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:

• Entropy, the “zappier for crypto”, has shut down — learn why below

• What's working in AI for GTM

• Moltbot, the AI agent that “actually does things”, is going viral — learn all about it below

A huge thanks to today’s sponsor, Playbookz. Grab their $1k deal while it lasts!

Turn LinkedIn into a steady stream of buyer conversations ^AD

Playbookz helps founders, CEOs, and exec teams turn LinkedIn into a predictable growth channel.

They run the entire system for you:

• weekly content that builds authority

• outbound DMs that start real conversations

• ads that amplify what’s already working

The outcome is simple: more qualified meetings booked, more inbound from people who already know who you are, and stronger credibility before the first call.

Most teams see: 150k reach, 300+ leads, 45 calls & 9 new clients in the first 90 days

Book a no-obligation call here. You’ll get $1,000 off your first month.

This Week In Startups

🔗 Resources

What's working in AI for GTM

10x job posts for 10x engineers

How to understand customers

Pilot helps startups run world-class finance, including bookkeeping, payroll, taxes, and fundraising, powered by dedicated US-based experts and AI-driven insights * 

📰 News

OpenAI’s Sora app is struggling after its stellar launch.

Chrome takes on AI browsers with agentic features for autonomous tasks.

OpenAI launches Prism, a new AI workspace for scientists

China’s Moonshot releases a new open source model Kimi K2.5 and a coding agent.

💸 Fundraising

The counter-drone startup Frankenburg raised $50M at a $400M valuation.

Crypto payments startup Mesh raises $75M at $1B valuation.

CVector raises $5M for its industrial ‘nervous system’.

Jelou raises $10M to turn WhatsApp chats into transactional AI hubs

* sponsored

Fail(St)ory

Putting Agents in Crypto

Entropy shut down this week. It was an a16z-backed crypto custody startup that slowly turned into “Zapier for crypto” with AI and fancy security sprinkled on top. You can probably guess how that ended.

Four years in, the founder pulled the plug, laid out the pivots, and said the quiet part out loud: they couldn’t find a venture-scale business model. 

What Was Entropy:

Entropy started in July 2021, right when “crypto infra” was a cheat code for fundraising. 

At the beginning, Entropy tried to replace the “custody company” model. Normally, if a business doesn’t want to lose millions to one misplaced key, it pays a custodian to hold the assets and approve withdrawals. That’s convenient. It also creates a giant target and a single place where things can go wrong.

Entropy’s approach was basically: don’t let any one person or one system have full control. The “key” was split into pieces, and moving funds required multiple approvals working together. So even if one piece got compromised, it wasn’t enough to drain everything.

That was the wedge. But the wedge had a ceiling.

By 2021-2025, custody got crowded and entrenched fast. Big players owned the institutional market. Wallets on the retail side grew like weeds. Entropy could build something safer and still struggle to convince people to switch, because switching custody isn’t like switching a note-taking app. It’s scary, slow, and full of committees.

So they pivoted.

Instead of only securing assets, they tried to make the wallet do work. They leaned into automation: software that could run actions for you in crypto, under rules you set. The company described it as basically Zapier or n8n for crypto, with signing built in so workflows could actually move funds without you clicking through prompts all day.

And because it was 2025, the automation pitch came with the full AI garnish. “Agents.” “Autonomous.” “Context-aware.” The idea was that you wouldn’t just click buttons in a wallet anymore. You’d delegate intent. An agent would watch markets, hop across chains, route transactions, rebalance, maybe even trade, all inside guardrails you defined.

In their version of the future, your wallet stopped being a vault and started acting like a junior analyst that never slept. It would “optimize” your holdings, look for opportunities, and execute based on your strategies and risk limits.

The Numbers:

• 💰 Total raised: ~$27M

• 🧾 Biggest round: $25M seed (2022) led by a16z crypto

• 📆 Founded: July 2021

• 📉 Market context: crypto venture deal count fell ~60% YoY in 2025 (about 1,200 deals vs 2,900+ in 2024)

Reasons for Failure: 

• The model never matched the cap table: The founder said they ran an “initial feedback request” and learned the business model wasn’t venture scale. That’s the whole story right there. You could ship something real and still not have a path to a massive outcome.

