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:

• Hardt Hyperloop, a dutch startup trying to build hyperloop technology, has shut down — learn more below

• The world needs an AI lab — for Data

• This week, the biggest AIs got into your office tools — learn why this matters below

A huge thanks to today’s sponsor, Playbookz. Turn LinkedIn into a consistent client-acquisition channel without creating content or managing DMs.

Turn LinkedIn into a steady source of buyer conversations ^AD

Playbookz helps founders use LinkedIn to generate qualified meetings, without writing posts, managing DMs, or running ads themselves.

They handle: founder-led content, targeted outreach, paid amplification, and a setter who turns conversations into booked meetings.

That means:

• more qualified meetings

• more inbound from people already familiar with you

• more trust before the first call

• less pressure to post constantly or chase reach

The result is simple: steady demand gen without adding more to your plate.

Book a no-obligation call to see if it fits. Failory readers get $1,000 off their first month.

This Week In Startups

🔗 Resources

The world needs an AI lab — for Data

Six breakthrough business models reshaping global growth

AI Safety Has 12 Months Left

Catch up on today's AI stack (OpenClaw, Claude Code, agents) in four hours. First 10 get 50% off with code: FAILORY *

📰 News

ChatGPT can now create interactive visuals to help you understand math and science concepts.

Meta acquired Moltbook, the AI agent social network.

Anthropic launches code review tool to check flood of AI-generated code.

X is testing a new ad format that connects posts with products.

💸 Fundraising

Cryptio raises $45 million to help big firms keep track of digital assets.

French AI startup AMI raises $1B to develop 'universal intelligent systems'

Chinese brain interface startup Gestala raises $21M.

Kai raises $125M to build an agent-driven AI security platform.

* sponsored

Fail(St)ory

Sci-Fi Trains

A hyperloop startup went bankrupt last week, which feels a bit like hearing that someone’s flying car company had a tough quarter.

Still, Hardt was not just selling sci-fi with a seed round. It spent years building test infrastructure and trying to prove that this very dumb-sounding idea might, somehow, work.

What Was Hardt Hyperloop:

Hardt was a Dutch startup building hyperloop tech. The basic idea, in case you’ve successfully avoided this category for a decade, is simple: put a vehicle inside a low-pressure tube, cut drag, use magnetic levitation and electric propulsion, and move people or cargo very fast.

Sort of a train, except much more expensive, much harder to build, and wrapped in sci-fi branding.

The bigger promise was classic frontier-tech stuff. Less flying. Less road traffic. Faster trips between cities. Cleaner transport. Europe, but compressed.

Hardt spent years trying to turn that idea into something more solid than renderings and conference slides. It wanted to show hyperloop could work as an actual network, not just one straight tube for a flashy demo. 

So it focused hard on lane switching. Hardt argued that if vehicles could not move between routes, the whole thing stayed a toy. A transport system has to behave like one.

And yes, the company did build things. Its main asset was the European Hyperloop Center in Veendam, a 420-meter test facility. There it tested levitation, propulsion, vacuum conditions, and later lane switching.

In 2024, Hardt finished building the test center. Later that year it announced its first successful test and then a full lane-switch test. In 2025, it said it had run more than 750 test missions and hit 85 km/h. 

So this was not fake. That matters. But 420 meters is still 420 meters. At that scale, you are proving that pieces of the concept work. You are nowhere near proving that the business works.

That is what made Hardt interesting. It had enough real engineering to avoid the usual vaporware jokes. It still looked painfully far from anything commercial. The company was making technical progress while the actual finish line barely moved.

Then, on March 4 2026 the money finally ran out and Hardt Hyperloop was declared bankrupt.

The Numbers:

• 🚄 Founded: 2016 in the Netherlands

• 💰 Capital raised: $36M

• 🇪🇺 Major public backing: €15M from the European Commission in 2021

• 🧪 Main test site: 420-meter European Hyperloop Center

• ⚙️ Testing claim:750+ missions in one year by 2025

Reasons for Failure: 

• The capital needs were brutal: This is the main cause. Hardt run out of money. That is what happens when every step takes years, costs millions, and still leaves you far from revenue. Deeptech is hard. Infrastructure deeptech is a different level of pain.

