November 15, 2026 | Listen online | Read online

Welcome, humans.

So Princeton researchers tried teaching an AI to learn its first words the way a baby does. They played it recordings of everyday speech from a baby’s home in the order she heard it. By its second simulated birthday, the AI still couldn’t recognize a single word.

Then they borrowed an idea from sleep. At the end of each simulated day, the AI replayed a little of what it had heard over the past few days. That worked. By the same age, it could recognize thousands of words. The same approach worked with recordings from seven other homes. But, does the AI talk like a toddler??

The experiment gives researchers a way to test how replay might help babies take that first step, and how AI can keep learning without forgetting what it heard before.

Next benchmark: toddler selective hearing. They can hear “ice cream” from upstairs. They can’t hear “clean up” from right next to you.

Here’s what happened in AI today:

😺 Oakland workers reached a tentative deal on AI job protections.
📰 ChatGPT Wallet lets your agent pay after you approve.
📰 LangChain reportedly raised $300M at a $3B valuation.
🍪 Mister Blocks suggests builds from a photo of your LEGO bricks.
🎓 Compare vendor quotes in the new Claude Sheets.

Advertise to 700K readers of The Neuron here!

😸 Oakland city workers vote on AI job protections

Oakland city workers are about to vote on a contract that spells out what happens if AI takes over their jobs. Under the proposed deal, the city would have to retrain or reassign workers displaced by AI.

Say Oakland brings in a tool that cuts the time it takes to process permit applications. The city could use those hours to clear a backlog, but it could also decide it needs fewer people processing permits. The tool saves time either way, but what happens to the people doing the work is a choice the city makes.

For Julian Ware, who works in the city’s IT department, losing jobs was only one concern. He said the union also worried workers would be held responsible for what AI produced, especially if it left them with more work. For once, I’d like the last slide of an AI demo to say “Here’s what happens to your paycheck.”

The city and IFPTE Local 21 reached a tentative three-year deal on October 31, and the union called off its planned November 5 strike. The deal includes a 9% raise over three years.

Other public workers have secured a say in what happens before the software shows up, too.

At Berkeley Unified, the same union secured 60 days’ notice before the district introduces AI, a chance to discuss the rollout, and training if the district goes ahead. The contract also bars using AI to evaluate or discipline those workers.
In Columbus, Ohio, the transit agency needs the union’s consent to put self-driving vehicles into service. The contract also protects bus drivers and mechanics from layoffs or pay cuts caused by new technology.
In New York, a state law protects covered public workers from losing their jobs or having their regular hours, pay, or benefits cut because their employer uses AI.

And what to do with the hours AI saves? After a yearlong ChatGPT pilot, Pennsylvania said it would look for ways to put that time back into public services. The state also brought union members and managers together to talk through where AI should be used.

I’d watch what Oakland offers the first worker whose job changes. Is there a role they can move into at the same pay? And do they get time on the clock to learn it?

You can see the same fight outside city hall. At Microsoft-owned Blizzard, workers won the right to bargain over how the company uses AI on the job. In Hollywood, producers now have to bargain with the actors’ union before using AI-generated performers.

As these tools take on more of the work, expect more unions to ask what happens to the people doing it now, while there’s still time to put an answer in the contract. An interesting contrast at the same time, specifically with government workers. The governments also have a responsibility to taxpayers to be financially responsible, and they aren’t represented in the votes, so there could eventually be pushback from directions..

🎓 AI Skill of the Day: Compare two quotes in Claude Sheets

Anthropic just launched Claude Sheets in beta on paid plans. Ask for a spreadsheet in chat and Claude builds one you can edit right there, with formulas you can click into and check.

Say you’re comparing software plans, printing quotes, or contractor estimates. One option has a lower base fee. The other charges less per seat, copy, or hour. Put both in a sheet so you can change the quantity once and see which deal is cheaper.

Paste this into a Claude chat:

Build a Claude Sheet to compare these two quotes:

[paste quotes here]

Use one editable cell for the quantity I expect to buy: [number and unit, like 40 seats]. Give each quote its own row with an editable base fee and price per unit. Use formulas to multiply the quantity by each unit price, add the base fee, and highlight the lower total. If costs recur, compare them over the same period. Flag any fees or minimums that still need to be included.

Put your quotes side by side and change the quantity to see both totals update. If you’re deciding how many software seats to buy or how large an order to place, you can see whether the same quote stays cheaper as your needs change.

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

FROM OUR PARTNERS

Coding agents work quickly. Safely shipping their output depends on what happens after the commit.

Join DZone and Harness on December 8, 2026, at 1:00 PM ET for How DevOps and Coding Agents Work Together.

See how codebase and delivery context, builds, tests, and security scans fit into a feedback loop that helps agents respond to failures while keeping humans involved.

Save Your Spot

🍪 Treats to Try

Mister Blocks turns a photo of your loose LEGO bricks into build ideas using the pieces you have, with 3D instructions you can rotate.
Nameshot gives your Mac screenshots descriptive filenames based on what’s in them, with all the processing done on your device.
Command L lets you edit text directly on your website, then uses Codex or Claude Code to update the code.
Trainn MCP turns a screen recording into a narrated tutorial you can edit through Claude or ChatGPT.
Expo Simulators gives your coding agent a virtual iPhone or Android device to test your app and sends back a recording you can watch.

📰 Around the Horn

OpenAI began rolling out ChatGPT Wallet, which lets your AI agent securely pay for things online after you approve a purchase.
MIT researchers used one AI to patch another like developers patch software, lifting the smaller model’s score from 8% to 48% on a math test without retraining it.
Notion warned companies that access keys they’d issued were exposed in a breach at its AI provider, Fireworks.
LangChain, which helps developers build AI agents, reportedly raised $300M in a deal that valued the company at $3B.
Harvard researchers published a guide to when AI might help or hinder learning at different ages.

☀️ Sunday Special

Top 5 Tools of the Week

Gemini 4 Hydrogen is Google’s faster, cheaper model for everyday coding and tasks you hand off to an agent.
Adobe Creative Agent handles multi-step edits in apps like Illustrator and Premiere from plain-English instructions.
Yogurt records calls and writes meeting notes on your computer without sending a bot into the call.
Gemini in Chrome opens beside the page you’re reading so you can get a summary or ask questions without switching tabs.
AWS Emulator lets you test code built for Amazon’s cloud on your laptop without an AWS account.