• They kept pivoting into bigger, harder problems: Entropy started with custody via threshold signing. Then it became an agent network. Then it became “Zapier for crypto” plus AI plus TEEs plus automated signing. Each step sounded more powerful, but it also widened the surface area and pushed them further into “platform” land.

• The market had already picked winners in custody: Institutional custody had been dominated by players like Fireblocks and Copper. Retail wallets like MetaMask and Phantom had expanded aggressively. Entropy’s differentiation was deep security architecture, but custody buyers rarely switched for elegance. 

• Crypto infra got hit when funding dried up: Entropy shut down in a year when crypto venture activity dropped sharply, with deal counts down about 60% year over year in 2025. That kind of environment punished long-horizon bets. 

Why It Matters: 

• Buzzwords don’t add up to demand. Crypto + agents + AI sounded cool. It still needed a buyer with a budget.

• Custody is a trust market, not a features market. “Safer” didn’t beat defaults, audits, and existing relationships.

• Test venture-scale early, not after years of building. Product feedback is cheap. Business model feedback decides the company.

Trend

Moltbot

There’s a new open-source AI agent going viral these last few days. It’s called Moltbot, and the pitch is basically “AI that actually does things.”

It became one of the fastest growing projects in Github history, crossing 85.000 stars in just two weeks, and it feels like giving a hyperactive kid the keys to your laptop: useful, fast, and a little terrifying.

Why it Matters

• Power and risk ship together now. Moltbot is fun because it can actually do things: run commands, touch files, click around the web, move money if you let it. That’s also why it’s scary, because one bad config or one sneaky prompt injection turns “personal assistant” into “remote-control chaos.”

• Distribution is hiding in plain sight. Moltbot doesn’t win by having a shiny new UI. It wins by living inside WhatsApp and Telegram and Slack, where you already spend your time. That’s the real growth hack.

What it is

Moltbot is an open-source, self-hosted “personal assistant” you run on your own machine or server.

You talk to it through chat apps you already live in (like Wahtsapp or Telegram), so it feels like texting a capable assistant who has access to your computer.

It’s agentic in the literal sense. It can run shell commands, read and write files, automate the browser, trigger webhooks, and run scheduled jobs. That’s why people call it persistent and 24/7. It’s not a tab, it’s a process.

It’s also not “one model.” It can route work across different model providers and different agents or workspaces depending on your setup. The core vibe is a local gateway that handles sessions, channels, tools, and events, plus a growing “skills” layer that expands capabilities fast.

Why it went viral

First, because it’s a new magic trick: texting your computer.

That sounds small, but it hits hard. People don’t want to “open an AI app.” They want to send a message like “handle this” and get a result back. Moltbot makes that feel normal in about 30 seconds.

Second, because some of the stories are ridiculous in a way the internet loves.

One user made Moltbot buy him a car. The bot searched for the car, negotiated the price and even close the deal.

And the one that should be printed on a warning label, where the bot buys a $3K “build your personal brand” course while the user sleeps:

Those stories do more marketing than any landing page. Because they’re not about intelligence. They’re about agency. The bot isn’t impressive because it talks well. It’s impressive because it can move money, click buttons, and make decisions you forgot you delegated.

Of course, this is also extremely dangerous.

Because the moment your agent can run commands and read files, you’ve created a new attack surface. Not just “someone hacks your server.” More subtle stuff too. Any untrusted thing the bot reads can mess with it: emails, web pages, attachments, DMs, random pasted text.

This is why the whole thing is so watchable. It’s cool. It’s scary. It’s both at the same time.

The Trend

I think Moltbot is telling us a few things about where AI agents are heading:

• Chat is the control plane. The killer move is hiding an agent inside WhatsApp/Telegram/Slack so it feels like texting, not “using software.” That’s why it spreads. It feels like texting a human assistant.

• People care about side effects, not IQ. The posts that travel aren’t “look how smart it is.” They’re “it did something real” and sometimes “it did something real I didn’t want.” That’s the new bar.

• The market is going to split into two camps. One camp builds “spicy” agents that move fast and break things. The other builds the boring layer that makes them safe: permissions, sandboxing, secrets hygiene, skills you can actually trust.

Moltbot is basically a preview of the next year: more power, less friction, and a lot more ways to shoot yourself in the foot.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico