Hey — It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read. If you only have one, check this:

• AllHere, an AI education startup working with the L.A school district, shut down — find out why below.

• Jason Fried shared startup ideas and how to build products.

• Carta published data insights on startup hiring in 2024.

• Y Combinator is moving to four cohorts a year in 2025.

• Open AI launched a new GPT model that “thinks” through problems and is much better at reasoning — below is everything you need to know.

This issue is brought to you by:

• Intercom’s Startup Program, which is giving startups a 100% discount for 12 months to Intercom, the AI-first customer service solution.

• Aimply, a fully managed newsletter platform that helps businesses stay top of mind of their audiences and leads.

Intercom for Startups ^AD

Join Intercomʼs Startup Program and get 12 months free.

Create a direct line to your customers. Experience the only complete AI-first customer service solution.

Join thousands of other fast-growing startups using Intercom such as Coda, Qonto and Pitch.

This Week In Startups

🔗 Resources

Jason Fried shares startup ideas and how to build products.

How to think of product design when building AI products.

ChartMogul's founder shares their path to $10M ARR through strategic partnerships.

How to gather and use the right information before sales calls.

Carta shares data insights on startup hiring in 2024.

📰 News

Salesforce Ventures announces a new $500M AI fund. 

Typeface acquires two AI startups Treat and Narrato.

European VC Noteus targets more than €600M for debut fund to back tech startups.

Fintech startup Revolut is growing faster than Facebook and Snapchat in Europe.

Y Combinator will move to four cohorts a year in 2025.

💸 Fundraising

TeamBridge raises $28M to develop HR software for hourly employees.

AI Agent startup 11x raises $24M in Series A and relocates to the US.

AI coding assistant Supermaven secures $12M from OpenAI and Perplexity co-founders.

Howbout secures $8M to develop its shared calendar app for friends.

Fail(St)ory

Nothing (All)Here Anymore

A few weeks ago, AllHere, a promising startup in the world of AI-powered education, announced it was closing its doors. 

Founded in 2015, the company hoped to transform how K-12 schools engage with students and families through its flagship product, Ed, an AI-driven co-pilot designed to offer personalized support. 

Backed by over $12M in funding, half of which came from a partnership with the Los Angeles Unified School District (LAUSD), AllHere seemed set for success. However, behind the scenes, financial overspending and questionable practices started to surface, ultimately leading to its downfall.

What Was It: AllHere developed Ed, an AI chatbot designed to communicate with students and families, providing 24/7 support in over 100 languages. 

The chatbot was intended to analyze student data, such as grades, attendance, and test scores, to create personalized action plans for each student. It could recommend study schedules, skill practice, and other resources based on the student’s unique needs.

For educators, the platform aimed to provide insights to help them make more informed decisions, adjust instructional strategies, and address the needs of each individual student.

What set AllHere apart from other edtech solutions was its collaboration with LAUSD, the second-largest school district in the U.S. This partnership was a bold move, which could have resulted in one of the most significant AI implementations in K-12 education. 

The Numbers:

• 📅 Founded in 2015.

• 💰 Raised $12M, half of which came from the Los Angeles Unified School District.

• 🧑‍💼 Developed Ed, an AI co-pilot used to support K-12 students and families.

• 🔥 Filed for bankruptcy earlier this year and is now in court, trying to sell its remaining assets.

Reasons For Failure:

• Mismanagement and Financial Irregularities: AllHere’s financial troubles came to light during ongoing bankruptcy court proceedings. These proceedings revealed some questionable expenses that raised serious red flags. CEO Joanna Smith was reimbursed $243,000 in expenses over the past year, sparking concerns about excessive spending. Additionally, the company owed $630,000 to salesperson Debra Kerr, but no proper invoices were found to support the debt. These financial irregularities, uncovered during the court process, contributed significantly to AllHere’s instability and eventual collapse.

• Privacy Violations: Earlier this year, a former executive, Chris Whiteley, came forward as a whistleblower, claiming that AllHere mishandled sensitive student data. According to Whiteley, the AI-powered Ed platform put student records at risk of being hacked, violating key data privacy principles. This revelation shook confidence in the company, especially given how essential data security is in the education sector.

• Operational Chaos: Whiteley also mentioned that the company was consistently “on fire” in terms of operations. Apparently, the startup struggled to manage its ambitious project and had frequent management issues that affected day-to-day functioning. 

• Lack of Trust and Transparency: Mismanagement, both financially and operationally, led to a lack of trust within the company and from external stakeholders. The court proceedings and whistleblower allegations made it clear that transparency was lacking and key decision-making processes were flawed.

Why It Matters:

• AllHere’s collapse is a reminder of the critical importance of trust and transparency, especially when handling sensitive data like student records.

• This case highlights the challenges of scaling AI-powered solutions in education, where privacy and data security must be top priorities.

• Financial mismanagement can quickly erode confidence and destabilize even the most innovative startups.

• The failure of AllHere shows that even well-funded companies with strong partnerships can fall victim to internal chaos if operations and leadership aren't tightly managed.

Weekly Tool

We make building newsletters easy.

Aimply is a fully managed newsletter platform that helps businesses stay top of mind of their audiences and leads.

They manage the entire newsletter release process for you: From design to writing and design to distribution.

They do all this and learn about your audience over time, so your newsletter is constantly improving.

Signup takes under a minute and it's free to try.

Try it free →

Trend

Think Before You Answer

Last week, OpenAI rolled out a new line of models, GPT o1, and it's clear this isn't just another iteration. These models represent a major shift in how AI processes information, focusing on thinking through problems more thoroughly before delivering a response.

Why It Matters:

• Sharper Problem-Solving: GPT o1 is designed to tackle more complex tasks than previous models, making it a standout in areas like coding, science, and math. 

• Boost in Reasoning Abilities: The leap in reasoning here is no joke. GPT o1 is outperforming expert humans on incredibly challenging problems, including PhD-level science and competitive programming questions.

• Efficiency and Cost Options: For those who want the power of GPT o1 without the price tag, there's also GPT o1-mini. This smaller version sacrifices a bit of speed but still gives much better results than other GPT models.

What’s New with GPT o1?

The core idea behind GPT o1 is that it doesn't rush to answer. Instead, it “thinks” through problems—reasoning step-by-step before delivering an answer. This is a big change from what we’ve seen in earlier models, which tend to respond more quickly, sometimes at the cost of accuracy. With GPT o1, the focus is on taking its time to get it right, which is especially useful in areas like science, math, and programming.

For example, here’s a demonstration of GPT o1’s coding abilities. While it takes longer than other GPT versions to respond, the code it outputs is much more reliable and often works without the need for modifications:

Benchmarks

When it comes to math, programming, and science, GPT o1 drastically outperforms not only previous GPT versions but also any other language models. In many cases, it even surpasses human experts in their respective fields.

For example, the model now ranks in the top 11% of competitive programming tests and placed among the top 500 in a USA Math Olympiad qualifier:

So, what do you think?

Help Me Improve Failory

That's all of this edition.

Cheers,

Nico