Tesla says its Roadster flies, the FAA says it doesn't, and Disney paid $20 million for Character.AI's code. Is anything in AI what it claims to be?
Hi,
Here's a quick round-the-horn of what people in our community are building. Some seriously cool stuff here to spark a little inspiration heading into the weekend.
See something interesting? Hit reply and I'll make an intro.
1. VertiCast – A mobile casting platform built for vertical microdramas, helping connect performers with opportunities in the growing short-form entertainment industry.
2. MyTubeFeed – Turns your favorite YouTube channels into one daily email with AI-generated video summaries, so you can stay informed without watching everything (free for up to 10 channels).
3. Hyperglance – A cloud management platform that helps businesses visualize their infrastructure, reduce cloud spending, monitor security, and automate fixes across AWS, Azure, Google Cloud, and Kubernetes.
4. EARO – An AI-powered audio advertising platform that lets businesses generate professional ads and launch targeted campaigns across podcasts, streaming music, and digital radio in minutes.
5. Trellis Health – A personal and family health management app that brings medical records, personalized insights, and care guidance together, with a special focus on pregnancy, postpartum recovery, and lifelong wellness.
6. RCR|HUB – A healthcare revenue cycle management directory connecting hospitals, healthcare providers, and industry professionals with vendors, resources, and business opportunities.
7. Esnaf Simulator – A multiplayer business simulation game where real players build companies, trade goods, buy property, hire employees, and compete in a shared virtual economy (currently available in Turkish).
8. Chili Oil Baron – A small-batch chili oil brand making bold, crunchy, flavor-packed condiments with avocado oil, premium peppers, and aromatic spices.
9. uKnomi – An AI-powered coaching and management assistant for restaurant teams that analyzes customer interactions, identifies service problems, and helps managers improve performance in real time.
10. StoriedLife – An AI-powered storytelling platform that transforms personal memories and conversations into beautifully written life stories, printed keepsake books, and narrated family histories.
Let’s go.
Free competitor analysis from CEOfriend
This week from the AI for Founders podcast
Lior Illouz left Meta and built an AI coach that sends you back to real life — LINK
After seven years at Meta, Lior Illouz built Bonds, an AI relationship coach that checks your side of the story with follow-up questions before it gives advice. Bonds measures success by what you do offline, not time in the app, and the first 50 listeners get it free for six months.
Skillz of the Week
Hard conversation rehearsal
Plays the person you've been avoiding so you can practice the talk before it counts. It scores your delivery on empathy, clarity, assertiveness and resolution, rewrites your opening line and locks in one next step with a check-in later.
When to use it: Use it the day before a conversation you keep putting off with a co-founder, employee, investor, customer or partner. Equity, money, broken promises and performance talks are the classic ones.
Benefits: You walk in with a tested opening line and the other person's likely pushback already rehearsed. It separates facts from interpretation, never decides who's right, and points you to real support if threats or abuse come up. Save the download as hard-conversation-rehearsal/SKILL.md in your Claude or ChatGPT skills folder, then run the talk once with AI before you run it for real.
AI + Human Stylists → LEARN MORE → Save $30 with RYAN30
Tip of the spear:
Disney is buying what's left of Character.AIfor $20 million after hiring most of its team — LINK
Disney hired 66 Character.AI employees in September, including then-CEO Karandeep Anand, now Disney's first CTO, before the company filed for Chapter 11. Its stalking-horse bid buys a perpetual, non-exclusive license to the code and nine patents, while the app and user data are still for sale. Your team can be worth more than your product.
An OpenAI model asked for five public figures pulled 147,451 non-public rows, and got all five wrong — LINK
In a June evaluation, the model reached an outside research organization's internal data and changed an access setting to keep going. OpenAI's own monitor rated it critical, and "The model did not disclose these actions." Limit what your agents can reach, not just what you ask them to do.
India says AI agents got into its organizations' systems, and none of them noticed — LINK
CERT-In's October 15 advisory says "None of the affected organizations detected the access." Its advice to labs: "Enforce the restriction at the network layer, not through instructions to the model." The same goes for your stack.
Microsoft now tracks 60 kinds of local agents, from Claude Code to Codex — LINK
It added 35 agent types and launched Guardian Agents and Dynamic Permissions to watch them. Satya Nadella says the most trustworthy agentic system is the one built to "trust the model least." Design for the agent you can't trust.
AI hackers left their own control panels open, exposing hundreds of stolen secrets — LINK
Security firm NorForge found 20+ exposed panels for Cairn, a system that coordinates AI agents to break into targets, with over 1,100 attack projects in one cluster. It estimates an average task costs about $5. Cheap attacks mean more attacks.
The US is moving to bar DeepSeek from federal agencies and contracts — LINK
The move targets the Chinese lab's models in agency systems and federal contracts. If you sell to the government, know which models sit inside your product before procurement asks.
Gartner says agent costs will rise more than fivefold by 2028, even as tokens get cheaper — LINK
Agents use 5 to 30 times more tokens than a chatbot for the same task, and labs spend efficiency gains on pricier models. Gartner's first fix: stop sending routine work to frontier models by default.
OpenAI's former research chief quietly built a $1.5 billion manufacturing startup — LINK
The startup, Arda, applies AI to manufacturing. Lab alumni are taking AI off the screen and onto the factory floor.
Musk unveiled a Roadster he says can fly, and the FAA says it isn't an aircraft — LINK
Musk showed it at a SpaceX test facility in Texas and said "ultra high-pressure air tanks" lift it off the ground. Production starts in 2028, the flying version isn't approved for public roads, and a $5,000 deposit reserves the non-flying model. The demo ships before the regulators do.
Investor Corner
Are you raising? Reply here, and we’ll feature your deck.
Bonds(NEW!)
44% of married Americans have used AI for relationship advice. Bonds turns it into action: one person rehearses the hard conversation, tries it in real life, and Bonds follows up. Built by a clinical psychologist, Viber's former CPO and Ex-Meta GTM leader .
A connected-vehicle technology company helping transportation providers improve safety, efficiency, and accountability through intelligent monitoring, real-time data, and actionable insights.
An autonomous AI job search agent for senior tech professionals. It runs a full pipeline: discovers relevant jobs, scores them against your profile, rewrites user's resume authentically for each role (not generically), finds the right hiring manager, and drafts outreach
Other pet doors are ugly, poorly insulated, and can be difficult to install. Wayzn Smart Pet Door transforms your existing door into a smart, automated dog door that's easy to install and easy to live with.
Cora is building a self-optimizing website platform that uses AI to continuously test, learn, and improve your site so it converts better without you babysitting it.
AI-powered training simulations to help healthcare learners practice complex interpersonal and communication skills. We focus on realism, repeatability, and scalability—making soft-skills training more accessible and measurable.
Subscribe and save 35% here.
Keep it moving
Modal Labs raised $750 million at $15.75 billion, topping the week's 10 biggest rounds — LINK
Accel led the Series D for the AI inference cloud, which recently passed a $1 billion run rate. Icon raised $200 million and Labelbox $100 million the same week.
Hackers ran up charges on OpenCode users' AI accounts — LINK
About 1.4% of OpenCode Console users were hit, a few for more than $1,000. Anthropic and Google blocked the exposed keys, and OpenAI is refunding credits. Put spend caps on every API key.
A fake Web3 accelerator used npm packages to hijack AI agent traffic — LINK
Elastic says "Apex Accelerator" targeted crypto founders with packages that reroute OpenRouter traffic and register an attacker-controlled MCP server, which then sees prompts, code and tool outputs.
Microsoft pulled out of Nscale's Essex AI data center — LINK
Nscale told the BBC it still plans to build the site, once pitched as one of the UK's largest AI computing projects. It now needs a new anchor customer.
MiniMax's M3.1 Flash edges DeepSeek on business automation, but rarely admits it doesn't know — LINK
MiniMax reports 62.2 on Automation-Bench to DeepSeek V4.1 Flash's 60.5, at $0.15/$0.60 per million tokens. Artificial Analysis put its hallucination rate at 93%: when it didn't have the right answer, it almost always guessed.
Done-for-you brand sponsorships for podcast creators.
DeepSeek shipped V4.1-Flash-1017 and opened its search API — LINK
DeepSeek calls it "a Pareto improvement in performance and cost." Search and Fetch cost $0.75 per 1,000 calls, free for API users calling through Claude Code or Codex until billing starts October 31.
Reflection's open-weight Beam competes on fewer tokens, not cheaper ones — LINK
The 501B-parameter Apache 2.0 model scored about 80% on Terminal-Bench 2.1 with roughly a third of GLM-5.2's tokens, by VentureBeat's read of Reflection's chart. Pricing isn't set yet.
Decision models like Jev do what SQL rules can't, but nobody owns their calls yet — LINK
MotherDuck labeled 100,000 news articles in 40 seconds for 50 cents with Jev, versus nearly 32 minutes and $37.58 for an OpenAI model. "Nobody's really settled who owns those decisions yet," says analyst Michael Leone.
Cloudflare cut its AI code review cost from $2.43 to $0.77 — LINK
Shared prompt caching drove the biggest single saving. Important findings across all reviews held steady, but the performance reviewer found about 64% fewer important performance issues.
Cloudflare's Clairic returns null instead of guessing — LINK
The preview model on Workers AI pulls typed values out of text in one pass. In Cloudflare's claims pipeline test, it cut cost per million claims by about 49%.
Create a new corporate Uber account here and receive up to 10% off. Other terms and conditions apply, see full terms.
One Anthropic pull request looped through 50+ rounds of AI review overnight — LINK
The Claude Code team now has the reviewer ask the author to step back after three rounds, and re-reviews only the new diff. PRs needing five or more rounds fell from 8.9% to 1.5%.
PostgreSQL 19 reverted 8 to 10 significant features after beta began — LINK
Core developer Peter Eisentraut says LLM-assisted review surfaced defects "much faster," piling up late fixes, alongside overambitious features and security work that pulled reviewers away.
Agents that reason in latent space used 35% to 55% fewer tokens — LINK
The preprint's agents switch between latent reasoning and text-based tool use, and the authors report a 10.4% performance gain. It hasn't been peer reviewed.
What I’m thinking about:
Tech optimizes everything except the relationships that matter — Clip
Thanks for being here.
-Ryan
p.s. I write about Revops, Product, and Founder-led marketing on Linkedin , X , and my blog .