• The company stayed stuck in prototype land: A test track can prove the tech is moving. It does not prove the business is moving. Hardt could show progress without getting meaningfully closer to commercial deployment.

• Hyperloop still makes ugly economic sense: The concept sounds great when you strip away reality. Then you add infrastructure, regulation, safety, land, politics, and adoption. Suddenly you are not building a startup. You are trying to brute-force a new transport layer into existence.

Why It Matters: 

• Milestones can lie. A startup can look alive because the product keeps moving. What matters is whether each milestone makes funding, sales, or deployment easier.

• You need a wedge, not just a vision. Big future markets sound great. You still need something smaller that customers will pay for now.

Trend

The AI Office Takeover

We get a new AI model every week now. Sometimes every six minutes. This is great news for the benchmark addicts. Less relevant for the millions of people doing actual work inside spreadsheets, docs, inboxes, and project tools.

This week, the thing that mattered was much simpler: AI got pushed deeper into work software. It was already there, obviously. But now it is getting its hands on the actual job. 

Dark times for spreadsheet masochists. Good news for everyone else.

Why it Matters

• The best position in AI now is owning the workflow. The product that already has the files, messages, meetings, tasks, and permissions has a real advantage. Once AI is wired into that context, a standalone assistant starts to look pretty small.  Looking at you, OpenAI

• Spreadsheets, calendars, and ticket queues matter more than another writing demo.

Google Workspace

This week Google updated Workspace, pushing Gemini deeper into the ecosystem.

Docs got a stronger drafting engine. Gemini can build a document from a prompt using your files, emails, and the web, then keep rewriting inside the doc. It can tighten sections, match your writing style, and even copy the format of a reference doc so the output looks like something your team would actually use.

Sheets is the big one. Gemini can now create, organize, and edit full spreadsheets from a prompt, pulling context from your files and inbox. Google also launched “Fill with Gemini”, which can populate tables with summaries, categories, generated text, and live info from Search.

That matters because a ridiculous amount of office work still lives in spreadsheets. Budgets, planning, reporting, forecasts, dashboards, all the usual corporate suffering. If Google makes building, filling, and cleaning sheets easier, that is a real piece of office work getting automated.

Slides got upgraded too. Gemini can now create slides inside a deck, match the presentation’s theme, edit through prompts, and turn rough inputs into editable diagrams and layouts.

Microsoft Copilot

A few weeks ago I discussed Claude Cowork, one of the clearest examples of AI moving from chat into actual delegated work.

Now Microsoft has brought that idea straight into the Office ecosystem with Copilot Cowork. You give it a task, it turns that into a plan, pulls context from your emails, meetings, messages, files, and data, and keeps working in the background while you approve changes along the way. 

Microsoft’s examples are very Office-core: clean up your calendar, reschedule meetings, block focus time, prep a customer meeting, build the deck, write the follow-up, pull company research, even package the outputs into docs, email, and Excel files. 

And yes, Anthropic is directly in the mix. Microsoft says it integrated the technology behind Claude Cowork into Microsoft 365 Copilot. So the product here is basically Claude Cowork’s multi-step delegation model, dropped into Outlook, Teams, Excel, and the rest of Microsoft 365, with Microsoft wrapping the whole thing in enterprise security and governance.

Notion and Others

Notion and Atlassian are also part of this trend.:

• Notion’s Custom Agents focus on recurring operational work: internal Q&A, task triage, standups, status reports, inbox cleanup. 

• Atlassian’s agents in Jira bring agents directly into tracked workflows so teams can assign work to them, mention them in comments, and run them inside the system where work is already coordinated. 

The Trend

AI is getting embedded into the software people already use to work.

Which means normal people are about to start using agents without thinking of them as agents. Your aunt will use Gemini inside Google Docs. Your mother will use Claude Cowork inside Office and never once wonder who Anthropic is. This stuff is no longer just for developers messing around.

That has big consequences for the AI war. Once Gemini is built into your docs, spreadsheets, slides, and Drive, and Microsoft brings Claude-style delegation into Outlook, Teams, and Excel, the standalone chatbots like GPT start to feel a bit naked. Owning the enterprise software may end up mattering more than owning the best model.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico