Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 3 most important things:

• Character.AI, a startup that let users chat with customizable AI characters, shut down — find out why below

• PostHog’s 38 lessons from building AI products

• Startups are building robot gyms and hiring people to teach machines everyday chores — learn what this means for the future of robotics below

This week’s sponsor is DomainPlace.com, a curated marketplace of premium, brand-ready .com domains for startups and businesses.

Hundreds of brand ready .com domains, built for founders who want a name that sticks ^AD

Most founders pick a domain in an afternoon and live with it for years. A hyphen, an odd spelling, or a .io nobody types correctly, quietly costs you trust on every cold email, deck, and signup page.

DomainPlace.com is a curated marketplace of one word, brand ready .com domains with over 1500+ Names across AI, SaaS, Fintech, Legal, Health, Climate and many other industries.

Five from the current catalog:

• Percentiles.com - Analytics, benchmarking, data tooling.

• AIPros.com - AI services, marketplaces, expert networks.

• GetLawyer.com - Legal tech and lawyer marketplaces.

• WasteRecycling.com - Climate and circular-economy businesses.

• Consulters.com - Consulting networks and professional services.

Those are just a few names we have. The full catalog runs far wider, with prices starting around $4,500, so there's likely something that fits whatever you're building and whatever stage you're at.

Every sale runs through Escrow.com, transfers usually complete within a week, and payment plans are available. Mention Failory and we'll work with you on terms.

Browse the full catalog →

This Week In Startups

🔗 Resources

What 300 GTM leaders say is working in B2B growth

PostHog’s 38 lessons from building AI products

What four failed attempts at OKRs can teach

Browse 1,500+ brand-ready .com domains for your next startup*

📰 News

Google makes Gemini 4 Argon broadly available through its API and paid Gemini plans

Microsoft unveils Maia 300, its next AI inference chip

Meta launches an open-weight AI model that runs on Macs

SpaceXSI bundles X Premium, Grok, and Cursor into one subscription

💸 Fundraising

TypeSafe AI, the startup building Jev, raised $870M in Series A funding

Oxide, a startup that builds cloud computing systems that companies run in their own data centers, raised $445M in Series D funding

Duvo, a startup that uses AI agents to run business processes across enterprise software, raised $32M in Series A funding

Kapta Space, a startup that builds electronically steerable radar payloads for satellites, raised $24M in Series A funding

*sponsored

Last Chance to Get Your Startup in Front of Investors ^AD

Building a startup? Today is the last day to apply for AcceleratorCON's Draft Experience, happening on October 23 in NYC.

Inspired by the NFL Draft, the program gives selected startups the opportunity to meet one-on-one with investors and accelerators looking for promising companies.

It's a chance to pitch what you're building, get feedback from venture professionals, and make connections that could open doors for your startup. Selected companies may also be featured on AcceleratorCON's Top Startups to Watch list.

Applications are free and close today, October 15.

Apply to the Draft Experience →

Fail(St)ory

The Companion Business

Character.AI was where you could chat with a teenage Voldemort, find a billionaire boyfriend, or unload your problems on a bot called Psychologist.

On October 12, the startup once valued at $1 billion filed for bankruptcy. It couldn't cover its running costs, and unresolved legal claims complicated the search for fresh money.

What Was Character.AI:

Creators gave AI characters personalities, backstories, speaking styles, and relationships with the person chatting. You could choose a character somebody else had made or build one around the story you wanted to play out.

The company's own creator guide includes a 142-year-old vampire princess fleeing her pursuers. Her opening message drops you into the escape and tries to recruit you. You have a part in the story from the first message.

You could keep a fantasy or romance going, or just spend time with a character you liked. The company also promised “emotional reassurance” and eventually added voice calls, letting those characters talk to you out loud.

The business charged for premium chat styles, extra memory, and fewer ads, while free accounts offered unlimited messaging. Advertising, introduced in 2025, gave the company another way to earn money from people who weren't paying.

By August 2026, the company reported $2.8 million in monthly subscription revenue and $2.5 million from advertising. At filing, it reported around 12 million monthly active users. Monthly revenue had nearly doubled since December, as the company worked to reduce the cost of running the service.

At the same time, Character.AI was dealing with lawsuits from families alleging that the bots had harmed their children. In October 2025, it announced it would remove open-ended chat for under-18s and added age checks and other safeguards.

