of messy, error-prone code is reshaping engineering jobs, turning coders into reviewers and raising a hard question about how new engineers will ever learn.
- AI-written code looks neat but can hide security vulnerabilities and subtle bugs, pushing review workloads to record highs.
- A Sonar survey of more than 1,100 developers found AI writes about 42% of committed code — yet 96% don’t fully trust it.
- Teams are adapting: checking the plan before the AI executes, deploying AI review agents, and letting the AI make the riskiest changes itself.
- Investors see the opening — CodeRabbit raised $143 million at a $1.5 billion valuation to automate reviews.
- At Synthesia, review requests jumped 120% year over year, with 95% of them involving AI-generated code.
💡 Why Should You Care?
If the AI writes the code, how do juniors learn to judge it? Fewer entry-level roles today could mean fewer senior engineers tomorrow.
✈️ Google Buys a Bankrupt Airline’s Data to Train AI
Brief Buzz:
When an airline goes under, you expect the leftover assets to be aircraft and airport slots. Spirit Airlines had a different one: its own operating history. Google just won it at auction, betting that a company’s internal data can teach AI how real businesses actually run.
- Google paid $10 million for the bankrupt airline’s business data and software code, per an August court filing.
- Its bid topped the $7.5 million offered by Mercor, an AI-data startup — proof that internal records now carry standalone market value.
- The trove is Spirit’s operational history: internal emails, spreadsheets, calendars, and financial and operational records.
- Google says it won’t receive personal data — a third party must strip identifiers first — and Spirit’s customer and loyalty profiles are excluded.
- A judge still has to approve the deal; the hearing is set for September 9 after the flight attendants’ union objected to employee data.
💡 Why Should You Care?
A sign of a bigger fight: as AI firms chase private data, courts and unions now have to decide who owns the digital remains of a dead company — including your old work emails.
🌊 OpenAI Says AI Solved Navier-Stokes — Credit Fight Erupts
Brief Buzz:
OpenAI says a swarm of AI agents, run by an unreleased model, has cracked the Navier-Stokes problem — a mathematical mystery that stood for 90 years and carries a $1 million prize. But the claim is now tangled in a dispute over who deserves the credit.
- Roughly 10,000 agents ran in parallel for about 88 hours, swapping around 2.7 million messages and 130 billion output tokens — costing OpenAI millions.
- An unreleased model found the proof; GPT-6 Astra then took 17 hours to formalize and check it in Lean, a proof verifier.
- A fluid’s speed can, in theory, become infinite in finite time even from a smooth start — though OpenAI won’t claim the $1M prize.
- Independent mathematicians call it remarkable but say it still needs scrutiny; only one Millennium Prize Problem had been solved before.
- Tristan Buckmaster of NYU and Levent Alpöge of Anthropic (in a personal capacity) reached a related but unverified result; Buckmaster says OpenAI pressured them, which OpenAI denies.
💡 Why Should You Care?
If AI can crack problems that stumped humans for a century, medical and materials breakthroughs could come faster. But who then owns discoveries made with AI’s help?
🛡️ AI Can’t Outrun Cybersecurity Basics, Officials Warn
Brief Buzz:
At a major federal cybersecurity summit, top government and industry officials had a blunt message: AI grabs the headlines, but careless fundamentals still cause the most breaches. The advice — get the basics right before chasing the shiny new tools.
- Core weaknesses — weak multi-factor authentication, slow patching, and poor asset visibility — still drive the most breaches, leaders at the Billington Cybersecurity Summit warned.
- The FBI is urging organizations to fix the fundamentals through its Operation Winter Shield campaign, built around 10 high-impact controls drawn from real cases.
- The NSA’s David Imbordino put it plainly: “The fundamentals are no longer dull, and AI can’t catch up to the fundamentals.”
- On defense, Canada’s cyber chief said AI has made analysts 10 to 100 times more effective — a genuine force multiplier.
- Attackers benefit too: Cisco’s Talos reports state-sponsored hackers using AI-generated video to pass job interviews at large companies.
💡 Why Should You Care?
Good news for most organizations: you don’t need cutting-edge AI to stay safe. Disciplined patching, phishing-resistant multi-factor authentication, and knowing what’s on your network still stop most attacks.
🤖 Bill Gates Pitches ‘Human Reserved’ Jobs and a Robot Tax
Brief Buzz:
Bill Gates has sketched a plan for the shock AI is about to deliver. In a recent essay he argues some jobs should stay the exclusive domain of humans — and proposes taxing AI and robots. The goal: slow automation and use the revenue to fund retraining before the disruption hits.
- Even where AI could do the work, the roles reserved for humans would stay with people — Gates names childcare and jury service.
- He puts the ceiling of jobs that could plausibly be protected at 40% — the most, he says, that he can imagine.
- He’d tax AI tokens (how AI models charge for their services) and robots, since today’s tax code rewards swapping workers for machines.
- Customer support, software engineering, and paralegal work shrink first; by the end of the decade, construction and hospitality decline too.
- His bigger worry: entry-level jobs for young workers are already scarce, with no solid plan for the shift.
💡 Why Should You Care?
Even if these ideas never become law, they show how seriously insiders take AI’s threat to entry-level work — and hint at how governments might respond.
Hey! I’m Krishna Rungta
Founder of Guru99.com, Editor-in-chief & Technology Expert
Was this email forwarded to you? Sign up for free here.
