Skip to content
AI Newsletter

Your next assistant might just be a text thread

ContentBuffer
ContentBuffer
November 22, 2026·~6 min read
ContentBuffer

ContentBuffer

Welcome, Tech Leaders.

Texting an AI assistant is no longer a novelty. A growing crop of agents, including Instinct, Caddy, Comma, Fambot, Haven, Iris, Martin, and Maverick, now lives where you already chat. Instinct raised $1 billion at a $10 billion valuation and started giving each assistant its own email address. Why should you care? These tools act instead of just answering. Caddy turns scattered phone messages into tasks inside iMessage or RCS. Comma carries tasks to completion and checks its own work. Martin handles calendars, email, reminders, and notes across SMS, WhatsApp, Slack, and email. If you're the person who coordinates everyone else's schedule, that's the job these agents are going after. The numbers matter. Ask Proxy's personal tier costs $1 per day, while business pricing starts at $15 per employee per month for large companies. Comma is free to start and open source. Fambot connects to Gmail, Google Calendar, Outlook, and WhatsApp on iOS, Android, and the web. Haven is free in beta. The open question is whether people will trust an agent with real actions in their inbox.

Shortical just premiered "Inevitable" in Los Angeles, and it's the first microdrama to use a real person's likeness in AI-generated content. Israeli star Aki Avni is the face, and he was paid in line with industry rates for a comparable non-AI acting role. That's the detail to notice: the deal is structured like a normal acting gig, not a novelty. Why should you care? Because the same pipeline that produces this show is now being pitched as a template. Shortical is offering face-licensing deals to creators, and CEO Guy Shimoni argues AI lets small teams produce movie-scale work, from dragons to werewolves. If you're building generative video tools or consent and rights management systems, this is the kind of commercial case that will shape your requirements. The numbers are thin on purpose. Shortical hasn't disclosed what it pays for face licenses, and it hasn't signed any contracts yet. Its model runs on the first 7-10 episodes being free, with a monthly subscription after that. In China, where microdramas started, firms already pay people between $15 and $700 for their likeness, which gives you a rough sense of the range. Watch whether Shortical actually closes its first face-licensing deals after promoting the program around the Inevitable launch. Let's dive deeper…

In today's ContentBuffer update:

  • AI Vision Models Match Brain Activity But Use Different Internal Logic

  • AWS Marketplace Lets AI Agents Buy Software Through x402 Payments

  • Google Duplex Evolves Into a Gemini-Built Agent Handling Real Phone Calls

  • Riemann Hypothesis Solved and Released With Lean-Verified Proof

  • 5 new AI tools & 5 new AI jobs

  • More tech news

Latest Development

Ai Agents

Instinct Raises $1B at $10B Valuation as Text-Based AI Agents Multiply

Image source: techcrunch.com

Summary: Instinct raised $1B at a $10B valuation, and a crowd of text-based AI agents is racing to run your calendar and inbox. Pricing runs from free and open source to $15 per employee monthly. If you're the team's scheduler, start testing now.

Key Points:

  • Instinct raised $1 billion, valuing the company at $10 billion, and it can take actions on users' behalf rather than just answering questions.

  • Ask Proxy's personal version costs $1 per day, while business pricing starts at $15 per employee per month for large companies.

  • Caddy lives inside iMessage for iPhone users and RCS for Android users, turning scattered phone messages into actionable tasks automatically.

  • Comma is free to start and open source, designed to manage work and personal tasks through completion while checking its own output.

  • Fambot connects to Gmail, Google Calendar, Outlook, and WhatsApp, and runs on iOS, Android, and the web as a family chief of staff.

Why it matters: If you coordinate calendars for a small team, Martin and Instinct now automate the reminder and follow-up work that usually lands on you. Comma is free and open source, so a dev team can self-host it and inspect exactly what actions it takes before connecting real accounts. Give any of these agents read-only access to your inbox first, because a $10B valuation says investors expect them to act, not just suggest.

Generative Video

Shortical Ships AI-Generated Microdrama Starring Real Actor Aki Avni

Image source: techcrunch.com

Summary: Shortical premiered "Inevitable," the first AI microdrama built on a real actor's likeness, with Aki Avni paid in line with standard acting rates. The company is now pitching face-licensing deals to creators. It's an early test of how likeness rights work in generative video.

Key Points:

  • Shortical's microdramas run about 2 minutes per episode, with the first 7-10 episodes free before viewers pay a small monthly subscription fee for the rest.

  • Aki Avni is the first real person whose likeness appears in the platform's AI-generated content, and he was paid in line with industry standards for a comparable non-AI role.

  • Shortical hasn't disclosed how much it offers for face-licensing deals and has not yet signed any contracts, despite advertising the program on its website.

  • In China, where the microdrama industry began, firms already pay people between $15 and $700 to license their facial likenesses for AI-generated content.

  • The microdrama viewer base skews heavily female, and many viewers watch short episodes at night, which makes the format a strong fit for mobile-first AI video pipelines.

Why it matters: If you're building generative video or avatar tools, Shortical's Inevitable is a live case study for how likeness licensing gets structured, paid, and consented to at production scale. The per-face pricing range from China ($15 to $700) gives your product or legal team a benchmark for what creator-side licensing costs look like. Anyone designing rights management or consent metadata for synthetic media should watch whether Shortical's first face deals actually close.

Neuroscience-ai

York University Finds AI Vision Models Match Brain Output But Not Mechanism

Image source: www.sciencedaily.com

Summary: York University researchers found that AI vision models can closely mirror brain activity while using very different internal mechanisms. Across 1,320 images, neuron predictions worked one way but failed in reverse. If you use these models to study perception, verify the match goes deeper than the output.

Key Points:

  • The study examined 1,320 natural photographs and realistic synthetic images featuring 10 familiar subjects to test how well models mirror neural responses.

  • An additional 300 images showing the same 10 subjects in different visual styles were introduced to stress-test the models beyond their training conditions.

  • Models predicted recorded neuron responses reasonably well, but reversing the direction showed neural data poorly predicted many internal model features.

  • Artificial neural networks have been among the leading tools for explaining the brain's visual system over roughly the past ten years, according to the researchers.

  • Comparable AI models are already being used to study hearing, language, and movement, meaning the mechanism gap could affect several neuroscience subfields.

Why it matters: If you're using ANN-based encoding models to study visual perception or build neuro-informed systems, this result means matching neural output is not proof of shared mechanism. The team validating these models against recordings should add the reverse-direction test, checking whether neural responses predict internal features, not just whether the model predicts neurons. Anyone deploying similar models for clinical research or human behavior modeling should treat current fit as a hypothesis to verify, not a conclusion.

Marketplaces & Agentic Commerce

AWS Marketplace Opens to AI Agent Buyers via x402 Payments

Image source: image.theregister.com

Summary: AWS Marketplace now lets AI agents buy some cloud software via the x402 protocol, billed in USD through existing AWS accounts. Amazon is simultaneously blocking shopping bots on its retail site. If you're running agents that need tools or data, this is the sanctioned route.

Key Points:

  • Amazon banned AI shopping agents from its main website while opening AWS Marketplace to bot-based buying, a split approach that treats cloud procurement differently from consumer retail.

  • Transactions settle in USD through AWS billing, so existing spending controls, budgets, and approval policies can be applied to agent-initiated purchases without building a new payments system.

  • Customers in the US can buy some marketplace software without an AWS account, using an email or social login plus a credit card, lowering friction for one-off tool purchases.

  • Sellers must register with AWS Marketplace and integrate the x402 SDK before agents can complete purchases, so the catalog of agent-buyable products will start small.

  • AWS plans to extend beyond software into data-as-a-service, with listings like US Building Permits, Contractors & Residents Data, and a US-EN Multi-Speaker Conversational Acoustic Corpus.

Why it matters: If your team runs agents that provision SaaS, data, or dev tooling, this gives you a sanctioned purchase path that still rolls up into the AWS bill your FinOps team already watches. The x402 integration means agent builders on Bedrock or custom orchestrators can call marketplace checkout directly instead of routing through a human. Platform and procurement teams should decide now which spending limits and approval thresholds apply to agent-initiated buys.

Voice Agents

Google's Duplex Voice AI Now Powers Gemini Tasks Across Search and Maps

Image source: storage.googleapis.com

Summary: Google's Duplex has been rebuilt on Gemini and now sits inside Tasks in Gemini, letting the assistant phone businesses to book dinner or haircuts for you. It's the most mature production example of a voice agent handling messy, unscripted calls, and that's the part engineers should study.

Key Points:

  • Duplex first debuted in 2018 and has since quietly powered business information updates across many Google products, including Search and Maps, for years now.

  • Business details gathered from Duplex phone calls have appeared in Search and Maps more than one trillion times, according to Google's own figures shared with the company's announcement.

  • The third architecture era removed fixed verticals entirely, letting users describe any task in free text, so the system can handle requests it was never explicitly trained for.

  • Live telephony needs low-latency streaming inference so replies arrive at a natural cadence, which pre-training on static web datasets alone does not solve for voice.

  • During COVID-19, Duplex expanded automatic merchant verification to keep hours and listings accurate, a use case that shows the system serves businesses as well as end users.

Why it matters: If you're building voice agents or telephony workflows, Duplex's shift to a unified Gemini architecture is the clearest public blueprint for handling interruptions and streaming latency at production scale. Teams maintaining business listing or local search data should also note that phone-call verification is now a first-class data source that Google feeds into Search and Maps. The three-era evolution from closed-domain pipelines to free-text prompting is a useful pattern for anyone trying to replace brittle intent-classification stacks.

Formal-verification

Mathematician Cracks Riemann Hypothesis, Solution Verified in Lean Proof Assistant

Image source: techcrunch.com

Summary: A mathematician released a Lean-verified solution to the Riemann hypothesis, a 1859 Millennium Prize problem. The proof is machine-checkable, but it came without a human-readable explanation, which the author has promised to fix.

Key Points:

  • The Riemann hypothesis dates to 1859 and is one of the seven Millennium Prize Problems, each carrying a one million dollar prize from the Clay Mathematics Institute.

  • The solution was released as a Lean-compilable proof, meaning Lean, an open-source language, can formally verify each logical step without relying on human reviewers alone.

  • The announcement came as an essay describing the discovery process, published on a Sunday night, with several participants confirming the result in a Zulip chat.

  • The Riemann hypothesis conjectures a hidden structure behind the distribution of prime numbers, which is why a proof would reshape analytic number theory and cryptography-adjacent research.

  • A separate General Riemann hypothesis proof from a Stanford postdoc was framed as 12 years of work, released partly to encourage discouraged mathematicians, and independently checked.

Why it matters: If you work on formal verification or proof tooling, this is the first high-profile case where a major open problem arrived as machine-checkable Lean code. Teams building verification pipelines for smart contracts or safety-critical software should study how the proof was validated without peer review. The missing human explanation is the real gap, and it will decide whether the result gets adopted by mathematicians or just archived as code.

New Tools & Job

  • AnswerDojo - AnswerDojo builds a mock interview around the job you're applying for. Choose one of 2,247 real postings from 28 companies, including Amazon, Google, and Netflix, or paste your own JD to see what the role tests. Then answer out loud as an AI interviewer asks questions, listens, follows up, and scores the session. Two voice mocks are free. No credit card required.

  • FullCastBook - FullCastBook turns a manuscript into an AI audiobook with a full cast. Upload an EPUB, PDF, DOCX, TXT or Markdown file, and it finds the chapters, works out who is speaking in each line and gives every character a distinct voice alongside an AI narrator. Regenerate a single line without redoing the chapter, then export the finished audiobook as one M4B with chapter markers, MP3 files or a ZIP. Freemium, paid plans from $9.90 a month.

  • Roomvana - Roomvana is an AI design tool for your home's exterior. Upload one photo of your house and try new siding or exterior paint colors, a new front door or garage door, new shutters or a new roof — or redesign the front yard, backyard or patio. The result keeps your layout, walls and windows, so what you see is still your house, just with the updates. It works indoors, too. Free to try on the web and as a mobile app.

  • British Accent Generator - British Accent Generator is an AI British/UK accent TTS workspace with 315 regionally labelled voices — RP and standard British, young British, Scottish, Welsh, and Northern Irish. Browse and preview speakers, paste a script, and download natural UK English audio for videos, lessons, podcasts, demos, and training. Also includes Voice Clone, Voice Design, Voice Effects, and AI Lip Sync. Free daily credits; paid credit packs and subscriptions unlock longer scripts and premium features.

  • Nano Banana (nanobananana.com) - Nano Banana at nanobananana.com is an independent AI image generation and editing tool for designers, content creators and marketers. Create images from text prompts, edit reference images with natural-language instructions and blend multiple images. A Chinese-language interface is available. It is separate from nanobanana.im and is not an official Google website.

  • Senior Platform Engineer, Replicate - About Us At Cloudflare, we are on a mission to help build a better Internet. Today the company runs one of the world’s largest networks that powers millions of websites and other Internet properties for customers ranging from individual bloggers to SMBs to Fortune 500 companies. Cloudflare protects

  • Creative AI Engineer - Why Harvey At Harvey, we’re transforming how legal and professional services operate. By combining frontier agentic AI, an enterprise-grade platform, and deep domain expertise, we’re reshaping how critical knowledge work gets done for decades to come. This is a rare chance to help build a generation

  • Research Engineer, Finetuning - About Anthropic Anthropic’s mission is to create reliable, interpretable, and steerable AI systems. We want AI to be safe and beneficial for our users and for society as a whole. Our team is a quickly growing group of committed researchers, engineers, policy experts, and business leaders working tog

  • Manager, Applied AI Architect - About Anthropic Anthropic’s mission is to create reliable, interpretable, and steerable AI systems. We want AI to be safe and beneficial for our users and for society as a whole. Our team is a quickly growing group of committed researchers, engineers, policy experts, and business leaders working tog

  • Systems Engineer - Network Software Performance - About Us At Cloudflare, we are on a mission to help build a better Internet. Today the company runs one of the world’s largest networks that powers millions of websites and other Internet properties for customers ranging from individual bloggers to SMBs to Fortune 500 companies. Cloudflare protects

QUICK HITS

  • Ricapar's ClawDOS Runs an AI Coding Agent on a 4.77 MHz IBM PC XT - Ricapar built ClawDOS, an AI coding agent that runs natively on an IBM PC XT with an 8088 at 4.77 MHz and 640K RAM. It writes, compiles with Turbo C, and runs programs in one tool call, and the whole harness is a 176 KB DOS binary. It's a fun stress test of how little hardware an agent really needs.

  • Dutch Regulator Finds Smart Glasses Unsafe for Users and Bystanders - Dutch regulator AP tested six AI smart glasses and called them unsafe for users and bystanders. Recording lights were easy to cover, nearby people could take control of the camera, and some models misread medication and weather warnings. Expect more privacy rules and venue bans to follow.

  • Researchers Show a 27B Model Can Rewrite Its Own Weights and Gain 6.6 Points - A 27B open model diagnosed its own failure and rewrote its weights, gaining 6.6 accuracy points on unseen problems. No training run, no gradients, and gains stacked across eight sessions. If this scales, your frozen model may soon improve itself for free at inference time.

  • PCB Shortage Could Rival Pandemic Chip Crunch as AI Servers Eat Capacity - AI servers are sucking up PCB capacity faster than factories can add it, and bookings are up 56.1 percent this year. Laminate lead times have tripled, and relief isn't expected before mid-2028. Hardware buyers should order early and qualify backup suppliers now.

  • California AG Subpoenas Nvidia Over Possible Chip Leaks to China - California's AG subpoenaed Nvidia to find out whether advanced GPUs or confidential data reached China and whether investors were misled. It lands on top of an SEC probe, a founder resignation, and a congressional inquiry. If you rely on Nvidia hardware, the compliance fallout could reach your supply chain.

  • Epic Ships Unreal Canvas in UE 5.9 as Unreal Engine 6 Drops Blueprints for Verse - Epic shipped Unreal Canvas in Unreal Engine 5.9, a Miro-style collaboration workspace. The UE6 preview is bigger: Blueprints out, text-based Verse in, and assets stored as text so AI agents can edit them. Blueprint-heavy teams should start planning now.

Read more news here →

THAT'S A WRAP

Partner with Us

Showcase your product to our engaged community of Tech enthusiasts

Our newsletter reaches a diverse audience of tech professionals, investors, engineers, managers, and business leaders worldwide. Hear when ads open.

More from the AI Newsletter

AI Newsletter

Get the next issue

One email a day. Free. Unsubscribe anytime.

How we handle your email: Privacy policy

Comments

Subscribe to join the conversation...

Be the first to comment