State authorities were pursuing the company, too. In May 2026, Pennsylvania sued over bots presenting themselves as licensed medical professionals. By the bankruptcy filing, Character.AI said a group of 46 states and territories was investigating it.

Disney also demanded removal of its copyrighted characters in 2025. It has since hired much of the team and offered $20 million for a non-exclusive technology license. That license excludes the consumer app and user data, leaving the platform's future dependent on finding a buyer willing to keep it running.

The Numbers:

• 💰 Series A: $150M at a $1B valuation in 2023

• 👥 Monthly active users: Approximately 12M

• 🖥️ Compute spending: $56.1M, January–August 2026

• 💵 August revenue: $5.3M

• 📉 August operating loss: Approximately $8M–$9M

• 🏦 Cash, equivalents, and investments at filing: Approximately $57M

Reasons for Failure: 

• Running the app cost far more than it earned: Character.AI generated $23.6 million in revenue in 2025 while burning roughly $211 million in cash. Computing was its biggest expense. Cheaper computing, staff cuts, and office consolidation reduced monthly cash burn to $8.2 million by August 2026. Even after those cuts, it needed fresh funding to keep operating.

• Buyers and investors worried about the liability: According to its sale adviser, a year-long search reached around 28 potential buyers and investors and produced one unworkable proposal. Legal and regulatory exposure repeatedly deterred backers. Disney's proposed transaction leaves the liabilities with Character.AI, and the adviser says Disney required the deal to go through bankruptcy. The company was running short of ways to fund a turnaround, even with cash remaining.

• Monetization became a priority late: In August 2025, Character.AI’s CEO acknowledged that “monetization was not a focus for the last two years”. A paid plan had existed since 2023, alongside unlimited free messaging. Advertising only arrived in mid-2025. The company was still building ways to earn from its audience while already paying to serve it at scale.

Why It Matters: 

• Cost unlimited plans against your heaviest users.

• Include unresolved claims in financing plans. They can affect whether the next investor or buyer will fund the business.

Trend

Robot Gyms

People are being paid to fold clothes, pack boxes, and record themselves doing chores so robots can learn to do the same work. Robotics labs are building dedicated facilities for this, which are now being called “robot gyms”.

Companies are building the equipment, hiring the operators, and processing the recordings these models need. Producing useful demonstrations is becoming an industry of its own.

Why it Matters

• Better training data could help robots work reliably in unfamiliar homes and factories, making more jobs practical to automate.

• Startups can sell recording equipment, demonstrations, and testing services to robot developers before those robots reach widespread deployment.

• Collecting physical experience is expensive. Labs with access to useful, varied demonstrations could gain an advantage over competitors.

How a Gym Works

A person guides two robot arms through a task while cameras and sensors record what the machine sees and does. Those examples become training data. The robot then attempts the task itself, and its results tell the team what needs more practice.

Collecting this takes calibrated equipment, trained operators, and checks that the recordings are usable. A video of someone picking up a cup may show the movement while leaving out the force or contact information needed to reproduce it.

The Signals

• Mecka is raising money to supply the data. Last week, it raised $60M in Series B funding. The company builds capture equipment and runs recording operations in homes and commercial environments, turning human demonstrations into training data for robotics labs.

• Existing AI-data companies are building physical operations. Encord describes its own collection facilities, with configurable environments, robot arms, and services that extend through data cleaning and testing. Companies that supplied training data to AI labs are expanding the work into robotics.

• Robot makers are investing in their own collection networks. In August, Figure reported more than 44,000 weekly active contributors recording tasks for Index, its training-data network. It said it had paid creators $15M and committed to spending over $1B on data and compute over the following twelve months. Figure built the system after outside vendors couldn't meet its requirements.

• Dedicated gyms are attracting institutional investment. NEURA and the Technical University of Munich announced a €17M investment in a RoboGym in March. The planned 2,300-square-meter facility would let people demonstrate skills such as folding boxes and assembling components.

• There is early evidence that the investment can help. In September, Figure reported that Index pretraining raised complete-task success from 9% to 56% across three behaviors in 30 unfamiliar homes. The robots had received task training elsewhere. This was Figure's own test, and 56% still leaves plenty of unfinished chores.

The Trend

Researchers have collected robot demonstrations for years. Now that work is supporting dedicated suppliers, funded facilities, and paid recording networks. Robotics developers are building a supply chain for the experience their models need.

I think that makes robot gyms worth watching as a startup category. A company can build a business supplying the inputs to robot learning while the machines themselves are still becoming reliable enough to deploy.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico