Welcome, humans.

So apparently Microsoft wants to put words in people's mouths. Literally. Futurism reported that Microsoft filed a patent for AI that rewrites what a speaker says during a live video, then reshapes their lips and face so the edit looks natural.

Here's the setup. AI agents watch the live feed and check each word against rules you set or a viewer's profile. When a word trips a rule, the system swaps or deletes it, then patches the audio and video around the change so there's no visible jump. Replacement words can even come from an internet search, and the original feed may be stored as a reference copy of what was really said. The patent's examples include:

Live translation, so you see and hear a speaker in your own language.
Cutting filler words like um and ah, plus stutters.
Removing slang such as y'all, along with words the system flags as offensive or mumbled.

One caveat: a patent is a claim on an idea, not a product you can use, and Microsoft didn't respond to Futurism. But the worry is easy to picture. Everyone watching the same live event could get a slightly different version of what was said. Somewhere, a Texan just felt a disturbance in the Force.

Here’s what happened in AI today:

😸 Proton canceled AI-use re-interviews after staff backlash
📰 Anthropic disclosed $895.4B in commitments in IPO filing
📰 OpenAI launched a program to curb data center power at peak
🍪 DeepSeek's Harness now builds plugins from a prompt
🎓 Keep a weekly log of what AI did for you

…and a whole lot more that you can read about here.

😸 Proton Scrapped Plans to Re-Interview Staff About Their AI Use After Backlash

Picture your boss announcing that nearly everyone at the company will sit for another round of interviews. The topic is how much you use AI. Cool cool cool, no pressure.

That was the plan at Proton, the privacy company behind Proton Mail and Proton VPN, until CEO Andy Yen canceled it, 404 Media reported. The story rests on six current employees and audio from a roughly two-hour all-hands meeting.

Here's what happened:

Yen planned re-interviews with nearly the whole company, where employees would explain how they use AI.
Employees told 404 Media they believed people who used AI too little could be fired.
A 404 Media investigation the day before the cancellation showed Yen pushing AI hard across Proton, sometimes over security concerns.
Staff and Proton users pushed back, and Yen dropped the interviews.

The detail from that earlier investigation is the one we can't stop thinking about. In late August, an employee in Proton's security Slack channel suggested spending a few days on penetration testing before releasing new code. That means attacking your own software the way a hacker would, so you find the holes first. Yen reportedly brushed the idea off as a loser's mindset and told staff to stop handcrafting code and use AI more.

That lands hard at this company. People pay Proton to keep their private stuff private, so a CEO pushing speed over a security check hits the one thing customers care about most. Users have been vocal about exactly that, and some called for Yen to resign. Not the brand moment you plan for.

Here's our own interpretation. Counting AI use measures activity. It can't tell the engineer who used AI to ship clean work from the one who used it to look compliant. And once usage gets graded, the careful person who asks for a security test starts to look like the slow one.

What we don't know: the report doesn't say how the interviews would have worked, how many employees were involved beyond nearly everyone, or whether Proton is changing its approach to AI. Proton didn't provide a comment in the article.

Watch two things. At Proton, whether the company says anything about how it reviews code before release, since that's what its users keep asking. At your own job, what your manager counts as using AI: hours spent in the tool, or what you shipped with it.

FROM OUR PARTNERS

How Small Teams Win Big

Competition’s tough for startups, and everyone needs an edge. Nothing kills a deal faster than showing up without a SOC 2, but audit prep takes time most teams don’t have. Vanta gets you SOC 2-ready fast.

See for yourself:

20% shorter enterprise deal cycles
$1.5M in sales unlocked in 3 months
SOC 2 Type II readiness in 3 weeks

🎓 AI Skill of the Day: Keep a weekly log of what AI actually did for you

If the Proton story has you wondering what you'd say when your own boss asks how you use AI, here's a fix that takes about five minutes on a Friday.

The problem with answering "a lot" is that hours in a chat window say nothing about whether your work got better. A log of results does. Instead of counting how often you used AI, you record what you asked it to do, what you checked or changed yourself, and what came out the other end. That last part is what holds up in a review, because it puts your judgment right next to the tool's output.

Here's the workflow:

Gather the week's raw material: your task list, meeting notes, or a list of what you sent. Skip anything your employer says can't go into an AI tool.
Paste it into your AI tool of choice with the prompt below.
Fix any guessed numbers, then save the one-page result somewhere you'll find it.

Below are my task list and notes from this week: [paste here]

List every task where I used AI. For each one, give me:
- What I asked AI to do
- What I checked, edited, or decided by hand
- The result, with a number only if I gave you one (time saved, errors caught, things shipped)

If you would have to guess a number, don't. Ask me for the real one instead. Keep it to one page.

Future you, staring down review season, will be very grateful.

Have a specific skill you want to learn? Request it here.

FROM OUR PARTNERS

Analytics on Live Data Without Leaving Postgres

When analytics on Postgres slows down, most teams add a second database. Then they manage pipelines, sync lag, and drift forever. TimescaleDB extends Postgres instead. Analytics run on live data, in the database you already have. No pipeline. No migration. No new query language.

Get $1000 Credit

📰 Around the Horn

OpenAI’s test covers seven kinds of errands, including apartment hunting, government forms, grocery planning, and caring for a sick relative. Each task is graded against its own checklist (e.g., does the school fit the budget, schedule, and distance?) rather than a single right answer. Think “can it do your Saturday,” not “can it ace the SAT.”

Anthropic disclosed $895.4B in commitments for purchases and leases, about $70B due within a year, in an amended IPO filing.
OpenAI launched Grid Protectors, a program where new compute campuses cut power demand during grid stress, starting with up to 1GW in Effingham County, Georgia.
MIT CSAIL researchers got GPT-6 Astra to edit 0.016% of a small model's weights, lifting its math benchmark score from 7.9% to 48.1% without retraining it (not yet peer reviewed).
Oracle said it will raise $20B through $10B in stock and $10B in bonds for its AI buildout, and its shares fell nearly 6%.
Anthropic confirmed new Claude Code limits that steer Opus 5.5 away from cloning specific games, while users report reverse-engineering blocks Anthropic hasn't confirmed.
Elon Musk said Grok Bot costs too much to serve relative to revenue, as users hit weekly usage caps.

🍪 Treats to Try

Cedric joins Google Meet, Zoom, or Teams to present your slides, take notes, or stand in for you using a brief you write beforehand.
Xmind Agent builds a mind map from a sentence, file, or link, then helps you spot gaps and restructure it by chatting.
Hera turns a description into editable graphics and motion design on one canvas, for things like launch videos and animated infographics.
DeepSeek Harness now builds a plugin from a plain-English description with Creator Mode (the repo is still in prerelease).
UnleashX staffs your team with ready-made agents, like one that qualifies inbound sales leads by voice and chat, wired into tools such as Salesforce and HubSpot.
Hume's Audio Decision API listens to raw audio and answers questions you write, like whether a caller sounds frustrated (waitlist only for now).
LongCat-2.5 from Meituan reads images and handles long coding tasks, and now has API access.
Agent Beacon records what your coding agents do, from Claude Code to Cursor, into local logs you can view on a dashboard (open source).

UPCOMING EVENT: The Neuron IRL in San Francisco

The Neuron is going IRL on November 18 at 4:30 PM PT! Join us in San Francisco for drinks, bites, special guests, and a live recording of The Neuron podcast. Huge thanks to Slack and Alumni Ventures for making it happen.

It’s free, but space is limited. Save your spot!

🛠️ Tuesday Tool Tip: Stop fixing the same video three times

You finish editing a video for YouTube. Now you need versions for TikTok, Instagram, and LinkedIn, each with different dimensions. And suddenly, you're doing the same editing job all over again.

Here's a trick: Google Flow lets you build your own reusable editing tool with a plain-English instruction.

Open a Flow project, then select Tools → Create Tool.
Enter: "Build a video-resizing tool with 16:9, 9:16, and 1:1 presets. Include a preview and controls for adjusting the crop before exporting."
Generate the tool, review how it works, and save it for your next project.

You can also start with Google's existing Video Resizer and explore remixing it for your workflow.

The useful part? You can keep refining your mini-app as your editing needs change, without having to code it yourself.

See how Google's custom Tools work.

Note: Creating custom Tools requires an eligible paid Google AI subscription. Existing prebuilt Tools are available to all Flow users.